Unbuffered 8GB DDR3 memory modules have been on the market for almost a year. For a long time, their choice was limited, the frequency was limited to the standard value of 1333 MHz, and the price per gigabyte was three times higher than that of a pair of 4 GB modules.
ADATA was one of the first to announce the release of 8 GB memory per module, and after a while the two-channel kit AXDU1333GW8G9-2G was tested in our laboratory. Unfortunately, this kit did not please at all with its overclocking potential due to its absence.
Later, 8 GB Kingston modules from the inexpensive Value RAM series appeared on the market. Although they already cost a little less than the aforementioned ADATA kit, it was still very far from a comparable comparison with a pair of 4 GB modules (in terms of price, prevalence and operating frequencies).
At that time, only one manufacturer was engaged in the production of microcircuits. random access memory with a density of 4 gigabits in volumes sufficient for mass production of modules. These are Elpida Memory and their J4208BASE / J4208EASE, operating at 1.50 V and 1.35 V respectively. For this, the company's production facilities located in Japan were used. Production costs in a not the cheapest country in the world and the lack of noticeable competition have led to high prices for 8 GB modules. And, despite the fact that the legendary Elpida Hyper microcircuits were previously produced at the same factories, new memory with high density proved to be unable to work at high frequencies. Even 1600 MHz became a problem for her.
This situation persisted until the end of last year and did not contribute at all to the spread of 8 GB memory. Especially when you consider that at the same time the cost of 4 GB modules has decreased, and the price gap has only increased. Even now, most users are satisfied with two (or, in extreme cases, four) 4 GB modules. But in this review it will be not at all about how much memory is needed to solve certain problems. And about what changes have occurred in the RAM market in terms of the range and prices for 8 GB modules and which of them should be chosen for overclocking.
What has changed this year? Almost all large producers memory, such as Samsung, Hynix, Micron, Nanya (Elixir) have launched a mass production of microcircuits with a density of 4 gigabits. As a result, competition appeared in this market segment, which began to affect prices. Many companies have been manufacturing 8GB memory modules, which has improved their prevalence. As a result, buyers had a choice. The cost of this memory has not yet stabilized, but now 8 GB modules can be found on sale starting at $ 50.
The appearance of memory kits, consisting of 8 GB modules and operating at high frequencies exceeding 2 gigahertz, clearly indicates that well overclocked ones have appeared among the new microcircuits with a density of 4 gigabits. To find out which ones, six varieties of 8 GB modules, two of each type, were taken for testing. Despite the fact that the spread in price turned out to be almost twofold - from 1,500 to 2,800 rubles per module, all these models can be attributed to budget category, both due to the lack of radiators, and due to the operation at a low nominal frequency.
Specifications
The characteristics of the tested memory modules are listed in the table:
| Manufacturer module | Silicon power | NCP | Patriot | GeIL | Kingston | Samsung |
| Marking module | SP008GBLTU133N02 | NCPH10AUDR-13M28 | PSD38G13332 | GB316GB1600C10DC | KVR1333D3N9 / 8G | M378B1G73BH0-CH9 |
| Manufacturer microcircuits | Elpida | Elpida | Micron | Elpida | Elpida | Samsung |
| Marking microcircuits | S-POWER 40YT3EB | NP15H51284GF-13 | PM512M8D3BU-15 | GL1L512M88BA15BW | J4208EASE-DJ-F | K4B4G0846B-HCH9 |
| Volume, MB | 8192 | 8192 | 8192 | 2x8192 | 8192 | 8192 |
| Memory type | DDR3-1333 | DDR3-1333 | DDR3-1333 | DDR3-1600 | DDR3-1333 | DDR3-1333 |
| ECC support | Not | Not | Not | Not | Not | Not |
| Rating | PC3-10600 | PC3-10600 | PC3-10600 | PC3-12800 | PC3-10600 | PC3-10600 |
| Frequency, MHz | 1333 | 1333 | 1333 | 1600 | 1333 | 1333 |
| Timings | 9-9-9-24 | 9-9-9-24 | 9-9-9-24 | 10-10-10-28 | 9-9-9-24 | 9-9-9-24 |
| Voltage, V | 1.50 | 1.50 | 1.50 | 1.50 | 1.50 | 1.50 |
| EPP / XMP / BEMP Profiles | Not | Not | Not | XMP | Not | Not |
| Price, rubles * | 1470 | 1533 | 1680 | 3444** | 2650 | 2800 |
* The price for memory modules, actual at the time of testing, is indicated.
** The cost of a set of two modules is given.
Packaging and appearance
An overview of memory modules and the results of their testing will be presented in order of increasing cost of participants.
Let's start with the cheapest option among 8GB models at the moment - Silicon Power.
Silicon Power SP008GBLTU133N02 (S-POWER 40YT3EB) DDR3-1333 8192 MB
This memory is supplied in sealed boxes made of transparent plastic with a paper liner inside, which states that the memory has been tested and is backed by a lifetime warranty.
Outwardly, these modules do not differ from many similar ones: green color printed circuit board and sixteen DDR3 memory chips on both sides of the module, eight on each side.
The sticker shows the module labeling (SP008GBLTU133N02), memory type (DDR3), nominal frequency (1333 MHz) and CAS Latency timing (CL9). The country of manufacture can be identified by the barcode (471 - Taiwan).
The module is equipped with DDR3 memory chips with a density of 4 gigabits in an FBGA package and measuring 9.0x10.5 mm:
Their marking was changed by the module manufacturer to S-POWER 40YT3EB. But judging by the design of the dot in the lower left corner (circle with vertical stripes), before us is one of the varieties of microcircuits produced by Elpida Memory:
This is also indirectly indicated by the format of the bottom line of the marking, similar to those that can be found on other modules using Elpida microcircuits with changed markings. By the way, from it you can find out the date of production of microcircuits - 51st week of 2011.
Unfortunately, at the moment on the Elpida Memory website there is information only about four-gigabit A-Die microcircuits (Elpida J4208BASE / J4208EASE), but they differ large size(9.3x13.0 mm).
But on the net you can find pictures of newer "B-Die" chips (Elpida J4208BBBG) and, comparing the sizes, we can assume that they are used on Silicon Power SP008GBLTU133N02.
SPD chip:
silicon_power_sp008gbltu133n02_spd.
The module is based on a six-layer HJ M1 KO-60244 PCB:
The next in line is the Hexon Technology memory option. It is quite insignificantly more expensive than the previous one and, with the exception of packaging, does not differ in anything special from it.
NCP NCPH10AUDR-13M28 (NCP NP15H51284GF-13) DDR3-1333 8192 MB
This copy came for testing without any packaging.
This is another green memory module standard size and with microcircuits installed on both sides:
The marking of the module is NP15H51284GF-13, the volume is 8 GB, the nominal frequency is 1333 MHz. The manufacturer's warranty is lifetime, but in Russian stores it is often limited to 3-5 years.
Hexon Technology has used Infineon chips in the past, but on the NCP NCPH10AUDR-13M28 module you can see the same as on the Silicon Power SP008GBLTU133N02:
Dimensions 9.0x10.5 mm, dot design and bottom line show the use of the same Elpida "B-Die" microcircuits, only produced a little earlier - in the 42nd week of 2011.
SPD chip:
Archive with a dump of its contents in SPDTool and Thaiphoon Burner formats: ncp_ncph10audr-13m28_spd.
PCB - 6-layer B63URB 0.70 from Brain Power:
Let's move on to the review of memory modules manufactured by Patriot Memory. They are more expensive than Silicon Power options, but still closer to them in price than GeIL, Kingston and Samsung.
Patriot PSD38G13332 (Patriot PM512M8D3BU-15) DDR3-1333 8192 MB
This memory belongs to the inexpensive Signature Line series. The modules are delivered in a transparent plastic package, but they got to the store without it.
The memory module is made on a dark green printed circuit board of a standard size with microcircuits installed on both sides:
The sticker shows the module labeling (PSD38G13332), its volume (8 GB) and the nominal frequency (1333 MHz):
Installed DDR3 memory chips with a density of 4 gigabits in an FBGA package and measuring 10.5x12.0 mm:
V in this case top part the microcircuits were marked by the manufacturer of the module, and the rest of it can be determined that these are Micron microcircuits.
There is no point with a double helix, but this type of marking has already been possible before, for example, on Crucial modules. To determine the type of microcircuits used in the Patriot PSD38G13332, we will use the catalog on the manufacturer's website. After setting all known parameters, we find the only one suitable option MT41J512M8RA-15E revision D. Then, using the decoder, we get the FBGA code D9PCH.
SPD chip:
Archive with a dump of its contents in SPDTool and Thaiphoon Burner formats: patriot_psd38g13332_spd.
Patriot PSD38G13332 uses printed circuit board Levin 2601:
Let's move on to the second half of the test participants and the two-channel memory kit manufactured by Golden Emperor International Ltd (GeIL).
What is the difference between DDR3 RAM with a frequency of 1333 and 1600 or 1866, should it be set with frequencies higher than 1333?
- The difference is in the frequency with which the motherboard communicates with memory. If the memory stick is 1333 and you add 1600, then it will work with min. frequency - ie 1333. If two 1600 - then both will work with 1600 or max. which the motherboard supports.
- you always need to take the maximum memory supported by the processor. The i5 650, for example, works with RAM at 1600 and higher. If you do not see it, set up hertzovka in the BIOS, and stick the same strips into the dual channel. The hertsovka of the RAM is chased like the hertsovka of the processor itself, usually automatically, optimally adjusting to its work, but it can be configured separately. In terms of performance, the RAM has little effect. It slows down the most because of the HDD, the bus and the performance of the video card, and overheating of the equipment.
- the alignment is like this - for everyone modern sockets except for the 775, the memory controller is not built into the motherboard but into the processor. So, built in Phenom processor on socket AM3, the memory controller can in normal mode, if not overclock, work with memory at a maximum of 1333 frequency - that is, faster memory - 1600 or 1800, and so on without overclocking the processor via the FSB bus will still work at 1333 frequency. That is, if you are going to overclock the processor only by raising the multiplier - and it allows you, since the Black Edition - the multiplier is not locked, then it makes no sense to take a faster memory, if you are going to raise the processor bus frequency - then here you can take memory with a frequency of 1600 MHz ... When the processor is overclocked via the bus, the memory bus frequency will increase by 20 percent, and it will just reach 1600 MHz. There is no point in taking a faster memory - just transferring money.
- if the iron allows, then it is better to take it with a higher frequency, the higher the frequency, the higher the processing speed, etc., etc.
- DON'T SEE ME 1600
- I advise you to set the 1333 MHz frequency calculated for the DDR-3 standard. The rest is like overclocking, it will not add performance, but there is a risk
- overclockers.ru
- It will be efficient if the memory bandwidth does not exceed the processor bandwidth. Prop. memory capacity is determined by its frequency, see the Internet.
- mind you, if the operating frequency is higher, then the timings will also be higher.
A higher frequency is needed for applications that actively use RAM (archiving, decoding)
The frequency of the opera has practically no effect on the operation of the processor, the processor has its own cache memory, which it basically uses (especially since you have Fenom, and it has a built-in memory controller) - Bogdan Zhukov, you are an idiot! And what's the turbo boost, and the frequency of your processor, if it comes about RAM? Fuck your nose if you're a ram? And throw out your dumb 3600 processor in boost!
- in games you won't even notice
- The higher the frequency, the higher the speed
- 1333 better game pulls A higher Already useless (only for programming)
- If you plan to surf the Internet, then you can leave 1333, and if you play games, then of course it's better more, I have a 3.6 frequency in the turbo boost
08.10.2012
The question: is it worth taking a faster memory - is faced by many buyers. Due to a decrease in the price of DDR3 modules with a frequency of 1600 megahertz and above, it has become even more relevant. The answer would seem obvious - of course it's worth it! But what gain can a higher memory frequency provide, and is it worth overpaying? This is what we will try to find out.
If until recently the choice of RAM was simple, you have extra money, you take DDR3 with a frequency of 1600 megahertz, if not, you are content with DDR3-1333. V currently on store shelves there is huge selection RAM with a frequency higher than 1600 megahertz, and at a very reasonable price. This encourages buyers to make their choice in favor of more fast models, with frequencies of 1866, 2000, and 2133 megahertz. And this is quite justified in theory - the higher the memory frequency, the higher the bandwidth, the higher the performance.
However, in real conditions the situation may be slightly different. No, a system with DDR3-2000 modules cannot be slower than a system with DDR3-1333 modules. In this case, “do not spoil the porridge with butter”. But the difference in performance can be almost imperceptible in most applications that we use in everyday life. In fact, of the constantly used applications, only archivers clearly and unambiguously respond to the increased frequency by increasing productivity. Otherwise, it is not easy to tell the difference.
At the same time, fast RAM continues to be actively promoted by manufacturers and sellers as a solution for gamers. As a result, users feel that the memory frequency is almost as critical as the number of cores in the processor, the number of stream processors, and the width of the memory bus in the video card chip.
To debunk, or vice versa, to confirm this statement, we conceived this test. Its principle is simple - we will test the same set of memory in several games when operating at different frequencies, and try to find out what kind of gain the increase in memory frequency actually gives. And does it give at all.
To conduct the test, we used our test bench, in which we installed a Team Xtreem Dark memory kit with a base frequency of 1866 megahertz manufactured by Team Group. Two memory modules with a capacity of 4 gigabytes each have 9-11-9-27 timings, standard for the nominal frequency, are labeled TDD34G1866HC9KBK, and operate at 1.65 volts. Quite affordable and fast memory modules with a three-year warranty and original heatsinks, which may well become the choice of a gamer who does not want to give crazy money for modules with a frequency of more than 2 GHz. Therefore, they fit perfectly into the concept of the test.
It was decided to test the memory at three frequencies - 1333, 1600, and 1866 megahertz. From more low frequencies at 800 and 1066 megahertz, it was decided to refuse, since the purchase of such modules (if you can still find them on sale) would be unreasonable, since they will be the same in price with DDR3-1333 modules. Although the 2000 megahertz mode was theoretically planned, but the harsh reality made changes to these plans. The memory frequency multiplier in our ASUS P8Z77-V board does not support this frequency, but next step over 1866 megahertz, offers 2133. At this memory frequency, the system booted at constant voltage, allowed it to work, and even passed the 3DMark Physics test, but launching any game resulted in “ blue screen”. And neither the increase in timings, nor the increase in voltage helped. Therefore, the high frequencies had to be abandoned.
In principle, there is nothing wrong with this, because the goal this test not checking the most expensive and fastest memory modules, but finding out the dependence of performance in games on frequency. If, as a result, it turns out that there is an increase, then based on the results of tests with three different frequencies, it will be possible to deduce an approximate increase for models with frequencies over 2000 megahertz by interpolating the results obtained.
During the test, we decided not to change the timings, so as not to confuse the results. But in the end we decided to give a small head start to the lowest frequency, and besides the mode with 9-11-9-27 timings, we ran tests with 7-7-7-21 timings, which are standard for good modules DDR3-1333. Note that we carried out all tests at a resolution of 1280 by 720 pixels, at maximum quality settings, using 16x anisotropic filtering, and without anti-aliasing. It was necessary to lower the resolution in order to reduce the impact of the performance of the video card, which traditionally becomes a bottleneck in gaming tests.
Well, the input data is there, it's time to move on to the test results. To estimate the theoretical gain bandwidth memory when increasing the frequency, all configurations were tested in the AIDA 64 package. This synthetic test gave quite logical and expected results. There is an increase in throughput with increasing frequency, and the mode with the minimum timings made it possible to obtain higher results than the mode with lower ones. Moving on to the game test results.
In Performance mode, 3DMark 11 demonstrated that the memory frequency has a linear effect on the final result. The faster the memory, the more points. How much more? As you can see from the diagram, with a total result of more than 6000 points, the system with DDR3-1866 memory outperformed DDR3-1333 with equal timings, only 111 points. This difference can be expressed as a modest figure of 1.8 percent. If DDR3-1333 memory operates at 7-7-7-21 timings, which are more usual for itself, then the difference with the fastest memory decreases to 1.5 percent. That is, in this case, the use of more fast memory does not give a noticeable increase.
This subtest was the only one in 3DMark 11 that reacted very positively to higher memory frequencies and lower timings. The load on the video card is not high here, but the load on the processor when calculating physics is very high. Accordingly, the load on the memory is also high, which stores all the results of data processing. As a result, the gap between DDR3-1866 and DDR3-1333 with equal timings was just over 16 percent. Reducing the timings of the slowest memory allows the gap to be reduced by up to 12.8 percent. DDR3-1600 turned out to be exactly in the middle between DDR3-1333 and DDR3-1866, as it should be in terms of frequency. Considering, very strange for real applications resource usage in this test, we will not take into account its results. There are no such games with such a load distribution, and most likely there never will be.
Metro 2033
We honestly didn't expect to see such interesting results. Moreover, they are interesting not because of their large gain, but because of their dependence on timings. In a direct comparison of three frequencies with equal timings, we observe the same linearity - with an increase in frequency, performance also increases. But the growth is scanty, and almost imperceptible: DDR3-1866 is faster than DDR3-1333 by only 0.8 frames per second, which is a modest 1.3 percent. Very little. There is again DDR3-1600 memory between them. But DDR3-1333 with 7-7-7-21 timings demonstrated remarkable potential, demonstrating the same result as the fast DDR3-1866 with 9-11-9-27 timings. This suggests that lower timings are preferable for this game, and DDR3-1600 with 8-8-8-24 timings could well have won this test. By the way, shifting the physics calculation from the video card to the processor did not change the alignment of forces and breaks, as could be expected after the 3DMark 11 Physics test.
Crysis 2
Inspired by the results of previous tests, which showed the way and almost imperceptible to the naked eye, but still present, the increase in performance, we moved on to the game Crysis 2, and then a revelation awaited us. All four configurations, as you can see in the diagram, demonstrated exactly the same result, with an accuracy of up to one tenth of a frame per second. Yes, it also happens. Apparently, the CryEngine engine is completely insensitive to the memory bandwidth. Let us state this fact and move on to the last test.
DiRT Showdown
This test produced the most controversial and inexplicable result. First, I was surprised by the DDR3-1333 memory with minimal timings, which yielded to the memory operating at the same frequency, but with higher timings, which is unnatural in principle. True, it yielded quite a tiny bit - 00.8 percent. DDR3-1600 turned out to be faster than DDR3-1333 at the same timings, by a reasonable and understandable 1.7 percent. But DDR3-1866 showed an incredible gain! The superiority over DDR3-1600 is a solid 5.8 percent. That's really a lot. Considering all the previous results. After all, it was quite logical and expected to see the same 1.7 percent that DDR3-1600 and DDR3-1333 shared - then the gain would be linear. Based on experience, we know that such results can be accidental, and an inexplicable result of some internal program failure, as in our practice there was a case when 3DMark 03 completely undeservedly gave the GeForce FX 5200 a result that was superior to the results of the top cards of that time. ... Well, given that it is customary to ignore nonlinear results in statistics, this is what we will do.
The range of DDR3 frequencies was revealed much earlier than that of DDR2, since DDR3 modules with frequencies of 1066, 1333 and 1600 MHz (DDR) have already appeared on the market, and are intended to replace DDR2 memory at 533, 667 and 800 MHz (DDR). As with DDR2, there are higher, "non-standard" frequencies, but these are aimed at the enthusiast, not the mainstream market.
In our review, we will consider modules that operate at "mass" DDR3 speeds, since 1333 MHz (DDR) memory falls exactly in the middle between "budget" (1066 MHz) and high-end (1600 MHz). In total, we invited 13 different companies to participate, and eight of them sent their memory for our testing.
As in previous tests memory, we overclocked each set to the stability limit to find the performance threshold. But before we get into DDR3 modules, let's talk about this market. What are the advantages of the new memory over DDR2? Why was it presented? And when a new technology enters the market at a hefty price, is it worth the money?
What's in a name?
The "official" name for DDR memory is based on bandwidth, not clock speed. An easy way to convert its effective frequency to bandwidth is to multiply by eight. So, DDR-400 is called PC-3200, DDR2-800 is called PC2-6400, and DDR3-1600 is called PC2-12800.
It is very simple to explain this mathematics: PC modules based on SDRAM technology are connected via a 64-bit bus; there are eight bits in a byte, and 64 bits are equivalent to eight bytes. For example, DDR2-800 transmits 800 megabits per second over a single line; 64 lines provide simultaneous transmission eight bits, and if 800 multiplied by eight, that's exactly 6,400.
But there is a rounding problem that first appeared with DDR-266 (PC-2100). The effective transmit frequency of 266 MHz is actually 266, (6) (six per cycle) MHz, so the actual bandwidth is 2,133 MB / s.
Today, DDR3-1333 memory provides a peak bandwidth of 10,666 MB / s, which can be rounded down to PC3-10600, up to PC3-10700, or left as PC3-10666 at the manufacturer's request.
Buyers planning to choose from multiple DDR3-1333 memory kits should pay attention to all three names, although most manufacturers mark their DDR3-1333 modules as PC3-10600 or PC3-10666.
Upcoming bandwidth ... today!
It is often argued that DDR2 memory is fast enough for modern processors, since the fastest current system bus, Intel FSB (Front Side Bus), operates at an effective frequency of 1,333 MHz. Do you need memory at this frequency of 1,333 MHz? The short answer is no.
Intel, since the appearance of RDRAM on the first Pentium 4, has been using dual channel memory, in which the width of the memory bus is doubled, since even then it was impossible to find memory that worked no slower than the FSB. The very first Pentium 4s used a 64-bit FSB with an effective frequency of 400 MHz even before the advent of DDR-400, but two 64-bit DDR-200 (PC-1600) modules were sufficient for such an FSB if the memory bus width was doubled to 128 bits. ... If only there had been a DDR SDRAM chipset for Pentium 4. The dual-channel technology has survived since then, and FSB1333 exactly matches the bandwidth of two DDR2-667 (PC2-5300) modules in dual-channel mode.
Another argument lies in the "synchronous" memory operation in relation to the CPU FSB: many people think that DDR3-1333 memory works synchronously with the FSB-1333. However, it is not. Intel uses Quad Data Rate (QDR) technology for the FSB, and memory uses Double Data Rate (DDR) technology. The FSB-1333 runs at a physical clock speed of 333 MHz, which is equivalent to DDR2-667 memory.
Yes, some users notice a small performance gain from working memory with a multiplier of up to 1.5x in relation to the CPU FSB frequency, violating the principle of synchronous operation. Actually, this is why DDR2-667 memory became popular even before the advent of Intel FSB-1333, and that is why DDR2-800 memory is well bought even by those who do not plan to engage in overclocking.
While many builders have not needed anything other than inexpensive DDR2 modules for some time now, DDR3 memory has two key benefits. First, the maximum memory density of the chips has been expanded to 8 Gbps, which gives a 16-chip module a capacity of 16 GB. Secondly, the default supply voltage has been reduced to 1.50 V compared to 1.80 V for DDR2, which gives a 30% reduction in power consumption at the same clock speeds.
Should I buy or not?
One of the important arguments in favor of DDR3 memory is the gradual movement of Intel chipsets in this direction. The company first added DDR3 support as an option on the Northbridge of the P35 Express chipset, and the DDR3 market was then further expanded with the introduction of new DDR3 chipsets. Motherboard makers will try to grab the cream of the crop from early adopters of new technologies, so most motherboards based on the very expensive X48 chipset are likely to support the latest memory standard. Meanwhile, DDR3 will gradually descend to the "budget" market.
The latest technology always comes at a price, and DDR2 memory is sufficient for most systems, so why bother? Intel will likely be preparing the desktop market for the next big step, in particular by moving the memory controller from the chipset to the processor itself. Like the current AMD processors This step removes the FSB bandwidth limitation and allows future processors to receive data at the same rate as it would be transferred from memory.
The buyer himself has the right to decide whether to bear the burden of promoting new technologies to the masses. Many people still remember how RDRAM memory was completely in vain for Pentium III chipsets, the same i820 and i840, as Intel was preparing the i850 chipset for Pentium 4 with the same memory. Intel's plan was to expand the availability of RDRAM memory to the point where it really needed it, but the market reacted negatively. However, the similarities with the promotion of DDR3 end there, since Intel does not force memory to the market, but provides a similar option to increase performance.
However, you shouldn't think that Intel's DDR3 memory at the current FSB frequencies is so useless, because the significantly increased frequencies allow you to overclock the FSB well. FSB-1600 (physical frequency 400 MHz) will appear soon, and if you want to overclock a 2.80 GHz processor from FSB1600 (400 MHz FSB x7) to 4.20 GHz (600 MHz FSB x7), you will need memory capable of running at an effective frequency of 1200 MHz (physical frequency 600 MHz). DDR2-1200 is rare, as this memory requires excessive voltage boost, good cooling and the user's prayers that it will not "die", since these are just overclocked DDR2-800 modules.
Therefore, while most builders of systems on Core 2 compare prices of DDR2-800 with different models DDR3, overclockers see DDR3-1333 as a faster, cheaper and more reliable alternative to DDR2-1200. Moreover, as DDR3 progresses to the mainstream market, overclockers with more limited budget will also join them.
Frequency vs latency: myths and facts
There is a myth that each new format increases the response time. This myth is based on the method by which latencies (timings) are measured: takt time.
Let's take a look at the delays of the last three memory formats: DDR-333 memory for the upper segment of the mass market worked with CAS 2 delays; similar positioning DDR2-667 memory - with CAS 4, and modern DDR3-1333 memory - with CAS 8. Most users will be surprised to learn that such different CAS latencies actually give the same time response, namely, 12 nanoseconds.
The point is that the clock time (period) is inversely proportional to the clock frequency (1/2 of the effective DDR frequency). DDR-333 has a cycle time of six nanoseconds, DDR2-667 has three nanoseconds, and DDR3-1333 has 1.5ns. Latency is measured in clock cycles, and two 6-ns clock cycles are as long as four 3-ns or eight 1.5-ns. If you still have doubts, do the math!
Many not very thoughtful buyers believe that faster memory responds more slowly, but from the examples given, it is obvious that this is not the case. The problem is not that response times are getting faster, but that they are not getting faster! When we look at astronomical frequencies, we hope that the system will become more responsive as a result. However, for last years memory delays, alas, have not perceptibly changed.
We're still hoping to find really fast modules, so our tests include both maximum frequencies and minimum response times. All this while maintaining the stability of the system.
But what do these numbers mean?
So latencies are measured in ticks, not seconds, but what do they mean? For most buyers, we recommend looking at only the first four values, which are listed in order of importance, such as 9-9-9-24 for high-speed DDR3 modules. The delays are commonly referred to as CAS Latency (tCL), RAS to CAS Delay (tRCD), RAS Precharge Time (tRP), and Active Precharge Delay (tRAS).
10 sets for choice
Most of the 13 memory manufacturers we contacted were willing to participate in our summary tests, but a few companies do not yet manufacture DDR3 memory with an effective frequency of 1,333MHz. Some completely ignore the mass market, focusing on "budget" DDR3-1066 and extreme DDR3-1600 models. The only company that produces modules, but did not meet the deadline, is Team Group. Of the eight companies that participated in our tests, OCZ and Kingston sent out a couple of kits each, which indicates a wide range of these companies.
If you have never heard of the company Aeneon, You are not alone. This is a new retail brand Qimonda... If the last name does not tell you anything either, it is probably due to the fact that this is the name of the former division for the production of memory. Infineon... Experienced assemblers should be familiar with Infineon memory and its reputation for quality and reliability.
While other manufacturers are trying to choose to name their memory PC3-10600 or PC3-10666, Aeneon decided to leave this dispute and name its memory by effective frequency, not bandwidth. After all, many collectors pay attention, first of all, to frequency, and not to bandwidth.
Modules are sold under model number AXH760UD00-13G... The package includes two 1 GB DDR3-1333 modules with the declared physical frequency of 667 MHz and delays of 8-8-8-15 at a default voltage of 1.50 V. The closest value in the SPD table is 8-8-8-24 ... If you want the X-Tune modules to work with the declared delays, then you should enter the BIOS and manually reduce the tRAS latency from 24 to 15 cycles.
Low 416MHz SPDs (DDR3-833) ensure systems with low FSBs boot to automatic configuration, and Aeneon decided to go up one notch by providing a 750 MHz profile. For processors with FSB1066, 500 MHz mode with a memory multiplier of 3: 2 DRAM: FSB would be useful, however a 533 MHz profile (DDR3-1066) would be more useful for automatically tuning more configurations.
Company G.Skill has earned a very respectable reputation among the budget enthusiast because it offers high-speed memory at standard retail prices. In the case of DDR3-1333, we found that the memory is sold at the prices of top-end DDR3-1066 modules.
But being a "low-cost performance" module does not mean that you have to make any compromises. G-Skill did a good job, the modules are equipped with heat spreaders, and the quality of the packaging can compete with more expensive models. Under the number F3-10600CL9D-2GBNQ hides a set of two 1 GB DDR3-1333 modules with standard delays of 9-9-9-24 at a default voltage of 1.50 V. Memory, as indicated, can operate at any voltage from 1.50 to 1.60 V, which will allow it to overclock.
The SPD value for the physical frequency of 667 MHz (DDR3-1333) turned out to be quite expected, but the modes at 592 and 444 MHz seemed somewhat strange to us. But we tested the modules on different motherboards and can confirm that the 592 MHz mode (DDR3-1184) works as needed and like DDR3-1066.
If you need to make the G.Skill PC3-10600 modules work higher than the standard values, then you will have to use manual configuration.
Kingston Probably the most mass-market-oriented manufacturer in our review, it offers a full line of modules, from the unremarkable to the highly curious. The company provided us with two kits with the same frequencies, while the ValueRAM PC3-10600 belongs to the "standard performance" class.
Modules look very modest, but Kingston has specified for modules KVR1333D3N8 / 1G very productive delays of 8-8-8-24 at the motherboard nominal voltage of 1.50 V. Two 1 GB modules provide a dual channel set, which is why the company sent us a couple of DIMMs.
SPD values for frequencies 667, 583, 500 and 416 MHz provide automatic tuning for DDR3-1333, DDR3-1066, DDR3-1000 and DDR3-800 memory, with little overclocking potential at 416 and 583 MHz.
Since all memory modes have SPD settings, manual configuration is not required.
In line Kingston hyperx includes modules that surpass the capabilities of standard components. So, the PC3-11000 set is declared to work at a frequency of 1,375 MHz. However, this value is very close to the standard 1333 MHz, which allowed us to consider them simply improved DDR3-1333 DIMMs.
Into the set KHX11000D3LLK2 / 2G includes two 1GB modules with blue heat spreaders, with declared delays at 1.70V. Non-standard voltage requires manual setting in BIOS, and by default the modules operate in a slow mode of 533 MHz (DDR3-1066) to ensure loading at the stock 1.50 V.
In fact, the SPD value for DDR3-1333 is not present in the HyperX table, the highest 1.50-V setting is 609 MHz on CAS 8. Since the modules normally operate with the declared delays at a lower clock frequency, you will have to manually change the frequency and voltage in the BIOS motherboard.
The 457 MHz SPD mode will be useful for DDR3-800 autoconfiguration when using FSB800 processors, while the 533 MHz DDR3-1066 value works for FSB1066, FSB1333 and FSB1600 processors.
In recent years Mushkin shifted the focus from "extreme performance" to "absolute stability." Although the company continues its efforts to produce high-speed modules. What else does an enthusiast need besides stability and speed?
Unlike many previous Mushkin products, the set 996583 of the two 1 GB modules, it is announced at DDR3-1333 with very modest delays of 9-9-9-24 at the default voltage of 1.50 V. This DDR3-1333 mode is set in SPD, so the memory will automatically work with the FSB1333 processor.
Other SPD values include 444 and 518 MHz, which BIOS recognizes as DDR3-800 and DDR3-1000. Again, for most users the normal DDR3-1066 mode would suit better than the weird DDR3-1036, since a system with DDR3-1066 will default to the slow SPD latencies for DDR3-1333.
Like Kingston, OCZ wants to cover as much of the DDR3-1333 market as possible by offering multiple modules. But, unlike Kingston, the "junior" kit from OCZ belongs to the middle level, providing the same CAS 7 latency as the competitor's high-end modules.
Yes, you can find OCZ's Gold kit even cheaper on the market, but the mainstream Platinum Edition line has 7-7-7-20 delays. These are not just declared delays, for which you need to manually dig into the BIOS, they are spelled out in the SPD of the kit. OCZ3P13332GK of two 1 GB modules.
But there is a bit of an oddity here: OCZ Platinum modules have to run at full performance at 1.70V, and the SPD table lists the delays mentioned for 1.50V. OCZ is one of the companies that shipped modules with which some systems failed to boot because the SPD values were too tight to operate at the motherboard's nominal voltage (1.50V for DDR3).
The good news is that our modules ran stably at the specified latencies, we didn't have to manually raise the voltage from the stock 1.50V to the OCZ recommended 1.70V. This is true for both Gigabyte and Asus motherboards.
The SPD value of 761 MHz (DDR3-1522) with delays of 8-8-8-23 provides overclocking potential for those overclockers who are not familiar with manual setting of memory modes, and the SPD values of 571 and 476 MHz go into DDR3-1066 and DDR3-800 for processors with lower FSB frequency.
If you've been impressed by the OCZ Platinum Edition's mid-market modules operating at the same latency as the high-end versions of some of the competition, then you're sure to be even more mesmerized by the reported latencies for the ReaperX line. Equipped with a dual heatpipe heatsink, the ReaperX modules are rated for an effective frequency of 1333 MHz with CAS 6 delays.
CAS 6 latency sounds impressive, but memory is not limited. The supported mode is 6-5-5-18, which is faster than 6-6-6-x, commonly referred to as "CAS 6". A complex cooling system plays a significant role, since for DDR3-1333 to work with delays of 6-5-5-18, the voltage must be increased to 1.85 V.
However, for the ReaperX modules to work at the declared level, you need to enter the BIOS and manually set the frequency, delays and voltage. But this can be forgiven for modules with extreme performance, since the target audience clearly familiar with BIOS setting... But beginning overclockers can get into trouble.
Even in 533-MHz SPD mode (DDR3-1066) modules ReaperX OCZ3RPX1333EB2GK use delays 6-5-5-20 instead of 6-5-5-18, but by at least, DDR3-1066 auto-configuration ensures stable first boot before manual BIOS adjustments.
SPD does not contain values for DDR3-1333, instead of them the unusual DDR3-1244 mode at a physical frequency of 622 MHz is used, there is also DDR3-1422 at 711 MHz. But none of our motherboards started using DDR3-1422 latency for DDR3-1333 mode by default on FSB1333 processor, but reduced the frequency of ReaperX modules to DDR2-1066 automatic configuration. CPU-Z points out that the reason for this behavior is probably the electronic labeling of the PC3-8500 modules instead of the PC3-10700.
PDP Patriot sent us a kit PDC34G1333LLK where LLK at the end of the model number denotes low latency two-channel dialing. It provides the same CAS 7 latency as high-end Kingston HyperX modules and mid-range OCZ Platinum Edition memory, but here we saw something that was not found in other kits: 4 GB of capacity. Although many companies offer 4GB kits to their customers today, only Patriot has the courage to send us such a kit for overclocking and minimal latency tests.
Patriot's decision to provide us with high-capacity modules for overclocking tests speaks volumes of confidence in their performance, since higher-capacity modules are more difficult to achieve stable operation. Like Kingston and OCZ, the company indicated that system builders need to go into the BIOS and raise the DIMM voltage from 1.50V to 1.70V, after which they can manually set the declared DDR3-1333 settings to 7-7-7-20.
In fact, there are no values for DDR3-1333 in SPD, but delays 7-7-7-20 are indicated for operation in DDR3-1066 mode (physical frequency 533 MHz) at a nominal voltage of 1.50 V. However, for users familiar with BIOS, it will not be difficult to specify the desired mode.
SPD has a 457 MHz mode (DDR3-914), which allows owners of processors with FSB800 to automatically set the memory to DDR3-800 before any manual changes in the BIOS.
Insofar as more values DDR3 is not provided for in SPD, our memory on both Gigabyte and Asus motherboards normally worked in DDR3-1066 mode with FSB-1333 and FSB-1600 processors.
This company is known for modules with extreme capabilities, so from a set of two 1 GB DIMM Super Talent W1333UX2G8 we expected a lot.
Previously Super talent released DDR3-1600 modules capable of overclocking above 2 GHz even before most competitors introduced DDR3-1333. On the other hand, average delays of 8-8-8-18 and a very high voltage of 1.80 don't inspire much enthusiasm for mid-range models. Only tests will tell whether or not DIMMs live up to Super Talent's reputation for high overclocking.
In the SPD table, Super Talent does not have DDR3-1333 modes (physical frequency 667 MHz), the electronic marking of the modules is DDR3-1066. That is, in most configurations, the memory will be configured for DDR3-1066 mode.
Super Talent is the only company in our testing to add Intel XMP SPD Extensions, which work similar to the EPP (Enhanced Performance Profiles) profiles familiar to DDR2 motherboard enthusiasts, where the memory is automatically configured for overvoltage and overclocked modes. In this case, Super Talent can automatically overclock DDR3-1333 modules to DDR3-1600 at a very high voltage of 2.00 V.
Company Wintec Industries known mostly to OEMs. However, it has been producing the high-speed AMPX line for several product cycles and hopes to be able to win the trust of enthusiasts and overclockers on a budget. The company sent us a couple of the latest 1GB AMPX PC3-10600 modules.
The declared delays are 9-9-9-24 at a nominal voltage of 1.50 V, that is, a pair of 3AHX1333C9-2048K gigabyte modules implies independent overclocking efforts, but, unlike more expensive DIMMs, the memory is not validated for any speed modes ...
In fact, Wintec AMPX PC3-10600 does not even have SPD modes higher than DDR3-1066, and the modules themselves are electrically declared as slower. Therefore, after assembling the system, you need to manually adjust the frequency and delays, even if the modules are declared to work in DDR3-1333 mode at standard voltage.
In general, it is strange that the declared mode is not in the SPD table. Perhaps this will be done later, with newer DIMM batches.
Comparison of SPD latencies
Although below in the test of "minimum stable delays" we will give the minimum timings at increased voltage, we decided to give a table of SPD values, which clearly shows which market certain modules are oriented to.
| Auto tuning (MHz: tCL-tRCD-tRP-tRAS) | ||||
| Company / Model / Number | Defined as | SPD delays | Memory mode | SPD extensions |
| Aeneon X-Tune DDR3-1333 AXH760UD00-13G |
667: 8-8-8-15 | 416: 5-5-5-15 500: 6-6-6-18 667: 8-8-8-24 750: 9-9-9-27 |
DDR3-1333 CAS 8-8-8-15 1.50V |
Not |
| G.Skill PC3-10600 F3-10600CL9D-2GBNQ |
667: 9-9-9-24 | 444: 6-6-6-16 592: 8-8-8-22 667: 9-9-9-24 |
DDR3-1333 CAS 9-9-9-24 1.50-1.65V |
Not |
| Kingston ValueRAM PC3-10600 KVR1333D3N8 / 1G |
667: 8-8-8-24 | 416: 5-5-5-15 500: 6-6-6-18 583: 7-7-7-21 667: 8-8-8-24 |
DDR3-1333 CAS 8-8-8-24 1.50V |
Not |
| Kingston HyperX PC3-11000 KHX11000D3LLK2 / 2G |
533: 7-7-7-20 | 457: 6-6-6-18 533: 7-7-7-20 609: 8-8-8-23 |
DDR3-1333 CAS 7-7-7-20 1.70V |
Not |
| Mushkin Enhanced EM3-10666 996583 |
667: 9-9-9-24 | 444: 6-6-6-16 518: 7-7-7-19 667: 9-9-9-24 |
DDR3-1333 CAS 9-9-9-24 1.5V |
Not |
| OCZ Technology PC3-10666 Platinum Edition OCZ3P13332GK |
667: 7-7-7-20 | 476: 5-5-5-15 571: 6-6-6-18 667: 7-7-7-20 761: 8-8-8-23 |
DDR3-1333 CAS 7-7-7-20 1.8V |
Not |
| OCZ Technology PC3-10666 ReaperX OCZ3RPX1333EB2GK |
533: 6-5-5-20 | 533: 6-5-5-20 622: 7-6-6-24 711: 8-7-7-27 |
DDR3-1333 CAS 6-5-5-18 1.85V |
Not |
| Patriot extreme performance PC3-10666 Low Latency Kit PDC34G1333LLK |
533: 7-7-7-20 | 457: 6-6-6-18 533: 7-7-7-20 |
DDR3-1333 CAS 7-7-7-20 1.7V |
Not |
| Super talent PC3-10600 CL8 W1333UX2G8 |
533: 7-7-7-20 | 533: 7-7-7-20 609: 8-8-8-23 |
DDR3-1333 CAS 8-8-8-18 1.80V |
XMP-1600 CAS 8-8-8-28 2.00V |
| Wintec Industries AMPX PC3-10600 3AHX1333C9-2048K |
533: 8-8-8-20 | 400: 6-6-6-15 533: 8-8-8-20 |
DDR3-1333 CAS 9-9-9-24 1.5V |
Not |
Aeneon and OCZ have specified SPD profiles that exceed their stated specifications, and Intel XMP profiles on Super Talent memory provide automatic memory overclocking. Kingston and PDP Patriot were targeting low latency audiences, while OCZ hit both markets with their two kits.
The price of DDR3 still hasn't dropped to the level of the mass market, and today one of the main reasons for buying relatively expensive DDR3 memory is overclocking, which would not be limited by the memory frequency. Sure, you could pay astronomical sums for DDR3-1800 or even faster memory for overclockers, but we still wanted to see what the less expensive kits were capable of.
New components appear today, but the best boards for overclocking are models based on the Intel P35 chipset, at the same time, Core processors 2 Duos can handle a noticeably higher bus frequency than the Core 2 Quad. Therefore, we assembled the system in such a way as to make it maximally aimed at overclocking, regardless of the age of the components.
| Overclocking test system | |
| Motherboard | Gigabyte GA-P35T-DQ6, Rev. 1.0, Intel P35, BIOS F5c (10/26/2007) |
| Socket 775 processor | Intel Core 2 Duo E6750 "Conroe", FSB-1333, 65 nm, 2.67 GHz, 4 MB L2 cache |
| HDD | |
| Video card | |
| Power Supply | |
| System software and drivers | |
| OS | |
| DirectX version | 9.0c (4.09.0000.0904) |
| Platform Drivers | Intel INF 8.3.1.1009 |
| Graphics driver | nVidia Forceware 163.75 |
Our Core 2 Duo test unit was very successful, as it was able to achieve 520MHz FSB at an 8x multiplier by default and 540MHz FSB at a 6x multiplier on top motherboards. Using the highest memory multiplier of the P35 chipset, with a 6x CPU multiplier, we can get an effective memory frequency of 2 160 MHz!
Of course, we needed a motherboard that works very stably with memory, and the Gigabyte GA-P35T-DQ6 was the best choice for this role.
To get different memory frequencies at a fixed 6x multiplier, we had to change the CPU frequency in each test. The change CPU frequency significantly affects the results of usual tests, so we limited ourselves only to the memory bandwidth tests in the overclocking section.
| Overclocking tests | |
| PCMark05 Pro | Version: 1.1.0 Memory tests |
| SiSoftware Sandra 2005 | Version 2005.7.10.60 Memory Test = Bandwidth Benchmark |
Overclocking memory often requires more power, but some modules are less tolerant of overvoltage than others. Likewise, there are more aggressive overclockers and there are more moderate ones. Therefore, we have chosen three voltage levels to suit most of the audience: nominal (1.50 V), reasonable overvoltage (1.80 V) and crazy for aggressive overclockers - 2.10 V. Note that even our "reasonably safe "the level is a 20% increase in nominal voltage, although we are quite confident that most modules will withstand this mode for several years.
In order to put all modules in the same frame, we reduced the latencies in the overclocking tests to the level of 9-9-9-24. What will be the results?
OCZ Platinum DDR3-1333 easily outpaced the competition by 2.10V, even outperforming the same manufacturer's extreme overclocking ReaperX lineup. Wintec AMPX memory came in second with the highest frequency at 1.80V, but failed to provide any benefit from boosting the voltage to 2.10V.
We were quite surprised that OCZ ReaperX modules were not able to overclock 2.10V better than 1.80V because they use powerful system cooling. However, OCZ is not the only company whose high-end modules yielded to slower models, since the Kingston PC3-10600 DIMMs provided better performance than the HyperX PC3-11000.
Now let me compare the performance of each kit, where we added the declared latencies (rated) in addition to the maximum overclocking on CAS 9. We'll start with the PC Mark 2005 memory test.
Is there any more test needed to prove that the fastest modules give better performance? Probably not, but here are the results. Yes, the 928 MHz OCZ ReaperX memory outperforms the 930 MHz Wintec AMPX somewhat, but this may be due to other latencies besides the four we manually set.
In the PC Mark 2005 test, the results coincide with the frequencies of the memory modules. Let's take a look at the SiSoftware Sandra memory benchmark.
Sandra's results again reflect the increase in memory frequency, although the 920 MHz Super Talent memory slightly outperformed the 930 MHz Wintec, which may, again, be related to latencies other than the four we manually set.
Of course, the main reason why DDR3 memory should be chosen when overclocking is to bypass the memory frequency limitations that can arise when the CPU frequency is increased. Given the slight difference in memory performance per clock, you should choose the fastest memory available for overclocking.
Problem with Boot Strap Modes
The next step in our testing is to find the most productive memory settings at a given clock speed, that is, the minimum latencies. Sounds relatively simple, but in reality this test requires many hours of testing to verify the stability of each pair of modules at each frequency.
Most of the modules tested can reach an effective clock speed of 1,600 MHz. An ideal solution for testing such modules will be the FSB1600 processor with memory frequencies of 1,600, 1,333, and 1,066 MHz. These frequencies correspond to commonly used DRAM to FSB multipliers of 2: 1, 5: 3 and 4: 3. Simple enough, right?
Unfortunately Intel does not publish every available divider at every available bus speed. The company chooses memory speeds based on its own considerations about what consumers need, and only supports them at each FSB mode.
| Intel X38 memory multipliers | ||||||||
| FSB mode | 1:1 | 6:5 | 5:4 | 4:3 | 3:2 | 8:5 | 5:3 | 2:1 |
| FSB800 | N / A | N / A | N / A | N / A | N / A | N / A | 667 | 800 |
| FSB1066 | N / A | N / A | 667 | N / A | 800 | N / A | N / A | 1066 |
| FSB1333 | 667 | 800 | N / A | N / A | N / A | 1066 | N / A | 1333 |
| FSB1600 | 800 | N / A | N / A | 1066 | N / A | N / A | N / A | 1600 |
To select a divider that Intel did not "bless" for a given frequency, you will have to select a different FSB frequency and overclock it.
But this raises a problem that seasoned overclockers are aware of - "Boot Strap" modes. The northbridge of the chipset operates at its own clock frequency, which depends on the FSB frequency. And each frequency level of the northbridge depends on the "Boot Strap". For example, for FSB800 the north bridge will operate at 200 MHz ("200 MHz Boot Strap"), and for FSB1600 at 400 MHz ("400 MHz Boot Strap"). Manually setting the 400 MHz FSB (FSB1600) when using the "Boot Strap" mode for the 200 MHz FSB (FSB-800) will overclock the north bridge by 100%.
| Intel X38 chipset memory multipliers with Boot Strap | |||||
| FSB mode | Boot Strap | Memory mode | Phys. memory frequency | Phys. FSB frequency | DRAM multiplier: FSB |
| FSB-800 | 200 | DDR2-667 | 333 MHz | 200 MHz | 5:3 |
| FSB-800 | 200 | DDR2-800 | 400 MHz | 200 MHz | 2:1 |
| FSB-1066 | 266 | DDR2-667 | 333 MHz | 266 MHz | 5:4 |
| FSB-1066 | 266 | DDR2-800 | 400 MHz | 266 MHz | 3:2 |
| FSB-1066 | 266 | DDR3-1066 | 533 MHz | 266 MHz | 2:1 |
| FSB-1333 | 333 | DDR2-667 | 333 MHz | 333 MHz | 1:1 |
| FSB-1333 | 333 | DDR2-800 | 400 MHz | 333 MHz | 6:5 |
| FSB-1333 | 333 | DDR3-1066 | 533 MHz | 333 MHz | 8:5 |
| FSB-1333 | 333 | DDR3-1333 | 667 MHz | 333 MHz | 2:1 |
| FSB-1600 | 400 | DDR2-800 | 400 MHz | 400 MHz | 1:1 |
| FSB-1600 | 400 | DDR3-1066 | 533 MHz | 400 MHz | 4:3 |
| FSB-1600 | 400 | DDR3-1600 | 800 MHz | 400 MHz | 2:1 |
Note, for example, that Intel no longer supports DDR2-533 (physical clock frequency 266 MHz), which means the company no longer provides a 1: 1 multiplier for the 266 MHz FSB1066. Besides, the X38 chipset supports "Boot Strap" FSB1600, but in this mode there is no 5: 3 multiplier, which is required for DDR3-1333 memory. To get a 5: 3 DRAM to FSB multiplier, you need to use the 200MHz "Boot Strap" instead of the 400MHz FSB1600 native.
The effect of choosing the wrong "Boot Strap" should not be underestimated, since neither the P35 nor the X38 can be overclocked by 100%, but even if it were possible, there would be a noticeable drop in overall system performance.
This did not allow us to use some "native" DDR3-1333 modules with a FSB1600 processor on motherboard The Gigabyte X38T-DQ6 because it automatically set the 400MHz FSB with a 5: 3 DRAM: FSM multiplier, which in turn resulted in a low 200MHz "Boot Strap" mode at a high 400MHz FSB. As a result, after 100% acceleration, the north bridge refused to load.
Therefore, we do not recommend using DDR3-1333 memory for processors with FSB1600 on the P35 chipset, but what about the X38? Our Asus board Maximus extreme set the "Boot Strap" mode to 400 MHz, which deprived it of the required 5: 3 DRAM: FSB multiplier, so the modules started working at DDR3-1066 frequency.
Due to the above mentioned limitations of the "Boot Strap" modes, we had to select different FSB frequencies for the DDR3-1333 and DDR3-1600 benchmarks. But how do you make the right comparison?
Since the 5: 3 DRAM: FSB multiplier is not available with the FSB1600 processor, DDR3-1333 cannot be tested either. Therefore, we had to compare DDR3-1333 and DDR3-1066 on FSB1333, and DDR3-1600 and DDR3-1066 on FSB-1600.
Only two CPU frequencies correspond simultaneously to FSB1333 and FSB-1600: 2.0 and 4.0 GHz. The CPU multipliers for 4.0 GHz on FSB1333 and FSB1600 are 12 x 333 MHz and 10 x 400 MHz, respectively.
| Test system for measuring minimum delays | |
| Motherboard | Asus Maximus Extreme Rev. 2.01G, Intel X38, BIOS 0501 (10/30/2007) |
| Socket 775 processor | Intel Core 2 Extreme QX9770 "Yorkfield", FSB1600, 45 nm, 3.20 GHz, 12 MB L2 cache |
| HDD | Western digital WD1500ADFD-00NLR1, firmware: 20.07P20, 150 GB, 10,000 rpm, 16 MB cache, SATA / 150 |
| Video card | Foxconn GeForce 8800GTX, P / N: FV-N88XMAD2-OD, nVidia GeForce 8800GTX - 768 MB |
| Power Supply | OCZ GameXStream OCZ700GXSSLI - 700W |
| System software and drivers | |
| OS | Windows XP Professional 5.10.2600, Service Pack 2 |
| DirectX version | 9.0c (4.09.0000.0904) |
| Platform Drivers | Intel INF 8.3.1.1009 |
| Graphics driver | nVidia Forceware 163.75 |
Since the Asus Maximus Extreme board proved to be more competent in overcoming the "Boot Strap" problem, we chose this board for the minimum latency test.
Quad-core processors use memory a little more efficiently than dual-core processors, and our latency test at maximum DDR3-1600 frequency matches the maximum memory multiplier available for FSB1600 processors. We used the only processor with a "native" FSB1600 bus, which is available today, namely, the Intel Core 2 Extreme QX9700 based on the Yorkfield core.
Game tests depend significantly on graphics performance, so we used the powerful GeForce 8800GTX from Foxconn.
While the hard drive's performance doesn't really affect the results of the selected tests, using the 10,000 RPM model clearly doesn't hurt. In this regard, the "ancient" 150 GB hard drive Western Digital Raptor still remains in the lead.
| Latency tests and settings | |
| 3D games | |
| F.E.A.R. | Version: 1.0 Retail Video Mode: 1024x768 Computer: Medium Graphics: Medium Test Path: Options / Performance / Test Settings |
| Quake 4 | Version: 1.2 (Dual-Core Patch) Video Mode: 1024x768 Video Quality: default THG Timedemo waste.map timedemo demo8.demo 1 (1 = load textures) |
| Sound | |
| Lame MP3 | Version 3.97 Beta 2 (12-22-2005) wave to mp3 160 kbps |
| OGG | Version 1.1.2 (Intel P4 MOD) Version 1.1.2 (Intel AMD MOD) Audio CD "Terminator II SE", 53 min wave to ogg Quality: 5 |
| Video | |
| TMPEG 3.0 Express | Version: 3.0.4.24 (no Audio) fist 5 Minutes DVD Terminator 2 SE (704x576) 16: 9 Multithreading by rendering |
| DivX 6.6 | Version: 6.6 Profile: High Definition Profile 1-pass, 3000 kb / s Encoding mode: Insane Quality Enhanced multithreading no Audio |
| XviD 1.1.3 | Version: 1.1.3 Target quantizer: 1.00 |
| Autodesk 3D Studio Max | Version: 8.0 Characters "Dragon_Charater_rig" rendering HTDV 1920x1080 |
| PCMark05 Pro | Version: 1.1.0 Memory tests Windows Media Player 10.00.00.3646 Windows Media Encoder 9.00.00.2980 |
| SiSoftware Sandra 2005 | Version 2005.7.10.60 Memory Test = Bandwidth Benchmark Lowest Latency Test Results |
Received minimum delays
We used a relatively safe voltage of 1.80 V, at which we determined the best latencies for the DDR3-1333 test modules while maintaining stable operation at effective memory frequencies of 1600, 1333, and 1066 MHz.
| Minimal latency while maintaining stable operation at 1.80V | ||||
| DDR3-1600 | DDR3-1333 | DDR3-1066 | Standard settings | |
| 9-8-8-15 | 8-7-6-13 | 6-5-5-10 | 8-8-8-15 | |
| Crash | 8-7-7-14 | 7-6-6-12 | 9-9-9-24 | |
| Crash | 7-7-6-13 | 6-6-5-12 | 8-8-8-24 | |
| 9-7-6-15 | 8-6-6-12 | 6-5-4-9 | 8-8-8-24 | |
| Mushkin EM3-10666 | 9-8-7-14 | 8-6-5-14 | 6-5-4-14 | 9-9-9-24 |
| OCZ Platinum PC3-10666 | 8-7-6-15 | 6-5-4-12 | 4-4-3-9 | 7-7-7-20 |
| OCZ ReaperX PC3-10666 | 8-7-6-13 | 6-5-4-12 | 5-4-3-8 | 6-5-5-18 |
| Patriot PC3-10666 | Loss of stability | 6-6-5-12 | 5-5-4-9 | 7-7-7-20 |
| Super Talent PC3-10600 | 7-6-6-13 | 6-5-5-10 | 5-4-4-9 | 8-8-8-18 |
| 8-7-6-15 | 6-5-4-12 | 5-4-3-9 | 9-9-9-24 | |
OCZ memory modules provided impressive 4-4-3-9 latencies at an effective memory frequency of 1066 MHz, and potentially inexpensive Wintec AMPX DIMMs were in the top three with two OCZ kits on DDR3-1333. Overclockers looking for the lowest latency at 1600MHz can use Super Talent 7-6-6-13.
Patriot DDR3-1333 modules were able to achieve stable operation at an effective frequency of 1,652 MHz on the top-end P35 motherboard from Gigabyte, but the Asus Maximus Extreme on the X38 chipset seems to be more demanding. On the new platform the modules could not even reach the frequency of 1600 MHz, but in terms of latency they were in second place in the DDR3-1333 category.
Reducing latency improves system performance. But to what level? We will learn about this from the following test results.
Test results with minimal delays
In DivX, the results turned out to be quite strange, since minimal delays did not always lead to victory. It looks like there is a small performance gain from the increase in frequency, but the results are too inconsistent to be analyzed in more depth.
XviD demonstrates a possible performance gain from a faster FSB, as well as a performance gain from a combination of a high-speed FSB with high memory frequencies. Delays on this test are not significantly affected.
Audio encoding in Lame does not show any noticeable performance gain from different memory frequencies and latencies.
Memory frequency and latency have little effect on OGG results. Taking into account the results obtained, it can be noted that the processor is the only limiting factor in performance in both audio coding programs.
Performance F.E.A.R. limited by other factors, and not by memory performance - most likely by the video card. However, hardly anyone will complain about this, since the frame rate is very high.
Quake 4 gives a tiny performance boost when installing high-speed modules, but latency seems to be of little effect.
3D Studio Max does not show a tangible performance gain from more speed memory or more severe delays. Again, the results seem to be dependent on raw processor performance.
Memory-only tests are perhaps the only ones where you can see noticeable performance gains with minor changes in latency. And in PC Mark 2005, Super Talent modules with excellent latency in DDR3-1600 mode are at the top. On the other hand, Mushkin's second place is weakly connected with the sixth position in DDR3-1600 latency.
Super Talent modules with surprisingly low latency in DDR3-1600 mode again took the lead in the first Sandra memory test, but Kingston ValueRAM took the second place, despite the fifth position in the lowest latencies.
Super Talent again takes first place in the second Sandra memory test due to its low latency. Mushkin is very strange second time come second.
We wanted our comparative testing DDR3-1333 memory turned out to be as valuable as possible, so we waited several months until enough memory modules appeared on the market so that we could collect a selection at the appropriate prices. Unfortunately, several of the modules we tested did not appear at attractive prices. However, before making a conclusion on the prices, let's take a look at the performance.
As we mentioned above, today the main reason for buying DDR3 memory is to remove barriers when CPU overclocking associated with slow memory. If you look at how much DDR2-1200 or faster modules cost, you will probably prefer the DDR3-1333 models.
Since our main goal was to overclock the memory, the OCZ Platinum Edition PC3-10666 modules took the lead here, defeating even the company's own line of ReaperX with an improved cooling system, as well as competitors. OCZ fans will take this for granted, but in my memory, OCZ modules are winning the overclocking competition for the first time. It is good to see that the company really confirms its good reputation, which was previously won mainly by marketing.
Buyers looking to overclock their memory to around 1,600 MHz should look at the Super Talent PC3-10600 modules. Or the humble looking Kingston ValueRAM modules.
This is the first time we've included Aeneon's memory modules in extensive benchmarks, and it's good to see Infineon's former memory division doing pretty well. So far, these modules have not received any awards, but the company can compete favorably with other kits in the mid-market segment in terms of price.
The Wintec AMPX PC3-10600 kit came in second in our overclocking tests, and while we couldn't find these modules on the market, we know this company will compete with OCZ on price. Again, buyers should weigh any decrease in overclocking potential with a corresponding decrease in price.
Within the framework of this article, I will tell you how to find out the frequency of RAM using at least two available methods.
The performance of the computer depends on what kind of RAM is installed in the computer. And almost all users know this. And the matter is not only in the volume, but also in the clock frequency of the boards themselves. Too slow speed can have a significant impact on how quickly your computer boots, run programs, or perform some tasks.
In addition, if you install two RAM dies at different speeds, they will be supported at a lower frequency, which can make it pointless to use a higher-quality fast motherboard. However, not all users know how to find out this same frequency. Therefore, below I will give two simple and feasible methods for each.
Note: It is also worth knowing that almost all programs for collecting computer characteristics provide such information. However, they can be too complex for ordinary users, so I did not describe them in the article.
Find out the clock frequency of the RAM from the sticker on the board
It is worth knowing that manufacturers often indicate special markings on the sticker of the RAM card. And from this marking you can find out the clock frequency of the board.
You need to look at the board and find a phrase like " DDRx - xxxx", where
DDRx is the generation of random access memory. Today, there are DDR, DDR2, DDR3, DDR4 and DDR5 (the first two are almost never found). Each of these technologies supports a specific frequency range. In this case with the picture it is "DDR3".
xxxx- this, in fact, is the clock frequency of the RAM. In this case, it is 1333 MHz.
Sometimes, the combination with DDR is not present on the sticker. In this case, you can simply write down the name of the board and see its characteristics on the Internet.
It should be admitted that this method may not be suitable for everyone (although it does not require the installation of programs), since it is not always possible to twist the board in your hands. Therefore, I turn to an alternative, simpler option.
How to find out the frequency of RAM using CPU-Z
First of all, you need to install the CPU-Z program. By the way, it is a very functional and useful program that allows you to quickly find out the main characteristics of your computer. So I recommend not to delete it, it may still be useful to you when solving other problems.
Run the program and then open the "Memory" tab. There, in the area called "Timings" there is a field "DRAM Frequency", in which the real (physical) frequency of the RAM will be displayed. It differs from what is stated in the characteristics. The latter is the effective frequency, which can be 2 or more times higher. The bottom line is that in one time step, more operations can be performed. Therefore, in order not to introduce additional marking codes, the real frequency is simply multiplied by this factor. For example, for DDR3 this factor is 2. Thus, with a real frequency of 667 MHz, the effective frequency is 1333 MHz. And if we look at the example from the picture, then the real frequency of the RAM is "798.2 MHz", that is, about 800 MHz, which corresponds to a DDR3-1600 board.
Note: An important point, manufacturers usually indicate the effective frequency.
Below is a table of frequencies for generation operating boards DDR3:
| Real memory frequency | Effective (doubled) frequency | Standard name of RAM | Peak data rate | Module name |
|---|---|---|---|---|
| 400 MHz | 800 MHz | DDR3-800 | 6400 MB / s | PC3-6400 |
| 533 MHz | 1066 MHz | DDR3-1066 | 8533 MB / s | PC3-8500 |
| 667 MHz | 1333 MHz | DDR3-1333 | 10667 MB / s | PC3-10600 |
| 800 MHz | 1600 MHz | DDR3-1600 | 12800 MB / s | PC3-12800 |
| 933 MHz | 1866 MHz | DDR3-1866 | 14933 MB / s | PC3-14900 |
| 1066 MHz | 2133 MHz | DDR3-2133 | 17066 MB / s | PC3-17000 |
If you use another generation of DDR boards, then you need to look at the corresponding table (the DDR number is indicated in the same tab in the "General" area in the "Type" field).
