Socket 754 Roundup: Comparing Generation 2
by Wesley Fink on May 28, 2004 5:57 PM EST- Posted in
- Motherboards
Gigabyte K8NSNXP: Overclocking and Stress Testing
FSB Overclocking Results
| Front Side Bus Overclocking Testbed | |
| Default Voltage | |
| Processor: | Athlon 64 3200+ 2.0GHz |
| CPU Voltage: | 1.5V (default) |
| Cooling: | AMD Stock Athlon 64 Heatsink/Fan |
| Power Supply: | Antec TruePower 430W |
| Maximum OC: (Standard Ratio) |
242FSB x10 2420MHz (+21%) (246 with EZTune4) |
| Maximum FSB: (Lower Ratio) |
242FSB x 10 at 1:1 Memory (275x8.5 at 1:1 with EZTune4) |
On paper, the Gigabyte K8NSNXP should be an incredible performer, but we found that no matter what we did, the Gigabyte would not boot at any ratio, at a speed higher than a BIOS setting of 242. This seems to be an artificial boot limiter, since we were able to reach a very competitive 275x8.5 using EZTune4 in Windows XP. While we were happy to see this level of performance with EZTune4, it is still a much less convenient method of overclocking than tuning the same settings in BIOS. EZTune4 does allow settings to be saved, so Windows will load at the same overclock on reboot, but we really wish that Gigabyte would give us a choice on how we wish to overclock. Gigabyte has been using what appears to be BIOS boot limits on several recent boards, probably to limit the number of boards destroyed by indiscriminate overclock settings in BIOS. We do understand why, but we still prefer to dial our own, as do most other overclockers.
The other complaint with recent Gigabyte boards is the extremely limited range of memory voltage adjustments that are provided. Memory is now available that is warranted to 3.0V to 3.2V, and an adjustment of just +0.2V for vDIMM is just not enough for some memory. The Chipset and HT voltage adjustments partially offset this limitation, but they are not an adequate substitute for such a limited selection of vDIMM settings. Gigabyte builds very reliable motherboards, and with a little more attention to memory voltage and BIOS overclocking, they would consistently be at or near the top in every roundup.
Memory Stress Test Results:
The memory stress test simply checks the ability of the Gigabyte K8NSNXP to operate at its officially supported memory frequency (400MHz DDR), at the best performing memory timings that our Mushkin PC3500 Level 2 or OCZ PC3500 Platinum Ltd Modules will support. Memory stress testing was conducted by running RAM at 400MHz with 2 DIMM slots filled.| Stable DDR400 Timings - 2 DIMMs (2/3 DIMMs populated) |
|
| Clock Speed: | 200MHz |
| Timing Mode: | N/A |
| CAS Latency: | 2.0 |
| Bank Interleave: | N/A |
| RAS to CAS Delay: | 2T |
| RAS Precharge: | 10T (10T for Best Performance)* |
| Precharge Delay: | 2T |
| Command Rate: | N/A |
*Several memory tests have shown that memory performs fastest on the nVidia nForce and VIA K8T800 chipsets at a TRas (RAS Precharge) setting in the 9 to 13 range. We ran our own Memory Bandwidth tests with memtest86 with TRas settings from 5 to 15 at a wide range of different memory speeds. The best bandwidth was consistently at 9 to 11 at every speed, with TRas 10 always in the best range at every speed. The memory bandwidth improvement at TRas 10 was only 2% to 4% over TRas 5 and 6 depending on the speed, but the performance advantage was consistent across all tests. Since best performance was achieved at 2-2-2-10 timings, all Athlon 64 benchmarks were run at a TRas setting of 10.
The Gigabyte K8NSNXP was completely stable with 2 DIMMs at the best performing settings of 2-2-2-10 at default speed, but we did need to boost memory voltage slightly by +0.1V for best stability. Higher overclocks could be achieved with 1 DIMM compared to 2 DIMMs, but at default speed, 1 or 2 DIMMs were reliable at the same aggressive 2-2-2-10 timings.
Filling all three available memory slots is more strenuous on the memory subsystem than testing 2 DIMMs on a motherboard. As we found on most other nF3-250 boards, 3 DS DIMMs (1.5GB) of memory worked fine at timings almost the same as the aggressive timings that we used for 2 DIMMs.
| Stable DDR400 Timings - 3 DIMMs (3/3 DIMMs populated) |
|
| Clock Speed: | 200MHz |
| Timing Mode: | N/A |
| CAS Latency: | 2.0 |
| Bank Interleave: | N/A |
| RAS to CAS Delay: | 3T |
| RAS Precharge: | 10T |
| Precharge Delay: | 2T |
| Command Rate: | N/A |
Three DIMMs required 2-2-3-10 timings for stable operation, but they remained stable at the +0.1V memory voltage that was required with 2 DS DIMMs. 2-2-3 is very close to the best performing 2-2-2-10 timings with 2 dimms.
Facebook
Twitter
RSS
30 Comments
View All Comments
karlreading - Monday, August 30, 2004 - link
i braught the chaintech vnf3-250 based on the review it recieved at AT and TBPH its a cracking motherboard, plus it easily gets my 2ghz a64-3200+ to 2.4 ghz and lets you run 1ghz HT if u running the cpu stock, something whic supposidly only the nf3 ultra let u do.all in all a very fast, very reliable, very overclockerbul motherboard which i have NMO regrets about buying :)lem79 - Friday, August 6, 2004 - link
I have the Epox 8KDA3+ here with an Athlon64 2800+, runs nice, except for one quirk, which Anandtech failed to mention here (they probably didn't even know)..Cool'n'Quiet only works on this board when _one_ DIMM slot is in use (that is, DIMM1). If there's RAM in DIMM2 and DIMM3 slots, Cool'n'Quiet gets disabled by the BIOS (July revision, earlier BIOS images leave it enabled, but Cool'n'Quiet activation causes system instability and lockups). Epox themselves told me that this was the case.
I think the review needs updating..
operator - Sunday, June 6, 2004 - link
I have the same question as #25. with a small addition.when will the msi k8n or the epox board be available in Canada?
Zebo - Wednesday, June 2, 2004 - link
This is a very good review Wesley, thanks. I really like how you went into detail with memory and boards FSB capabilites for the clockers out here.:) Looks like the Chaintech is the board for me. Cheap and a real performer. It's really to bad nVidia failed to have a decent sound solution though.Sk0t - Tuesday, June 1, 2004 - link
Nice review.... BUT Im unable to understand why the Epox 8KDA3+ was awarded gold, and MSI K8n Neo Platinum only awarded silver ?Epox pros over MSI:
6th pci-slot
Slightly better overclocking
MSI pros over Epox:
Firewire
3 working dimm sockets (at 400mhz)
Slightly better layout (dimm-sockets)
Supports Cool&Quiet*
*Since the review seems to completly ignore cool&quiet (unless i missed it?) I will stick to my current rule of thumb, that only Asus & MSI fully supports this feature
Did I miss some wonderfull feature on the Epox ? or was slightly better overclocking considered more important than features & layout ?
gmenfan - Tuesday, June 1, 2004 - link
Does anyone know when the MSI K8N Neo will be available? Thanks.mbf - Tuesday, June 1, 2004 - link
Nice review! I just have one question; how can there be boards *without* ECC support, when the A64 has the memory controller on-die and it *inherently* supports unbuffered ECC memory? I'm especially thinking of the Epox and MSI boards, since they interest me the most (hardware firewall et al).Actually, I wrote to MSI (US) about this and was told that indeed ECC is supported and the "non-ECC" statement on the product spec page was erroneous. The page was immediately updated (http://www.msicomputer.com/product/p_spec.asp?mode... However, only the US site has been updated, and neither the manual (PDF) nor the BIOS available from the MSI Taiwan site state ECC support. Neither do the manual or BIOS for the Epox board.
As ECC support is pretty important to me having had some bitter experiences with regular memory, I was wondering if some kind person at Anandtech might be persuaded to throw in a stick of unbuffered ECC memory into either the K8N Neo or the Epox 8KDA3 and let me know if the memory is detected as such? I'd really appreciate the effort!
I know the ASUS K8N-E Deluxe (http://www.asus.com.tw/products/mb/socket754/k8n-e... states ECC support, but it's anyone's guess when that board will be out, since they cannot even make their mind up if the board is to be shown on their home page or not (right now it is, tomorrow it may not).
l3ored - Tuesday, June 1, 2004 - link
price is mentioned as a component in electing the epox as the gold winner, so whats the current street price?SilverRyu - Monday, May 31, 2004 - link
Do you think shuttle will make a 250Gb or K8T800Pro XPC?cnq - Monday, May 31, 2004 - link
Wesley,Nice writeup...
A question about overclocking: you tested at 1:1 ratios, meaning you didn't really find the bounds of the FSB/HTT so much as you found the bounds of the PC4400 RAM that you used. If you had dropped the RAM ratio, do you have any indication which boards could have topped 300 MHz? I am of course thinking of your great AK89 Max review from a few weeks back, where you made waves by showing FSB/HTT speeds of 347 MHz. Will you have time to do similar tests on these new boards? (Or, for that matter, can I ask why you didn't do similar tests on these new boards? Is it merely because ClockGen hasn't yet been ported to NF250?)
I'd also like to second the suggestion made by posting #1 (he mentions aceshardware.com. techreport.com has also chimed in with recent articles on the subject. Not surprisingly, their conclusion is to just avoid cheap Realtek products if you want good CPU utilization. And sudhian.com found great ethernet performance with low CPU overhead from the NF250-GB chip's embedded controller.)