3d mark 2000 benchmark results (DirectX 7)

Previous benchmarks from nd22 with s754 Athlon 64 3400+ and Radeon X1950 Pro
Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

3d mark 2001 benchmark results (DirectX 8 )

Previous benchmarks from nd22 with s754 Athlon 64 3400+ and Radeon X1950 Pro
Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

GeForce GTX 275 shows its teeth in 1600x1200 with almost 2x higher score. It's 2 years newer than Radeon X1950 Pro and much more performant.

3d mark 2003 benchmark results (DirectX 9 )

Previous benchmarks from nd22 with s754 Athlon 64 3400+ and Radeon X1950 Pro
Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

GeForce GTX 275 demolishes Radeon X1950 Pro in 1600x1200 with almost 4x higher score. Keep in mind this is a synthetic benchmark with low cpu load.

I re-ran the 1280x1024 benchmark and the result is indeed correct. The large discrepancy between 1024x768 and 1280x1024 is caused by quality settings.

3d mark 2005 benchmark results (DirectX 9.0c ). Compared to 2003 it is CPU intensive. For 1280x1024 and above we use 8 sample anti aliasing with the highest quality. It represents games that require a X2 CPU.

Turion 64's CPU performance is insufficient and bottlenecks the GeForce GTX 275. This benchmark is very challenging even for a fast AM2 Windsor X2 CPU. s939 X2 will struggle here as well.

See Re: Battle of the platforms: socket 754! for continuation of benchmarks at 2400Mhz once the problem with CR1 got solved.

SuperPI 32M detected instability at 2400Mhz, Command Rate 1T, despite prime95 (1h run) and memtest86 (9h run) working fine. It detected it within 5 minutes. This problem went away with Command Rate 2T, but I do not want to use that. Since CPU voltage adjustment in BIOS doesn't work, voltage cannot be increased and per my rule CPU clock should go down by 200Mhz. Indeed 2400Mhz proved to be edge of stability as I originally anticipated. It may be possible to stabilize it on another board with CPU voltage adjustment but I'm not going to do that.

I'm going to do further testing at 2200Mhz CPU, 259Mhz base, 8.5x multiplier, HT 777 (3x), 400 memory (divisor 11) and post 3d mark benchmark results. This may be of interest to anyone having Clawhammer 3400+.

Results at 2209Mhz CPU, 260Mhz base, 8.5x multiplier, HT 779 (3x), 400 memory (Command Rate 1T).

Memory timings used:
CAS Latency (Tcl): [2]
RAS# active time (Tras): [6T]
RAS# to CAS# delay (Trcd): [3T]
Row Precharge Time (Trp): [2T]
Row active to row delay (Trrd): [2T]
Row Cycle time (Trc): [8T]
Row Refresh cyc time (Trfc): [14T]
Read-to-Write time (Trwt): [2T]
Write Recovery Time (Twr): [2 bus clock]
Write-to-Read Delay (Twtr): [2 bus clock]
Refresh Rate (Tref): [200 MHz , 15.6 us]
Read Preamble value: [5ns]
Async Latency value: [7ns]

Screenshots from cpuz and gpuz

3d mark 2000 benchmark results (DirectX 7)

Previous benchmarks from nd22 with s754 Athlon 64 3400+ and Radeon X1950 Pro
Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

3d mark 2001 benchmark results (DirectX 8 )

Previous benchmarks from nd22 with s754 Athlon 64 3400+ and Radeon X1950 Pro
Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

3d mark 2003 benchmark results (DirectX 9 )

Previous benchmarks from nd22 with s754 Athlon 64 3400+ and Radeon X1950 Pro
Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

GeForce GTX 275 demolishes Radeon X1950 Pro in 1600x1200. Keep in mind this is a synthetic benchmark with low cpu load.

The large discrepancy between 1024x768 and 1280x1024 is caused by quality settings.

3d mark 2005 benchmark results (DirectX 9.0c ). Compared to 2003 it is CPU intensive. For 1280x1024 and above we use 8 sample anti aliasing with the highest quality. It represents games that require a X2 CPU.

Turion 64's CPU performance is insufficient and bottlenecks the GeForce GTX 275. This benchmark is very challenging even for a fast AM2 Windsor X2 CPU. s939 X2 will struggle here as well.

PCMark 2004 benchmark results

Previous benchmarks from nd22 with s754 Athlon 64 3400+ and Radeon X1950 Pro
Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

PCMark 2005 benchmark results

Previous benchmarks from nd22 with s754 Athlon 64 3400+ and Radeon X1950 Pro
Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

We got 4% boost over Clawhammer 2.2Ghz, 1MB L2 cache in CPUscore in both PCMark benchmarks.

Far cry benchmark

Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

We get 113fps compared to 89fps Clawhammer 3400+ reached with Radeon X1950 Pro. CPU frequency is the same, we got SSE3, both systems used Command Rate 1T, memory latency in Aida64 is very similar. 27% performance boost.

Quake 4 benchmark

Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

We get 42fps compared to 28fps Clawhammer 3400+ reached with Radeon X1950 Pro in 1600x1200. 50% performance boost.

FEAR built-in benchmark. All details set to maximum except for soft shadows which nd22 disabled. FSAA 4x, Anisotropic 16x. Enabled HW audio mixing and EAX 2.0 since we have Audigy 2 ZS.

Re: Battle of the platforms: socket 754! (2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (2.4Ghz, 512kB L2 cache)

We do not have screenshots as I do not know how to take them in FEAR.

• 1024x768 - min 46, avg 117, max 304, 100% above 40 fps
• 1280x1024 - min 47, avg 110, max 306, 100% above 40 fps
• 1600x1200 - min 43, avg 100, max 260, 100% above 40 fps

We get 100fps compared to 36fps Clawhammer 3400+ reached with Radeon X1950 Pro in 1600x1200. 2.7X performance boost.

Crysis built-in 32bit benchmark. All details set to HIGH except for anti-aliasing quality which was set to 4x.

• 1600x1200 - min 10.4, avg 27.9, max 42.9

• PCMark 2004/2005 shows there is a massive improvement in HDD Score due to usage of SSD (nd22 didn't use it to be able to compare s754 platforms). Definitely use nForce4 with an SSD.
• 3d mark 2003 shows a massive boost which is also confirmed in FEAR (170% boost) and partially in Quake 4 (50% boost) in 1600x1200 where Radeon X1950 Pro is struggling. With antialiasing, the difference would have been even greater. Usage of overpowered GPU can indeed pay off. nForce4 should be used at least with GeForce 8800 GTX 768MB. Alternatively, 9800 GTX, 250 GTS, GTX 260 will work just as well.

Cinebench 2003 benchmark

Re: Battle of the platforms: socket 754! (Clawhammer 2.2Ghz, 1MB L2 cache)
Re: Battle of the platforms: socket 754! (Newcastle 2.4Ghz, 512kB L2 cache)

Turion 64 at 2209Mhz scores 312 points to 302 of Clawhammer 2.2Ghz in CPU test. A small 3% boost. It is outperformed by Newcastle 2.4Ghz which scores 318 points.

Small lead to Clawhammer 2.2Ghz exists also in C4F Shading test (4% lead) and OpenGL SW-L (4% lead) test so this is not a coincidence.

Since CPU frequency and memory latency is basically identical, this can only be explained by more efficient architecture. This old benchmark doesn't benefit from SSE3.

Congrats on the results! I believe GTX 275 can run FEAR with SS on with an average of over 60 FPS even at 1600*1200!
For an overclocked CPU the result are not too far from the results I got at default clocks with Athlon 64 3400!
That Gigabyte K8NE has an exceptional layout: SATA ports are above the pci-express port!

There are two versions of Gigabyte K8NE - v1 an v2. I have v2. One SATA port gets blocked by 2 slot PCIe card. V1 has more voltage options in BIOS according to manual. I saw one sold locally.

4% CPU boost at 2.2Ghz (seen in both Cinebench and PCMark) helps to close the gap with s939 as NewCastle, Winchester, Venice, Manchester got just 512kB L2 cache and need 2.4Ghz to compete. s939 San Diego also has 1MB L2 cache so it should be better as well as FX or Opterons.

There is an Italian seller of Turion 64 ML-34 on eBay who sells them for 10€ so it may be worth to anyone interested in s754. They turned out to be surprisingly compatible.

I managed to find a Turion 64 ML-37 locally in laptop spare parts shop. We will see whether it works at 2.4Ghz without voltage adjustment.