Performance Test Configuration
The six 2GB kits were tested with the DFI LANParty nF4 SLI-DR Athlon 64 Socket 939 motherboard. 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 90nm 4000+, the OCZ Power Stream 520 Power Supply, and the NVIDIA 7800 GTX video card. We used an AMD 4000+ 90nm processor with a Revision E type memory controller. All other basic test conditions attempted to mirror those used in our earlier Athlon 64 memory reviews.
Since the video card, and platform/video drivers are upgraded from past memory testing, we found that the results were not completely consistent with past memory test results. For that reason, we retested the three 2GB kits that were previously tested. Results are still broadly comparable to past memory test results in earlier reviews, but we decided to include just the nine 2GB kit results in this roundup.
As discussed in other memory reviews, we ran a complete set of Memtest86 benchmarks with only tRAS varied to determine the best tRAS setting for these memories on the nForce4 chipset. The NVIDIA nForce4 has the best bandwidth at tRAS settings ranging from 5 to 8. Therefore, a tRAS setting of 7 was used for testing.
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+ as possible, with CPU speed/Memory Speed increased at lower multipliers to achieve 2.4 GHz. 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 six 2GB kits on the DFI nF4 test bed:
The six 2GB kits were tested with the DFI LANParty nF4 SLI-DR Athlon 64 Socket 939 motherboard. 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 90nm 4000+, the OCZ Power Stream 520 Power Supply, and the NVIDIA 7800 GTX video card. We used an AMD 4000+ 90nm processor with a Revision E type memory controller. All other basic test conditions attempted to mirror those used in our earlier Athlon 64 memory reviews.
Since the video card, and platform/video drivers are upgraded from past memory testing, we found that the results were not completely consistent with past memory test results. For that reason, we retested the three 2GB kits that were previously tested. Results are still broadly comparable to past memory test results in earlier reviews, but we decided to include just the nine 2GB kit results in this roundup.
| AMD nForce4 Performance Test Configuration | |
| Processor(s): | AMD 4000+ Athlon 64(90nm) 2.4GHz, Socket 939, 1 MB cache, Dual Channel, 1000HT |
| RAM: | TWINX2048-4000PT (DS) 2X1GB Crucial Ballistix CLIII5N.32 (DS) 2X1GB G. Skill F1-3200PHU2-2GBZX (DS) 2X1GB KHX3200AK2/2G (DS) 2X1GB Mushkin 2GB Redline XP4000 (DS) 2X1GB Team XTreem TXDR 1024M400HC2 (DS) 2X1GB Corsair TWINX2048-3500XL PRO (DS) 2X1GB Gigaram 2GB Dual Channel PC-4200 (DS) 2X1GB OCZ PC4000 2x1024MB EB Platinum (DS) 2X1GB |
| Hard Drives: | Seagate 120GB SATA 7200RPM 8MB Cache |
| PCI/AGP Speed: | Fixed at 33/66 |
| Bus Master Drivers: | NVIDIA nForce Platform Driver 6.70 |
| Video Card(s): | NVIDIA 7800 GTX 256MB PCIe, 256MB aperture, 1024x768x32 |
| Video Drivers: | NVIDIA Forceware 81.98 Release |
| Power Supply: | OCZ Power Stream 520W |
| Operating System(s): | Windows XP Professional SP2 |
| Motherboard: | DFI LANParty nF4 SLI-DR |
| BIOS: | 7/06/2005 |
As discussed in other memory reviews, we ran a complete set of Memtest86 benchmarks with only tRAS varied to determine the best tRAS setting for these memories on the nForce4 chipset. The NVIDIA nForce4 has the best bandwidth at tRAS settings ranging from 5 to 8. Therefore, a tRAS setting of 7 was used for testing.
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+ as possible, with CPU speed/Memory Speed increased at lower multipliers to achieve 2.4 GHz. 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 six 2GB kits on the DFI nF4 test bed:
- 2.4GHz-12x200/DDR400 - the highest stock memory speed supported on nF3-4/SiS755-FX/ATI Radeon Xpress 200/VIA 939 motherboards
- 2.4GHz-11x218/DDR436 - a ratio near the standard DDR433 speed
- 2.4GHz-10x240/DDR480 - a ratio near the standard rating of DDR466
- 2.4GHz-9x267/DDR533 - a memory speed achieved by only a few top memories on the Athlon 64
- 2.4GHz-8x300/DDR600 - with recent improvements in OC abilities, a very few memories can reach this next multiple with the 8 multiplier
- Highest Memory Performance - the highest memory bandwidth and game performance that we could achieve with the memory being tested. This is the highest memory speed that we could achieve with 1T Command Rate.
Facebook
Twitter
RSS
51 Comments
View All Comments
PrinceGaz - Monday, January 23, 2006 - link
"... 2GB also makes sense for the OS future. With Vista coming and 2GB, the new memory recommendation for the new OS, you will be set for Vista."I was under the impression that the minimum is expected to be 512MB, and that 1GB will be recommended for good responsiveness under normal use. Having said that, I would personally choose 2GB for any new system as pairs of 1GB DDR (and of course DDR2) modules are now readily available and competitively priced. Unfortunately most big-name OEM systems are still shipping with either 512MB, or if you're lucky 1GB.
bigtoe36 - Monday, January 23, 2006 - link
Being honest we have had some supply issues, and some issues regarding supply that are out of our control. We do have stock and will have more real soon.ocztony
bigtoe36 - Monday, January 23, 2006 - link
This was to counter any moans about availability before they were posted.emilyek - Monday, January 23, 2006 - link
You guys tested the worst GSKILL kit.Their pc4000 2 gig kit (speed-binned version of that kit) is the same price as the 2-3-2-5 chip ($200) and will do DDR580
Their $250 kit 2-3-2-5 would contend with the Mushkin and the Crucial at a much lower cost.
Wesley Fink - Monday, January 23, 2006 - link
The Silver Editors Choice Corsair 4000PT, and the OCZ PC4000 Gold and PC4000GX XTC, also based on UCCC chips, are all in the $205 to $230 price range for the 2GB kit from e-tailers. So $250 for a UCCC kit seems in line but hardly a bargain.As for testing the worst G. Skill, we asked all the memory manufacturers to supply their "best" 2GB kit for our roundup. The manufacturers decided what to supply, and we expect Enthusiast memory makers to know what is best in their line.
irev210 - Monday, January 23, 2006 - link
The crucial memory is amazing memory.I was able to get it stable at 3-3-3-8 @ 300mhz, but you really need to spend some time on the settings, it is VERY picky memory. Both tras and drive strength play huge factory in memory stability.
Yes, it is suspected that due to really high RMA they discontinued, however what anandtech doesnt report is what actually happens to the memory.
Even at 2.8vcore or less PLUS active cooling, 1 2 or 3 months down the line your memory will just start dying.
Thinks like memtest86 will check out fine, but then while gaming, your rig will just lock up. Random bluescreens, random lockups...
There is a big story to these ballistix, it is too bad they simply have a HUGE rate of failure. Do a ddr2 roundup, then you will see how many people use micron fatbodies for their high performance ddr2 memory. You dont see that with the high density memory, and obviously i suspect due to stability issues.
My advice for memory is the g.skill F1-4000USU2-2GBHZ based on samsung ram, it has been doing well, and g.skill has awesome customer service.
I personally am going to wait for ddr2 before going with 1gb sticks, then ballistix 2gb set is on my list. But that is just me.
Bull Dog - Monday, January 23, 2006 - link
I'm another Ballistix memory user here. I ran mine at 2.75v the whole time and they just died after a while. Screw you Crucial/Micron. I will NEVER buy your products AGAIN. (I don't care if they have zero problems eitehr.gooser - Monday, January 23, 2006 - link
Now what about the tracers?? I just purchased the 2 GB tracers last week knowing about the issues with the regular ballistix after a couple of months. I have read that the tracers use a higher quality power modulator, thus dont have the same problem. Anyone?lopri - Monday, January 23, 2006 - link
Lots of work and outstanding comments. Thanks. This review, along with the Part I, is significant in that it puts an end to the 1GB system memory standard. It's declaring that an *era* of 2GB system memory is finally here.However, with these super high-clocking 1G sticks (512MB sticks also for that matter), there is one *realer* question that's haunting my mind. With the current crop of DDR memory reaching speed of DDR550~DDR600, one has to wonder - What's the good if CPUs can't handle the speed of memory?
Yes, I'm questioning the capability of A64/Opteron's memory controller. Because even the best current AMD CPUs often fail to keep up with the speed of memory. You've got a PC4800 DDR memory? Good luck, pray your CPU can run it at that speed. This issue is, at least for me, very real. I just don't see the point of high-clocking memory when the truth is CPUs' memory controllers are the limit. (And to some extent motherboards, of course)
In the light of this, would it be possible for you to test the quality of each revision of AMD CPUs' memory controllers? Preferably with recent cores - namely, Venice, San Diego, Toledo, Manchester and their Opteron equivalents. The more detailed, the better. Rev E3, E6, E4, and even the steppings.
While reading this excellent review, I couldn't shake the issue of memory controller off my head. These days a stick of RAM seems only as good as the CPU's memory controller it relies on. Unless one is just happy to see her/his RAM passing memtest @DDR600 all day long.
Again, thanks for your hard work on this review, Wesley. However, I'd really like to hear from you about the issue that I'm bringing here. I'd much appreciate it.
lop
dlerious - Monday, January 23, 2006 - link
If my CPU can't keep up with my RAM, I just drop the multiplier (and/or divider) down until I find the point where both run as high as possible at the voltage and temps I'm comfortable with.