Pretty much every review is stating that the new 512gts cards have a healthy performance lead over the old 640 version. But they always seem to be going against a stock clocked 640 version. If you overclocked a 640 version to the same clocks as a 512 (or underclocked the 512), which would be faster? The 640 would have the edge with memory bandwidth, but would the extra SP's push the 512 ahead?
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
8800gts 512 vs 640 clock for clock?
- Thread starter ritual
- Start date
nissanztt90
2[H]4U
- Joined
- Apr 13, 2007
- Messages
- 3,858
Until you get to 1980x1200 and over, the extra shaders would help. At that point, clock for clock, i think it would be pretty close. Considering the G92 core can clock much higher, its obviously the better card.
As you mentioned, there are plenty of comparisons between the stock 640 and the stock 512, maybe some with factory overclocked 640s - look at those and make some guesses/interpolations. Clock for clock, the 512 is probably going to win, plus it overclocks higher than the 640, and unless you are running high res/high AA the memory bandwidth probably won't be a factor. So if you are starting from scratch get a 512. If you already have a 640, it is probably not worth it to upgrade right now.
I'm getting my new GTS tomorrow. If there's a lot of interest in this, I can run some benchmarks for people, but tell me what you want to see:
Both cards at 512MHz?
or 576MHz (stock clocks for my Superclocked edition)
What games/benchmarks? I have the Crysis demo installed, so I could benchmark that.
Both cards at 512MHz?
or 576MHz (stock clocks for my Superclocked edition)
What games/benchmarks? I have the Crysis demo installed, so I could benchmark that.
http://www.gpureview.com/show_cards.php?card1=548&card2=475#
Go there and clock the cards both to same frequency and you will see how much better then new GTS really is, look at number of shaders how big the difference is..
Go there and clock the cards both to same frequency and you will see how much better then new GTS really is, look at number of shaders how big the difference is..
well too late. I just purchased a new OC GTS for $300 from bfgtech since they had that promotion for the first 500 hardocp users. Now if i can sell my current GTS 640 for around $270 it will only be a $30 upgrade for me.
I wouldn't expect to get $270 for a used GTS 640 though.
Battleneter2
Gawd
- Joined
- Nov 3, 2006
- Messages
- 973
Am I missing something? I don't see any benchmarks. Looking at raw data like this can be very misleading across slightly different architectures.
Try comparing the GTX to the G92GTS and you can see how worthless this type of comparison is, ie 10% extra texture shader performance will not always translate into 10% extra performance in games, ie the GTX is still the fractionally faster card overall depending how you translate all the benchmarks.
.
Why not? I've seen them sell for that much...If i sold it for $250 would $60 be worth the upgrade to the new GTS that i just bought?
Only reason I say that is that I considered getting one of the BFG G92s at $300, but looking on the For Sale forum here it looks like most used 8800GTS 640s are going for $220-250, and it didn't seem like enough of an upgrade to justify the $50-80. I'm thinking the value of the 640s is going to really come down since the 512s have similar or better performance and can be had new for $350 or so. It was a close call for me though - but in the end I figured I was better off waiting for the new cards in the spring. YMMV.
Yeah man, if you can get some crysis comparison benchies between your 640 and the 512, I think we'd all like to see that.
If G92 has an Achilles Heel, it might be when running hi-res with AA and/or AF. G92 has substantially less memory bandwidth than G80.
Well, I finally pulled the trigger on a GTS 512 to replace my GTS 640. I wasn't really expecting too much of a performance increase since I was overclocking the 640 pretty good, but I was wrong. I just installed the 512 tonight, so I haven't had a lot of time to fool around with it, but here are some early numbers.
I was running my 640 at 650/1600/950. I overclocked the 512 to 750/1800/980 without artifacting, and with the fan fixed at 50% (inaudible) I was getting max temps of 60C. I haven't tried mixing and matching core/shader/mem clocks yet to see what is the best mix, although I will try that later (800/2000/2000 failed immediately - I was hopeful but not expecting).
My 3dMark06 score went from 11800 to 13200 - so pretty good jump there, but nobody plays 3dMark so I went for some Crysis benching.
At 1600x1200, no AA, no AF, all settings High on Assualt Harbor (4 runs) I went from 25.4 avg FPS to 31.2 avg FPS, about a 25% increase
At 1600x1200, no AA, no AF, Medium except textures/shaders/volumetrics/game/postprocessing High, I went from 28.0 avg FPS to 34.7 avg FPS, again about a 25% increase.
I'll try to get some other benchies tomorrow, and play around with the overclock some, but for now I can say that I was pretty darn impressed with the boost the 512 gives.
I was running my 640 at 650/1600/950. I overclocked the 512 to 750/1800/980 without artifacting, and with the fan fixed at 50% (inaudible) I was getting max temps of 60C. I haven't tried mixing and matching core/shader/mem clocks yet to see what is the best mix, although I will try that later (800/2000/2000 failed immediately - I was hopeful but not expecting).
My 3dMark06 score went from 11800 to 13200 - so pretty good jump there, but nobody plays 3dMark so I went for some Crysis benching.
At 1600x1200, no AA, no AF, all settings High on Assualt Harbor (4 runs) I went from 25.4 avg FPS to 31.2 avg FPS, about a 25% increase
At 1600x1200, no AA, no AF, Medium except textures/shaders/volumetrics/game/postprocessing High, I went from 28.0 avg FPS to 34.7 avg FPS, again about a 25% increase.
I'll try to get some other benchies tomorrow, and play around with the overclock some, but for now I can say that I was pretty darn impressed with the boost the 512 gives.
In my testing, (Sorry deleted last weekends benchmark runs) I found what you said to be true. Pretty much every game runs a little bit faster with the G92. But, I did find that it choked harder than my GTS 640 when running crysis at 1920x1200, using mostly high settings, and some set to medium. Every resolution below that showed the G92 to have a much easier time, and performance was substantially greater than the G80. It was no surprise that over clocking the memory provided greater performance gains than over clocking the core. I think that Crysis shows us that
1) next-gen high end cards will be 1GB
2) 512-bit & greater memory buses are now useful
IMO, G92 would be a GTX slayer if they had only kept the 384-bit memory bus.
1) next-gen high end cards will be 1GB
2) 512-bit & greater memory buses are now useful
IMO, G92 would be a GTX slayer if they had only kept the 384-bit memory bus.
veedubfreak
2[H]4U
- Joined
- Dec 10, 2007
- Messages
- 2,070
clock for clock is somewhat of a useless comparison as the gts 512 can easily hit clocks of 750+.
Which is why Nvidia capped the memory bus at 256bit so they don't pummel one of their high end cards with a mid-range one.
