Our test setup hasn't changed from the previous reviews, have a look :
As we explained in our previous
reviews, the "Generic PC4300" module is a 533Mhz part with quite basic timings, at the time of writing. This "Generic" module does not really exist: we constructed this module with the Corsair Dominator modules, but slowed the modules down to 533 Mhz DDR with (relatively) slow timings. The timings we "arranged" are 4-4-4-15 timings. You might wonder why we chose exactly this speed and these timings. We could have picked an even slower part with PC3200 speeds or 5-5-5 timings, but than again, these modules are barely available anymore. The cheapest memory kits you can find nowadays are almost all capable of doing 533Mhz with 4-4-4 timings, so in our opinion, this seemed like a nice performance base to compare our memory kits with.Test methodology
Every module we test, except for the above mentioned generic module, was put to the test at five different speeds:PC6400 - 800Mhz performance, at the best possible timings
PC8500 - 1066Mhz performance, at the best possible timings*
Maximum cas 3 performance
Maximum cas 4 performance
Maximum cas 5 performance
*1066 Mhz is tested only for the modules for which this is possible, in this case all the modules were capable of reaching this speed, except for the PC6400 kits from OCZ and Kingston. These PC6400 kits are omitted in the DDR1066 benchmark tables.
For the 800Mhz and 1066Mhz tests, settings are very simple: in bios only the memory speed was turned up to the appropriate setting, while front side bus (FSB) and cpu speed (and multiplier) remained at the stock settings of 266FSB and 2.66Ghz respectively.
The maximum cas "x" performance is measured at other than stock cpu and FSB settings, to be able to give the memory sufficient bandwidth, and to play down the influence of the cpu in comparison to the memory. In our tables on page 6 you'll see that the cpu speeds and FSB setting is always mentioned at the bottom, so please bare that in mind when comparing the results. So to be clear: results cannot be directly compared, as the cpu/Fsb factor accounts for a certain % of the performance shown.
For the cas 3 setting we test with a CPU multiplier of 9, and a memory divider of 4:5 or 667Mhz in bios. For the cas 4 and 5 setting we test with an CPU multiplier of 8, and a memory divider of 2:3 or 800Mhz.
A word on Asus P5B-deluxe performance
The motherboard that we are using for all our memory testing is the Asus P5B deluxe/wifi-AP, a motherboard equipped with the Intel 965 chipset. We chose specifically this motherboard because it is capable of attaining very high front side bus speeds compared to motherboards equipped with the Intel 975 or (older) nVidia chipsets. Among the 965 motherboards, the Asus is also one of the better performing boards, often reaching FSB speeds of more than 500FSB. Of course, when we want to test memory overclocking, we need a motherboard that is capable of this high FSB, and one that is compatible with high frequency ram (the 975 chipset is known to give problems with ram over 1000Mhz, sometimes).
Now how does the Intel 965 chipset reach these high front side bus speeds? You must realize that, when you start pumping up the FSB of your motherboard, you are actually overclocking your chipset. Just like memory, chipsets have standard timings and standard voltages, and when you're overclocking you can give a little more voltage to the chipset, in order to make it scale higher. Normally, once you hit 380-430FSB on the Intel 965 and the 975 chipset, you hit an overclocking wall (unless you do some more exotic, warranty voiding trickery of course...). How does Asus reach 500FSB than? That’s where Tony "Bigtoe" comes in : he has been hard at work to get the public familiar with the technology behind the 965 chipset, and one of the things he found out was that the P5B automatically relaxes (slows down) the timings of the chipset itself in order to be able to scale above 400FSB. There seem to be certain "sets" of timings which are implemented at different FSB settings, and these sets are called "straps". The Asus P5B now sets a different strap when you boot at 400+FSB, than when you boot at 400-FSB, the 400+ strap being the slower of the two. This obviously has a direct influence on your memory bandwidth and on the entire performance of your PC. This will show up in this review as well, when we're looking into the performance of the Team Group's ram for example, which scaled beyond 400FSB on our settings, and in the process gives worse results than you would actually expect...
Take a look at this and this thread by Tony Bigtoe over at the support forums of BleedinEdge, and also at this thread at the XtremeSystems forums. Finally, here's a link to Tony's P5B-deluxe tweaking guide, from the excellent Tech Repository forums which have been recently founded by Kris "Freecableguy" and Tony "Bigtoe". These threads are in my personal opinion obligated reading for anyone considering to buy a top performing memory kit, just to understand the performance of your PC setup.
The goal of this section of the review is to look how far the modules are willing to go. For this we played around with different memory latencies a bit, to end up with the best cas 3, cas 4 and cas 5 performance of these particular modules.
In our experience, Orthos is the most demanding stability test we can run. Orthos is the dual core version of the better known Stress Prime 2004, or SP2004 in short, which in his term is based upon the famous "Prime95" code. Even when all tests succeeded, Orthos testing would sometimes fail. On the other hand, when Orthos succeeded in running for about 1/2 hour, other tests would (almost) never become a problem.
So how did our modules of the day do ?
Our OCZ PC8500 kit is rated to work at speeds of 1066Mhz DDR2, and this with 5-5-5-15 timings at a voltage of 2.3V. As with all previous reviews, when testing we try to comply with the warranties that are provided by the manufacturers. In this case OCZ warrants its modules for an EVP (Extended Voltage Protection) of 2.35 Volts, with a margin of 5%. This means we can apply up to 2.4 volts on our memory without voiding the warranty. We did apply 2.4 Volts to the modules for a short time to be able to find out the maximum performance of the modules (while monitoring voltage with a multimeter), and to be able to provide you with the next section.
Let's first look at the rated timings again and there's certainly some extremely good news here. At 2.4V, the OCZ PC8500 kit went all the way up to 1218Mhz DDR, with timings of 5-5-5-15. This makes this kit the second fastest kit we ever tested, exactly on par with the excellent Team Group PC6400 kit. This is an absolutely great result, even more so when you consider the ease with which it was reached. The voltage applied was also lower than that of the Team Group ram, which means theoretically the memory could scale even higher with additional voltage.
Next up is the cas 4 setting: at cas 4 our memory reached 1031Mhz DDR with 4-4-4-15 timings, and 2.4 volts. This result was, compared to the excellent cas 5 results, maybe a little disappointing, but the memory does outperform the similarly priced Kingston modules by quite a margin. Compared to the older Micron-equipped memory modules, the newer OCZ Reaper modules are at the bottom of the pack.
Finally, at the cas 3 setting, we fall just short of the 800Mhz DDR mark. With 795Mhz at 3-3-3-10 and 2.4 Volts we hit almost exactly the same wall as with the Kingston PC8500 modules. The modules are at the back of the Micron pack this way, however this is still great performance, which cannot easily be obtained by lower rated memory (without Micron chips).
After testing the Reapers and the Kingston modules before, I started wondering why performance on tighter latencies has suffered a little over the last revisions of Micron memory. After asking Tony (Bigtoe) about this, he explained to me that a newer PCB is in use since the introduction of the PC9200 Flex-XLC modules, which indeed corresponds to our findings regarding performance on lower latencies. The new 8 layer PCB's do not seem to "like" lower latencies as much at least that was what I was thinking. Tony did not agree though, and told me I should continue testing the modules on other motherboards, as performance is often found to be better on Nvidia 680i boards with the newer PCB's. He told me that I hit the limitations for the board rather than the memory, which is an interesting point of view. Anyway, performance on nForce 680i is something you and I will indeed be finding out in the very near future...
Again you can find the screenshots of the Orthos runs below, at the maximum memory speeds we achieved, for the different cas and system settings.
Left to right (click to open) : Maximum CAS3, CAS4, CAS5
Last but not least we have to mention that the OCZ Reaper modules are fully nVidia SLi-EPP certified, which means that on a compatible nForce 590 or 680 motherboard, the best performing settings are automatically chosen, at least when you set this function to "enabled" in bios. This way people that don't have or don't want to take the time to test the memory to its maximum can still enjoy great performance from these modules, without any hassle.
On to the benchmark results now ->