Nvidia Kepler, a step back for GPU folding?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
-
- Posts: 5275
- Joined: Wed Jun 13, 2007 10:12 am
- Location: ITALY
Nvidia Kepler, a step back for GPU folding?
This is the question, someone told.
I have no answer but here there's a somehow related rather interesting reading (just FYI).
I have no answer but here there's a somehow related rather interesting reading (just FYI).
Re: Nvidia Kepler, a step back for GPU folding?
I don't have much to add other than it does appear that at least GK104 (GTX 680 silicon) is a step back for computation. Its FP64 processing rate is 1/24 FP32, as opposed to Fermi's rate of 1/8, and 1/4 for GCN (Radeon 7970). It's possible that a future GK110 chip changes this, though.
-
- Posts: 5275
- Joined: Wed Jun 13, 2007 10:12 am
- Location: ITALY
Re: Nvidia Kepler, a step back for GPU folding?
However I don't agree even with the "graphics first" commandment/explanation/approach about which the article talk.
As a paradox, even leaving the computational trade-off aside, after two years, two generations and one die-shrink, a pair of GTX460 looks like still (relatively) competitive with what Nvidia (and the GPU market as a whole) has to offer in its upper side nowadays.
As a paradox, even leaving the computational trade-off aside, after two years, two generations and one die-shrink, a pair of GTX460 looks like still (relatively) competitive with what Nvidia (and the GPU market as a whole) has to offer in its upper side nowadays.
Re: Nvidia Kepler, a step back for GPU folding?
Not starting a flamewar, but if you have to upgrade now AND want max folding power, your current choice isn't Nvidia any more. As Luca pointed out, this may change with the next gpu generation. But for the time being, the newest Keppler is way behind.
-
- Posts: 5275
- Joined: Wed Jun 13, 2007 10:12 am
- Location: ITALY
Re: Nvidia Kepler, a step back for GPU folding?
I think there's no need for any flame, it's just a reflection on the state of things.
After some hype and great expectations, maybe Kepler achievements are questionable.
If we look at the reference GTX670 (sort of downclocked GTX680), graphics-wise it's about on par with the previous Fermi GTX580 within a smaller thermal envelope (and the oced GTX670s are in the GTX680 territory): but this is also roughly what a pair of old GTX460 actually can still offer (and do you remember the EVGA 2 Win?), which is even about competitive with Tahiti/Tahiti Pro.
Even if it's just an hint, it leaves (to me) more than a doubt about the effectiveness of the Nvidia development along last two years: I'm not so confident any (if any) GK110 can solve anything either for computational power (and the new Tesla 3000 8GB is just a GTX690 IIRC), or maybe even graphics-wise.
Strictly about folding, AFAIK probably Intel (cpu) folding under linux looks like the most efficient way, expecially for a dedicated rig.
After some hype and great expectations, maybe Kepler achievements are questionable.
If we look at the reference GTX670 (sort of downclocked GTX680), graphics-wise it's about on par with the previous Fermi GTX580 within a smaller thermal envelope (and the oced GTX670s are in the GTX680 territory): but this is also roughly what a pair of old GTX460 actually can still offer (and do you remember the EVGA 2 Win?), which is even about competitive with Tahiti/Tahiti Pro.
Even if it's just an hint, it leaves (to me) more than a doubt about the effectiveness of the Nvidia development along last two years: I'm not so confident any (if any) GK110 can solve anything either for computational power (and the new Tesla 3000 8GB is just a GTX690 IIRC), or maybe even graphics-wise.
Strictly about folding, AFAIK probably Intel (cpu) folding under linux looks like the most efficient way, expecially for a dedicated rig.
-
- Posts: 5275
- Joined: Wed Jun 13, 2007 10:12 am
- Location: ITALY
Re: Nvidia Kepler, a step back for GPU folding?
quest_for_silence wrote:Even if it's just an hint
Well, I would add just that I'm starting to slowly change my mind on Kepler, while receiving more and more hints, or even some sort-of "deja-vu" (OMG, so my memory isn't so good as I thought)... We'll see.
-
- Posts: 5275
- Joined: Wed Jun 13, 2007 10:12 am
- Location: ITALY
Re: Nvidia Kepler, a step back for GPU folding?
Just for reference purposes: The Inquirer, less than two months ago.
Re: Nvidia Kepler, a step back for GPU folding?
http://www.anandtech.com/show/5840/gtc- ... -tesla-k20
Let's wait and see, how they perform. And of course, for what price.
Let's wait and see, how they perform. And of course, for what price.
Re: Nvidia Kepler, a step back for GPU folding?
I don't understand why it should be surprise that NVIDIA want to differentiate their HPC/GPGPU cards from their mainstream graphics cards. They have done this for a long time, as have their primary competitor, in certain ways at least. If AMD weren't flapping about trying to find The Rock Of Relevance in stormy waters they'd probably do the same thing.
-
- Posts: 5275
- Joined: Wed Jun 13, 2007 10:12 am
- Location: ITALY
Re: Nvidia Kepler, a step back for GPU folding?
I mean it's not the point.nutball wrote:I don't understand why it should be surprise that NVIDIA want to differentiate their HPC/GPGPU cards from their mainstream graphics cards.
Re: Nvidia Kepler, a step back for GPU folding?
Not the point of what? Folding is never going to be the top of the priority list for any of the IHVs.quest_for_silence wrote:I mean it's not the point.
-
- Posts: 5275
- Joined: Wed Jun 13, 2007 10:12 am
- Location: ITALY
Re: Nvidia Kepler, a step back for GPU folding?
nutball wrote:Not the point of what? Folding is never going to be the top of the priority list for any of the IHVs.
Actually we were not debating on any market segmentation, but just about current Kepler's technical achievements.
If these are due to a new market segmentation strategy, I think we'll see later this year (or in the 2013) with the Tesla K20 and with any "mainstream" GK110-based cards (if any).
At anyway, folding may never has been the top of the priority list for any of the IHVs (even if the point is the lack of computational power), but I mean it isn't enough to not point out that Nvidia has actually "stolen" PPDs right from our pocket.