Please help!!! Benchmarks... (WinXP)!

[three0duster]

Its the RAM. Its definetly the RAM. 2 1Gb sticks of DDR2 800 SODIMM.

If that is truly the case I'd like to see this thing, on an AM2 system that has the DDR2 mem, and a decent over clock maybe 2.4-2.8 ghz and some nice memory, Just for my own knowledge.

 

[Mephistopheles]

DaSickNinja: Thanks a lot for your results!!!...

New score for X6800 @ 3.9GHz - 12.5 sec.
New score for Xeon 3060 @ stock - 18.32 sec 8O
Score for E6400 @ stock - 19.75 sec 8O
Score for Turion X2 @ 2.0GHz - 23.5 sec 8O
Score for Core 2 T2700 @ 2.2 GHz - 21.2 sec 8O
Score for E6300 @ stock - 22.65 sec 8O
Score for Celeron D (Prescott) @ 3.06GHz - 35 sec
Score for Pentium 4 @3.8 GHz - 33.2 sec
Score for AMD FX 55 - 29.99 sec
Score for Pentium 2 @266 Mhz - 59.6 sec 8O
Score for Dual Woodcrest 5160 @ 3.0GHz - 17.8 sec

Oh man, there has to be something wrong with that code. There are too many bizarre results there! For instance, your E6400 @ stock 2.13Ghz does as well as two other E6300s OCed to 3Ghz!! That Turion X2 indeed does almost to well to be true! The 3060 Xeon @ 2.4Ghz performs close to the 3Ghz OCs too... and the most bizarre of it all: that Pentium II 266Mhz, if true, outperforms an Athlon XP 2600, a 1.7Ghz P4 Willamette, another Athlon XP 1700, and a dual-CPU P3 933Mhz!!! What the hell is up with that?

Grimmy: Bizarrely enough, yours is an anomalous result for this benchmark. A Celeron @ 2.4Ghz outperforms you... But don't worry, it's the code's fault. I'll look into it and get back to you.

Skyguy: Indeed, even if you dismiss all the bizarre performance fluctuations, C2Ds still annihilates A64s by a +50% per clock performance gain!!! So I have been able to show C2Ds performance advantage. 8)

Also, DaSickNinja is right about the RAM. This benchmark seems to be highly dependent on FSB and memory speeds, be it latency or raw clock rates for each. That's probably why the 3060 Xeon can destroy the others easily.

But still, there are some bizarre results there that don't follow this rule. Take, for instance, that last E6400 score DaSickNinja reported: 19.75s. How come? That's the same as the X6800! Also, his stock E6300 is doing much better than others! :!:

I'm going to talk to the guy who wrote this about this and see if he can shed some light on the subject.

 

[Grimmy]

Grimmy: Bizarrely enough, yours is an anomalous result for this benchmark. A Celeron @ 2.4Ghz outperforms you... But don't worry, it's the code's fault. I'll look into it and get back to you.

Well.. I think it does have to do with Dual Channel memory, which mine lacks.

Super_PI I basically score low as well, perhaps when you compare another system with the exact same CPU.

From what I remember before the upgrade:

Super PI scores for 1MB:

P4 2.66/533/stock - 1 min 06 sec
P4 2.66/533/OC 2.9 - 57 sec
P4 3.00/800/stock - 50-52 sec

Forgot this benchmark when I had it OC to 3.2, since my bus was jumping around from its setting of 215. It actually fluctuated to 218, according to CPU-Z, or it was a misreading.

:lol: . o O (perhaps my 4-5 year old 430W PSU is on its way out in pushing it)

I've seen other P2.8/800 OC to 3.0 ghz getting 45-47 secs, if memory serves, in which they do have dual channel memory support.

 

[dasickninja]

I think some of those might be anomalies. I'll run the test later today to see if I can get more accurate scores.

 

[Schepsel]

C2D E6600 @ Stock speeds. 19.0s

P5W DH DL and OCZ PC7200 4-4-4-12 @ 800Mhz

 

[noxida]

C2D E6600 Stock 17.89s
ASUS P5B Deluxe
Corsair 6400C4 2Gb

 

[GrimReaperGuy]

P5W DH Deluxe
2GB 800Mhz RAM (5-5-5-15)

E6600 (2.4) = 19s
E6600 (3.35) = 12s

 

[TSIMonster]

Still very interesting score for the turion.

 

[Mephistopheles]

Thanks, GrimReaperGuy, noxida and Schepsel! Also, thank you very much for your interest, DaSickNinja! 😀

I have 6 different E6600 stock scores that are downright phenomenal. This is great news for my colleagues!

Also, I've added a performance per megahertz comparison. It immediately shows which results look skewy and which ones look legit. C2Ds always get pretty high performance per megahertz (3.6-4.4x reference), while A64s always seem to get away with around 2.6x reference.

Take a look at the new results, including multiplicity:

 

[GiDDY_SOUL]

That isn’t the FSB! It’s the DRAM clock.

A) Celeron @ 1700 MHz, RAM - (2x) 134 MHz, FSB = (4x) 101 MHz -> 87.8sec
B) Celeron OCed to 1920 MHz, RAM – (2x) 113 MHz, FSB = (4x) 113 MHz -> 102 sec
C) Celeron OCed to 1853 MHz, RAM - (2x) 142 MHz, FSB = (4x) 109 MHz -> 48 sec* (The System was unstable bcuz of the DRAM clock. The chipset supports DDR-266 only)
D) Celeron OCed to 1818 MHz, RAM – (2x) 140 MHz, FSB = (4x)107 MHz -> 82 sec


RAM timings –> Cas latency – 2CL, RAS to CAS Delay - 2CL, RAS Precharge- 2CL, Cycle time – 5CL

E) Celeron OCed to 1818 MHz, RAM – (2x) 140 MHz, FSB = (4x) 107 MHz -> 78.5sec (Benchmarked twice)
F) Celeron OCed to 1818 MHz, RAM – (2x) 140 MHz, FSB = (4x) 107 MHz -> 73sec* (Benchmarked twice with Opera and Kerio firewall running. All other benchmarks were done with these 2 off!)
G) Celeron OCed to 1853 MHz, RAM - (2x) 142 MHz, FSB = (4x) 109 MHz -> 76 sec

The benchmark is sensitive to memory timings. It gives strange* results too! (See F &C)
May b u should remove the results with the “*” sign from the table

 

[dasickninja]

You are quite welcome. If you need anything, you'll know how to find me.... *steps back into shadows*

 

[Mephistopheles]

Aaaah yes, you're right, I missed the part where you said that these numbers were for you RAM. Oops, sorry!!

The table has been corrected accordingly. Thanks!! 😀

 

[Mephistopheles]

Damn... I've come to the conclusion that this benchmark is so memory-dependent that I should have asked people to tell me what kind of memory they were using, not what kind of CPU. There are too many wild fluctuations there that I simply cannot account for.

Question though: that Turion result leads me to believe that Socket AM2 A64s will do much much better. Can anyone else post results for DDR2-equipped A64s?...

I suspect this code is so ridiculously dependent on memory that that's why the results float around so much... This is even more dangerous, because the unsuspecting and way too trusting physicists around here only ask system integrators to go with "processor X" and accept whatever kind of memory and motherboards the system integrators want to shove down their throats...

This is one of those times that justify things like this memory...

 

[three0duster]

Im gonna work with my timings and memory clock speeds and see what I can get, To get the overclock I was running I had hte memory only at 215mhz, so Ill see what 250 mhz can do that I know whats oging on. I still want to see some AM2 boards run this thing.

 

[mrcmtl]

Im gonna work with my timings and memory clock speeds and see what I can get, To get the overclock I was running I had hte memory only at 215mhz, so Ill see what 250 mhz can do that I know whats oging on. I still want to see some AM2 boards run this thing.

AM2 2.8G..DDR2-933 CL5...24s

 

[darkstar782]

Celeron 340 (2.93GHz) s478, Prescott Stepping D0, 22x multi, 133FSB.
512mb single channel DDR1 @ 333MHz 2.5-3-3-7, 4:5 divider.
Intel 845 chipset/memory controller, 8mb shared Gfx ram.

115.3125 secs.

This is my parents computer, will try on all of mine when I get home tomorrow

 

[darkstar782]

My main PC, probably not a good result as it was done while connected over Remote Desktop.

Pentium D805 @3.66GHz, 183*20 (normally clocked at 4GHz but didnt want it to crash wen I was away).

3*1GiB DDR2 single channel (well, Dual Channel Asynchronous but basically single channel) @ 732MHz 4-4-3-8.

Asus P5W DH Deluxe (Intel 975 chipset/memory controller)

22.71875 seconds.


One thing... the file is Bench2.exe yet your first post says it should be bench3.exe?

 

[Mephistopheles]

AM2 2.8G..DDR2-933 CL5...24s

Holy crap! It's the memory alright. Though a similarly-equipped C2D will probably do better.

Also, darkstar782: your results for that OCed PD 805 are fantastic. Another evidence of very powerful memory influence. And your parent's computer doesn't stand a chance with that subpar DDR1-333...

Thanks guys. With all this information, I'm positive that my friends will be quite pleased! I might end up recommending Corsair Dominator with maybe an E6400 OCed to a sync-operating 369Mhz FSB (which is 8x369 = 2.96Ghz, should be easy with a high-end motherboard). Sounds like this would easily make it under 20s - or maybe even under 15s...

 

[Avalanche]

26.8s everything in sig except running at 3.0 Also on Vista

 

[NSH]

Ran 3 times to get an average.

21.6s each run on a stock E6400

 

[darkstar782]

Pentium D 805@ stock, 133*20 = 2.66GHz. RAM @ DDR2-533 Dual channel asymettric (basically single channel performance due to 2GB on one channel and 1GB on the other) 3-3-3-8 timings. Intel 975X/Asus P5W DH Memory controller

Total Time 31.15625s



Pentium D 805 @ 4GHz, 200*20. RAM @ DDR2-800 Dual channel asymettric, 4-4-4-12 timings. Intel 975X/Asus P5W DH Memory controller

Total time 21.375s // Process 21.35938s



Pentium D 805 @ 4GHz, 200*20. RAM @ DDR2-400 Dual channel asymettric, 4-4-4-12 timings. Intel 975X/Asus P5W DH Memory controller

Total time 36.15625

Thats about a 15 second, or 75% increase in time with DDR2-400 rather than DDR2-800, leading me to believe this benchmark is *very* much memory limited, especially as it didnt even hit 100% CPU usage on either core with DDR2-800 and 4GHz core. Faster than DDR2-800 isnt likely to bring me much benefit however as the D805 is limited in FSB bandwidth.

Will bring you DDR2-600 benchmark in a second just to be thorough!.... Here they are!

Pentium D 805 @ 4GHz, 200*20. RAM @ DDR2-600 Dual channel asymettric, 4-4-4-12 timings. Intel 975X/Asus P5W DH Memory controller

Total time 25.23438s

Thats quicker than I expected, might need to re-run the DDR2-400 result later but I'm bored now 😛

I also did my best to beat 20s, but its not happening 🙁 took it to 4.2GHz DDR2-840 4-4-4-12 and got 20.5 something... oh well.

 

[Guest]

Well just to back up the memory issue even if it seem really clear now.

I killed almost all of the process, got from 75 to ~20 and had almost the exact same run, just under a minute. Probably have crappy memory even if I didnt bother to look!