ATi > nVIDIA @ synthetic benches because...?

T

tsuehpsyde

Supreme [H]ardness
2FA
Joined
Oct 22, 2004
Messages
6,604
I'm not even trying to start a flame war, so if you have something ignorant (as in un-informed), please, save everyone the grief and don't reply.

Is there any reason why ATi dominates in benches? I know this is video-card related, but overclocking ~= benchmarking, and that's the basis of my question. It seems almost all benches are owned by the X800XT:pE whereas real gaming seems to be give and take between both card makers' top end pieces. Is it because all benches are Direct3D, which ATi is known for being faster at? Or is ATi just optimizing the hell out of their drivers, whereas nVIDIA just sits on the sidelines and doesn't? Just curious.
 
I am also curious. I think this would have been better posted in the video cards forum in order to get a larger response, though I suspect your request for no flames would not be heard.
 
Well, this is the area for people who benchmarks (otherwise, why OC?) so hense why I placed it here....no answers tho, so I guess it's an unknown answer.
 
Well, doesn't the XT PE have the highest fillrate? I would think that would be theoretically make it the best performer. Reality is different, but synthetics are not always reflections of reality.
 
Jonsey said:
Well, doesn't the XT PE have the highest fillrate? I would think that would be theoretically make it the best performer. Reality is different, but synthetics are not always reflections of reality.
Click to expand...

in a game that uses absolutely no pixel shader/vertex shader optimizations...yes...but in games that utilize these optimizations, not necessarily...
 
Actually, the 6800 series tends to dominate Aquamark 3, and I'd consider it a better/more realistic/more useful DX9 benchmark than most.
 
hmmmm.... maybe the x800xtpe just has the raw clocks/speed? 400/1.1 6800ultra.... and 520/1120 x800xtpe... if i am wrong.. don't flame me!! i am just trying to give an answer!! hehehe
 
tsuehpsyde said:
Well, this is the area for people who benchmarks (otherwise, why OC?)
Click to expand...

I overclock to get better performance from gaming :confused:


With relation to your question - doesn't the XT:pE push more raw pixels and sharders through?
 
Well there is THAT school of thought. I did that with my GeForce3 Ti200 to make it work alright. But mostly it's for benching, or either to save money on an upgrade :)
 
Well, I have found there are a number of factors that allow the XT PE to dominate the benchmarks:

1. There is no cold throttle issue with ATI, so you can freeze these chips all you like. The nVidia, without some odd BIOS tweaks that are a little less effective, cold throttle at 24C and below.

2. The clock rates of the XT-PE are much higher, which allows some nice peak FPS(look at the peak FPS on AM3). It looks like the nVidia has horsepower (look at the min FPS on AM3), but the peak helps improve the average frame rates.

3. Easy to mod. The ATI cards have less flavors than the nVidia cards. Mods on an ASUS do not work on a BFG. A lot of the modded ATI cards run 1.8V into the core and have extreme cooling (-100C) to get around 800 on the core (Holy sh*& !!). Shimano has a workaround to get his phase cooled Ultra's to 600 on the core, but that is still not quite the overclock of the XT-PE. My UEE, modded, can reach the stock XT-PE GPU clock of 520. I know comparing nVidia clocks and ATI clocks is a lot like comparing AMD versus Intel, but, the higher frequencies do help.

4. The latest benchmark, 3DMark05, is a SM 2.0 optimized bench. While it does allow the SM2 code down the SM3 pipes of the nVidia cards, it is not the same as running SM2 optimized code, and ATI wins at that. That's why the bench does not seem to refect real world results (games). SM3 is still considered an IHV optimization.

5. When you peruse the nVidia results on AM3, keep in mind that most nVidia results are inflated 10K because of the well known null cheat. Easy to spot, but they are everywhere. I think this makes the nVidia card look better than it is on this bench.

If you looked at a stock XT-PE versus the 6800 Ultra, then they are very similar. But, I would be hard pressed to find someone around here who runs their XT-PE stock;)
 
Thanks Heavy...probably (as Napoleon Dynamite would say) is 'The Best Flipping Answer Ever, God...' :p Seriously, that helped out alot. And I take it that is a Winchester 3200+ you have in your sig? 800MHz overclock is quite nice, did that require an assload of voltage or no? To get this 3400+ Claw CG from 2.2GHz to 2.4GHz required 1.6v....damn thing is voltage hungry. Plus, aZn said to get his other 100Mhz, it required another .1v to 1.7. As long as my temps are alright, does it matter what kinda voltage I put thru my proc? Using an XP-120 so I've got about the best air cooling you can get. I want to see 2.6GHz (need better memory first, but damn you voltage hungry claw's), just curious as to if I can ever get it.
 
tsuehpsyde

Yes, it is a Winchester 3200, and I run 1.65 V (at BIOS) which is really about 1.55V for 2520, a 9x multi, and 280 FSB 1:1, 1T with the RAM for gaming day to day. That is easily accomplished with air. The watercooling is needed when I up the voltage to 1.80V, to net around 1.75V or so in the core to approach 2.8 GHz. I don't think air would work for that;)

There are also frozen Winnie's getting up to around 3.4 GHz, so there is a lot of potential with the later week releases and certain batches.

And, yes, the 6800's do cold throttle at 24C or lower. The card thinks it is overheating, then drops down to the 2D clock and voltage (1.1V) until the condition is alleviated, which in the case of cold, does not happen. So, the workaround is to create a mod BIOS with 1.5V in 2D with the same maximum clocks you can get for the bench. So, if I can rely on 515/1320 in 3DMark05, then I would set it for that clock in the BIOS. Kind of risky if you have it too high, but there is always the power switch. But, I never run my card below around 28C, where I have clocked it as high as 520 in the core which is not too bad. Shimano gets around 600 on his frozen dual SLI Ultra's with this workaround.
 
omg, freakin SLi Ultra's frozen....*wipes drool off the floor*

Hopefully nVidia kicks it up a notch next round for benchmarks, because I honestly would rather have nVidia in my machine. But, if they figure SLi is how they are going to own benches, they I might have to go ATi next time around. Time will tell...

And I set 1.6v in the bios, and this is what cpu-z reports

3700.png


and that was the LOWER end of the voltage :p I took the pic when I saw it droop to it's lowest...it wavers betewen that voltage (1.622) and 1.645 in cpu-z....and I only have 1.6v set in the BIOS. How can I tell what I'm actually getting to my CPU?
 
So, the workaround is to create a mod BIOS with 1.5V in 2D with the same maximum clocks you can get for the bench. So, if I can rely on 515/1320 in 3DMark05, then I would set it for that clock in the BIOS.
Click to expand...

I'd wanna TEC the card. How would I figure out the maximum clock speeds using benchies if I'm gonna have it below 25c in the first place? I'm thinking that in this case, I'm screwed.
 
Well, I would create a start BIOS with decently high clocks, then have a flash for every 20 MHz gain on the core. Try each one, find the one that locks/fails, then tune the successful BIOS just below the failed. It takes time, but once you find it, you can use it over and over.
 
tsuehpsyde said:
omg, freakin SLi Ultra's frozen....*wipes drool off the floor*

Hopefully nVidia kicks it up a notch next round for benchmarks, because I honestly would rather have nVidia in my machine. But, if they figure SLi is how they are going to own benches, they I might have to go ATi next time around. Time will tell...

And I set 1.6v in the bios, and this is what cpu-z reports

3700.png


and that was the LOWER end of the voltage :p I took the pic when I saw it droop to it's lowest...it wavers betewen that voltage (1.622) and 1.645 in cpu-z....and I only have 1.6v set in the BIOS. How can I tell what I'm actually getting to my CPU?
Click to expand...

I should be benching a couple of Ultra SLIs in the next couple of weeks. Just waiting on the cards to get processed. Should be here late next week. I am only going to watercool them, but I am still looking forward to give them a go. :D

The voltage on the core will vary a bit depending on the accuracy of the BIOS. If the rail voltage is a little high, the BIOS will show low, etc. Other than using a multimeter for amp and voltage values, there is no really accurate way. The onboard sensors are decent, but can be a little off at times.
 
Shit dude, that's gonna be nice....real nice. I can't wait to see it :)

Also, found out CoreCenter had 1.65v set for some reason (damn program does what it wants sometimes...I don't even remember setting 1.65v on it) so now it's at 1.6v at the moment, Priming to make sure it is stable at 1.6v. But my load temps are notiably lower now :D
 
You must log in or register to reply here.
Back
Top