tacotime (OP)
Legendary
Offline
Activity: 1484
Merit: 1005
|
|
March 16, 2012, 09:32:30 PM |
|
GK104 will have 1536 stream processors clocked at 1GHz.
Here's my totally speculative math:
A GTX 570 with 480 SP at 732MHz gets about 150 MH/s, or 0.3125 MH s^-1 SP^-1.
Scaling linearly with clock speed, we would expect 0.4269 MH s^-1 SP^-1 at 1GHz.
For 1536 SPs, that's 656 MH/s at an estimated TDP of 200w.
Of course, it's likely that these new SPs will be a little slower than the old ones, but even if they're 30% slower they should still be competitive with AMD cards.
|
XMR: 44GBHzv6ZyQdJkjqZje6KLZ3xSyN1hBSFAnLP6EAqJtCRVzMzZmeXTC2AHKDS9aEDTRKmo6a6o9r9j86pYfhCWDkKjbtcns
|
|
|
MrTeal
Legendary
Offline
Activity: 1274
Merit: 1004
|
|
March 16, 2012, 09:37:06 PM |
|
GK104 will have 1536 stream processors clocked at 1GHz.
Here's my totally speculative math:
A GTX 570 with 480 SP at 732MHz gets about 150 MH/s, or 0.3125 MH s^-1 SP^-1.
Scaling linearly with clock speed, we would expect 0.4269 MH s^-1 SP^-1 at 1GHz.
For 1536 SPs, that's 656 MH/s at an estimated TDP of 200w.
Of course, it's likely that these new SPs will be a little slower than the old ones, but even if they're 30% slower they should still be competitive with AMD cards.
It doesn't appear that hotclocks have survived the transition to Kepler though, so while the shaders in a GTX570 actually run at 1464MHz the shaders in the GTX680 will run at 1000MHz. The CUDA cores in Kepler appear to be completely different than Fermi. I don't think we'll have a good idea how they perform until they're released and miners get optimized for them.
|
|
|
|
The-Real-Link
|
|
March 16, 2012, 09:40:38 PM |
|
Still really curious to see how it'll perform though. Thanks for giving us an idea (hopefully).
|
Oh Loaded, who art up in Mt. Gox, hallowed be thy name! Thy dollars rain, thy will be done, on BTCUSD. Give us this day our daily 10% 30%, and forgive the bears, as we have bought their bitcoins. And lead us into quadruple digits
|
|
|
jjiimm_64
Legendary
Offline
Activity: 1876
Merit: 1000
|
|
March 17, 2012, 03:21:36 AM |
|
my guess is it will be as good as a 6950. dont they have about 1536?
|
1jimbitm6hAKTjKX4qurCNQubbnk2YsFw
|
|
|
Jaryu
Member
Offline
Activity: 90
Merit: 10
|
|
March 17, 2012, 04:30:05 AM |
|
my guess is it will be as good as a 6950. dont they have about 1536?
then it won't be good at all, not much to compare when one costs 210 and the other 550 both brand new. So unless nvidia added something in the hardware so it can do the proper bitcoin hash math in 1/5 or better than the time of current cards from the 6000 series, the card it will be totally useless for mining.
|
|
|
|
Gabi
Legendary
Offline
Activity: 1148
Merit: 1008
If you want to walk on water, get out of the boat
|
|
March 17, 2012, 01:41:44 PM |
|
You can't compare ati and nvidia SP, they are different
Also i suppose kepler SP will be different from older nvidia cards. So again, comparing 570 with kepler seems pointless to me.
|
|
|
|
jake262144
|
|
March 17, 2012, 02:35:15 PM Last edit: March 22, 2012, 02:51:20 PM by jake262144 |
|
GK104 will have 1536 stream processors clocked at 1GHz. Here's my totally speculative math:
A GTX 570 with 480 SP at 732MHz gets about 150 MH/s, or 0.3125 MH s^-1 SP^-1. ...
Wrong, shaders in a GTX570 run at twice the core clock. Fermi^WKepler will be running them at core clock, like AMD cards do. This subject was already discussed. Mining performance will almost certainly suck compared to AMD cards.
|
|
|
|
Jaryu
Member
Offline
Activity: 90
Merit: 10
|
|
March 17, 2012, 06:03:25 PM |
|
GK104 will have 1536 stream processors clocked at 1GHz. Here's my totally speculative math:
A GTX 570 with 480 SP at 732MHz gets about 150 MH/s, or 0.3125 MH s^-1 SP^-1. ...
Wrong, shaders in a GTX570 run at twice the core clock. Fermi will be running them at core clock, like AMD cards do. This subject was already discussed. Mining performance will almost certainly suck compared to AMD cards. We'll know the answer soon enough... they are releasing the GTX680 next thursday or friday, so by the weekend the internet will have something either good or not.
|
|
|
|
tacotime (OP)
Legendary
Offline
Activity: 1484
Merit: 1005
|
|
March 17, 2012, 07:20:03 PM Last edit: March 17, 2012, 07:44:35 PM by tacotime |
|
Wrong, shaders in a GTX570 run at twice the core clock. Fermi will be running them at core clock, like AMD cards do. This subject was already discussed. Mining performance will almost certainly suck compared to AMD cards. Even if the SPs are running at 70% the frequency of the Fermi cards, that would still mean they would be capable of 460MH/s. With a TDP of 200w, that's 2.3MH/w/s, which is better than a 6970. The card also has a 1050MHz turbo hot clock. There's also supposed to be an extra ALU per SP in Kepler as well, not sure if it's just another Vec4 ALU though
|
XMR: 44GBHzv6ZyQdJkjqZje6KLZ3xSyN1hBSFAnLP6EAqJtCRVzMzZmeXTC2AHKDS9aEDTRKmo6a6o9r9j86pYfhCWDkKjbtcns
|
|
|
bulanula
|
|
March 17, 2012, 09:38:01 PM |
|
I don't know about you guys but last time I heard, hotclocks are kept in Kepler so the GTX 680 still might turn out to be a BTC mining monster.
Ready to sell all the red ATI stuff I have. Boycotting them if Nvidia turns up with something competitive. Linux drivers suck too much. No need for xserver and no stupid 8 GPU limit = Nvidia FTW. Just hoping that Kepler does integer math right and not only FP like before for weather forecasting, folding etc.
|
|
|
|
Gabi
Legendary
Offline
Activity: 1148
Merit: 1008
If you want to walk on water, get out of the boat
|
|
March 17, 2012, 10:07:07 PM |
|
I'm ready to boycott nvidia even more
|
|
|
|
|
jake262144
|
|
March 18, 2012, 04:31:57 PM |
|
Anyway, maybe someone more technically inclined ( ArtForz we are looking at you ) can tell us if there is some hope or not.
I think you want one particular answer, not just any answer. I see no reason whatsoever why one should expect significant progress in integer ops performance - those have long been off nVidia's radar. There are twice as many stream processors working at a lower clock than previous generation cards. We're not getting rid of buggy AMD drivers and the need for running Xorg any time soon
|
|
|
|
Vbs
|
|
March 19, 2012, 04:10:43 PM Last edit: March 19, 2012, 04:23:41 PM by Vbs |
|
GK104 will have 1536 stream processors clocked at 1GHz.
Here's my totally speculative math:
A GTX 570 with 480 SP at 732MHz gets about 150 MH/s, or 0.3125 MH s^-1 SP^-1.
Scaling linearly with clock speed, we would expect 0.4269 MH s^-1 SP^-1 at 1GHz.
For 1536 SPs, that's 656 MH/s at an estimated TDP of 200w.
Of course, it's likely that these new SPs will be a little slower than the old ones, but even if they're 30% slower they should still be competitive with AMD cards.
The stream processors ("cuda cores") are dynamically clocked, and can go to 1411MHz (non-oc'ed). http://www.overclock.net/t/1231113/gigabyte-gtx-680-2gb-already-arrive-at-my-shopAlso, you should be using hot-clocks in your calcs, e.g.: GTX 570, 480SP at 1464MHz (2*732MHz) shaders ~150MH/s, so (1411*1536)/(1464*480) = 3.084*150MH/s = 463MH/s The thing is... they've changed the architecture from Fermi, so untill someone tests it with real hardware, it's all a gamble. http://www.techpowerup.com/162500/GK104-Block-Diagram-Explained.htmlnVidia presentation slides: https://imgur.com/a/aQmuA
|
|
|
|
bulanula
|
|
March 19, 2012, 04:28:44 PM |
|
GK104 will have 1536 stream processors clocked at 1GHz.
Here's my totally speculative math:
A GTX 570 with 480 SP at 732MHz gets about 150 MH/s, or 0.3125 MH s^-1 SP^-1.
Scaling linearly with clock speed, we would expect 0.4269 MH s^-1 SP^-1 at 1GHz.
For 1536 SPs, that's 656 MH/s at an estimated TDP of 200w.
Of course, it's likely that these new SPs will be a little slower than the old ones, but even if they're 30% slower they should still be competitive with AMD cards.
The stream processors ("cuda cores") are dynamically clocked, and can go to 1411MHz (non-oc'ed). http://www.overclock.net/t/1231113/gigabyte-gtx-680-2gb-already-arrive-at-my-shopAlso, you should be using hot-clocks in your calcs, e.g.: GTX 570, 480SP at 1464MHz (2*732MHz) shaders ~150MH/s, so (1411*1536)/(1464*480) = 3.084*150MH/s = 463MH/s The thing is... they've changed the architecture from Fermi, so untill someone tests it with real hardware, it's all a gamble. http://www.techpowerup.com/162500/GK104-Block-Diagram-Explained.htmlnVidia presentation slides: https://imgur.com/a/aQmuASo going by above link it looks like card is already here. Anyone going to buy one and do some BTC mining ? Maybe a new kernel needs to be developed to pwn the 7970 at mining ?
|
|
|
|
|
st4rdust
|
|
March 19, 2012, 10:03:34 PM |
|
No way of knowing until the card is actually released, but OP did a solid bit of educated estimation.
Also, why are they calling their card Kepler? It sounds to me like some not-very-subtle reference to the behavior of planetary orbits. Is this supposed to be some kind of nVidia/NASA joint project? Some kind of space calculation capabilities? Hmm? Some kind of crazy orbital mechanics logic built into the GPU? Maybe leaving amateur astronomers richer for the experience? That right, nVidia? Hmm???
No, but seriously, sounds like an awful idea for a video card.
|
If you should choose to pity me -14GLjCUE7ohxRLvwZD2sfjKjf22Lt3UHip
|
|
|
ShadesOfMarble
Donator
Hero Member
Offline
Activity: 543
Merit: 500
|
|
March 19, 2012, 10:06:06 PM |
|
Last GPU was Fermi, now we have Kepler, then Maxwell. What do these names have in common? All three were great physicists...
|
|
|
|
st4rdust
|
|
March 19, 2012, 10:51:04 PM |
|
Last GPU was Fermi, now we have Kepler, then Maxwell. What do these names have in common? All three were great physicists...
That makes much more sense. I'm not familiar with the current nVidia line but now I kinda wish I was, so I could have saved myself the embarrassment of my previous rant about the name of the card. Oops.
|
If you should choose to pity me -14GLjCUE7ohxRLvwZD2sfjKjf22Lt3UHip
|
|
|
tacotime (OP)
Legendary
Offline
Activity: 1484
Merit: 1005
|
|
March 20, 2012, 03:22:40 PM |
|
nda is up and reviews are out. The 680 is shredding the 7970 in most applications... Can't wait to see the mining performance. http://overclock.net/t/1231711/toms-geforce-gtx-680-review
|
XMR: 44GBHzv6ZyQdJkjqZje6KLZ3xSyN1hBSFAnLP6EAqJtCRVzMzZmeXTC2AHKDS9aEDTRKmo6a6o9r9j86pYfhCWDkKjbtcns
|
|
|
|