OCZ VX Revisited: DDR Updates on DFI nForce4
by Wesley Fink on March 30, 2005 12:05 AM EST- Posted in
- Memory
Performance Test Configuration
All four memories were tested on the DFI LANParty nF4 SLI-DR. The new DFI nForce4 SLI and Ultra are the only current production boards that support the voltages required for top VX performance. An OCZ DDR Booster can be used with motherboards without support for high memory voltages. nForce4 is PCI Express, so we used the PCIe version of our standard nVidia 6800 Ultra for testing. Other components remain the same as used in the memory setup in Athlon 64 Memory: Rewriting the Rules.The A64 test bed includes components that have been proven in Socket 939 Athlon 64 benchmarking, such as the Socket 939 4000+ (same specifications as FX53), the OCZ Power Stream 520 Power Supply, and the nVidia 6800 Ultra. We have found the 6800 Ultra to be a particularly good performance match to nVidia motherboards.
All other basic test conditions attempted to mirror those used in our earlier Intel memory reviews. However, test results are not directly comparable to tests performed on the Intel test bed.
| AMD nForce4 Performance Test Configuration | |
| Processor(s): | AMD 4000+ (FX53) Athlon 64 (2.4GHz, Socket 939, 1 MB cache, Dual Channel, 1000HT) |
| RAM: | OCZ EL PC4000 VX Gold (DS) 2X512MB Corsair TwinX1024-4400C25 (DS) 2X512MB Crucial Ballistix (DS) 2X512MB OCZ PC3200 Platinum Rev 2 (DS) 2X512MB |
| Hard Drives | Seagate 120GB SATA 7200RPM 8MB Cache |
| PCI/AGP Speed | Fixed at 33/66 |
| Bus Master Drivers: | nVidia nForce Platform Driver 6.39 |
| Video Card(s): | nVidia 6800 Ultra 256MB PCIe, 256MB aperture, 1024x768x32 |
| Video Drivers: | nVidia Forceware 71.84 Release |
| Power Supply: | OCZ Power Stream 520W |
| Operating System(s): | Windows XP Professional SP1 |
| Motherboards: | DFI LANParty nF4 SLI-DR |
| BIOS: | N4D310p Release (3/10/2005) |
With nForce3 motherboards, we achieved the fastest performance on AMD Athlon 64 chipsets (nForce3, VIA K8T800 PRO) at a Cycle Time or tRAS of 10. However, as we saw in the recent nForce4 SLI roundup, the nForce4 appears to behave a bit differently with memory. In the SLI roundup, we found a tRAS of 7 to be ideal for memory based on Samsung TCCD chips. Therefore, we ran a complete set of Memtest86 benchmarks with only tRAS varied to determine the best tRAS setting for OCZ VX.
| Memtest86 Bandwidth DFI nForce4, Athlon 64 4000+, OCZ VX |
|
| 2 tRAS | 2572 |
| 3 tRAS | 2572 |
| 4 tRAS | 2572 |
| 5 tRAS | 2640 |
| 6 tRAS | 2640 |
| 7 tRAS | 2640 |
| 8 tRAS | 2505 |
| 9 tRAS | 2505 |
| 10 tRAS | 2505 |
| 11 tRAS | 2441 |
| 12 tRAS | 2383 |
These tests are particularly easy to do with the DFI nF4 boards, since memtest86 is built-in to the BIOS. To boot memtest86, you only have to enable it in BIOS and the system will boot directly into memtest86. This makes it very easy to test various memory timings, but memtest86 should be disabled in BIOS when you are ready to boot into the system.
| Memtest86 Bandwidth DFI nF4, Athlon 64 4000+, OCZ 3200 Platinum Rev. 2 |
|
| 4 tRAS | 2191 |
| 5 tRAS | 2191 |
| 6 tRAS | 2242 |
| 7 tRAS | 2191 |
| 8 tRAS | 2191 |
| 9 tRAS | 2141 |
| 10 tRAS | 2092 |
Similar tests were also run on Crucial Ballistix, Corsair PC4400, and OCZ Platinum Rev.2. All 3 memories showed their best bandwidth at a tRAS setting of 6 - just like OCZ VX. Therefore, a tRAS setting of 6 was used for testing wherever possible.
Test Settings
All AMD Athlon 64 processors are unlocked downward, and the FX CPUs are unlocked up and down. This feature allows a different approach to memory testing, which truly measures performance differences in memory speed alone. All tests were run with CPU speed as close to the specified 2.4GHz of the 4000+/FX53 as possible, with CPU speed/Memory Speed increased at lower multipliers to achieve 2.4Ghz. This approach allows the true measurement of the impact of higher memory speed and timings on performance, since CPU speed is fixed, removing CPU speed as a factor in memory performance.The following settings were tested with the OCZ EL PC4000 VX Gold on the DFI nF4 test bed:
- 12x200/DDR400 - the highest stock memory speed supported on K8T800-Pro/nF3-4/SiS755-FX motherboards
- 11x218/DDR436 - a ratio near the standard DDR433 speed
- 10x240/DDR480 - a ratio near the standard rating of DDR466
- 9x267/DDR533 - a standard memory speed used in testing other high-speed memory
- Highest Memory Performance - the highest memory bandwidth and game performance that we could achieve with the memory being tested. This is rarely the highest memory speed that we could achieve; it is normally a lower speed with 1T Command Rate and tighter memory timings.
We ran our standard suite of memory performance benchmarks - Quake 3, Return to Castle Wolfenstein-Enemy Territory-Radar, Super Pi 2M, and Sandra 2004 Standard and UnBuffered.
Facebook
Twitter
RSS
53 Comments
View All Comments
AnnihilatorX - Wednesday, March 30, 2005 - link
This memory rocks. Can someone please do me a favour of buying this memory for me from newegg, ship it to me (UK/HK) and I'll pay him cost + postage + a bit extra via paypal. Thanks very much. It's a bit annoying since you cannot buy this ram from HK. No suppliers.Just e-mail me at annihilator@x-annihilation.com
Thanks again
ozzimark - Wednesday, March 30, 2005 - link
*notes that it's 245mhz at 2.5-2-2-10ozzimark - Wednesday, March 30, 2005 - link
#28my ballistix can go to 245mhz with 2.7v before my memory controller starts hating me and having some serious stability issues. though they seem to scale nicely with voltage, i can't really test it out, because even at 3-4-4-10 with 2T, i can't get over 250mhz.
n yusef - Wednesday, March 30, 2005 - link
AnandTech should do a review of the TwinMOS UTT chips that cost $150 less than the VX. I bought VX before I heard about the TwinMOS stuff, but my friend is doing 2-2-2-6 at 255MHz with 3.4 volts at 1:1. My VX can do 260MHz, but I would give 5MHz RAM/FSB bandwidth and 50MHz core for $150 any day.JoKeRr - Wednesday, March 30, 2005 - link
one more thing, I've heard from many ppl that they're running their BH5 sticks (2x512) to 250 2-2-2-5 with like 3.1 to 3.2V, if u still do have some old BH5, would it be possible to lay a show down between bh5 and VX?? I think I saw a guy on xtremeresouce /or maybe system doing ddr293 with 4V with mushkin BH5 at 2-2-2-5 timing, that's crazy.JoKeRr - Wednesday, March 30, 2005 - link
How does the Ballistix react to voltage?? I've heard that Tccd doesn't react very well, but the old EB stuff could do 2.5-2-2-5 at ddr500 once u give it 3.1V. Could you plz try that and see how the Ballistix scale with voltage plz?? thankQ in advance.JoKeRr - Wednesday, March 30, 2005 - link
Rapsven - Wednesday, March 30, 2005 - link
23-If you are trying to get 2 gigabytes to do those speeds in tandem with the CPU, you'll have problems. Try using the regular 2*512 instead of 4*512 and see you can hit 250 fsb at 2-2-2.
Or ask the OCZ guys, they're very helpful.
ozzimark - Wednesday, March 30, 2005 - link
hmm, my bad, the [img] tags don't work here. just copy and paste that link :)and since it seems that some posts have been deleted, my above comment is in response to #13
ozzimark - Wednesday, March 30, 2005 - link
#14-that's quite interesting. maybe the spd's on the ram sticks are changing important values other than the cas, trcd, trp? i know that there are a lot of values that effect ram performance other than those three. i actually went through most of them on the dfi board, and found that lower isn't always better, similar to tras. though with that you said, i may have to go through and re-tweak my setup with the write benchmark in mind too. i never really gave it much thought assuming that it would scale similarily to read bandwidth.
also, i noticed some inconsistancies at various multipliers. i tested at two speeds, 230mhz (near the limit of my memory controller) at 2.5-2-2-10, and 200mhz at 2-2-2-10. (i use 2x512mb of crucial ballistix)
the only thing changed between runs is the cpu mutliplier. the ram is 1:1, and everything else remains constant.
[img]http://www.freewebs.com/cfeclipse/latency.PNG[/img]