This may also be the case in addition to what I've said. Nevertheless, the point remains that 15 SMX and higher frequencies are rather likely for Geforce than for Tesla/Quadro.
This may also be the case in addition to what I've said. Nevertheless, the point remains that 15 SMX and higher frequencies are rather likely for Geforce than for Tesla/Quadro.
Not to back Charlie on anything, but didn't he say that GK110 would not be released as a gaming card? Don't remember him saying it would never be made.
I'm too lazy to go read.
Its done? Where did you find the press release, i looked but found nothing.
On that article, all they said was receiving initial 32 samples, then 1,000 later... then they expect to finish by December 2012 to March 2013..
"Our DOE target date for the Keplers being available to our users is March 2013," Nichols said. "We will change the name (from Jaguar to Titan) once we have gotten through acceptance (most likely sometime between December 2012 and March, 2013)."
SirPauly
Diamond Member
- Apr 28, 2009
- 5,187
- 1
- 0
NVIDIA Powers Titan, World's Fastest Supercomputer For Open Scientific Research
http://nvidianews.nvidia.com/Releas...omputer-For-Open-Scientific-Research-8a0.aspx
ORNL Debuts Titan Supercomputer
http://www.ornl.gov/info/press_releases/get_press_release.cfm?ReleaseNumber=mr20121029-00
http://nvidianews.nvidia.com/Releas...omputer-For-Open-Scientific-Research-8a0.aspx
ORNL Debuts Titan Supercomputer
http://www.ornl.gov/info/press_releases/get_press_release.cfm?ReleaseNumber=mr20121029-00
blastingcap
Diamond Member
- Sep 16, 2010
- 6,654
- 5
- 76
I don't know what he initially said but he had an article recently saying that there will be cut-down versions of the GK110 sold as GeForces while the fully functional ones go to Tesla/Quadro, which makes absolute sense if you think about it. Why not sell your full chips at huge prices and the harvested ones as lower-margin GeForces? Everybody wins in that arrangement. Charlie claims those will have too-high power draw and be akin to GTX 465, but I don't know, GK110-consumer probably be better than that simply becaues GTX 465 was THAT BAD.
http://semiaccurate.com/2012/10/15/will-nvidia-make-a-consumer-gk110-card/
Charlie may be delusional, but I would not imagine he would say GK110 would never be manufactured. At most I can guess he said that it would not be manufactured for consumers on 28nm. Most likely due to power draw/die sizes.
blastingcap
Diamond Member
- Sep 16, 2010
- 6,654
- 5
- 76
Huh? He said that GeForces based on GK110 will be produced but will be made from harvested GPUs, with the non-harvested GPUs going to professional cards. This is unsurprising to me and should be unsurprising to anyone who was paying attention to the Kepler-based Tesla launch earlier this year. Let's face it: PC gamers (GeForce users) are a stagnant-at-best market, and it isn't even high-margin like the stagnant pro graphics (Quadro) market. It looks like the high-margin, fast-growing HPC market (Tesla) will get priority. Which is probably fine considering how consolized most PC games are and will probably continue to be. If NVDA can make more money with Tesla as a priority, more power to them. Hopefully those extra sales help Quadro and GeForce users as well, via shared R&D costs which are relatively fixed, and higher volume (including harvested dies).
Oh really? And just who do you think would be creating processors 3-5 years of capability ahead of what Intel, the ATI branch of AMD, and Nvidia? This is pure fantasy.
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 146
- 106
Some live in an illusional that specially military/enforcing sides of the governments got secret technology and such way ahead of commercial. Tends to be the other way around.
Thanks to Gipsel @beyond3d forums:
15 SMX @800-850 Mhz @250W don't sound unrealistic now for Geforce.
If those specs are accurate, then holy crap I'm excited! I still think the first Geforce based GK110 won't be fully unlocked, but this definitely bodes well with where Nvidia is on GK110's viability. BTW, if GK110 is 550mm^2 (one of the size rumors), then 100 total dies can be made on one 300mm wafer. Nvidia needed 18,688 functional dies.... Lots and lots of GK110 wafers have been produced by TSMC!
As of the second week of September they had 32 K20s, now they have over 18K. That indicates a production rate of about 3,000 units per week. If we were to assume that nVidia was going to release a BigK GeForce board at $499(I find that unlikely, $599 seems to be a reasonable minimum expectation) that would work out to ~$78Million at retail just for their top SKU over the course of the year.
Said another way, GK110 production seems to already be at the volume that would be needed for a desktop launch, by a comfortable margin actually. If they were going to launch in January at that production rate they would have tens of thousands of boards ready at launch, which would be quite a bit higher then we have seen for any top tier SKU for a while now(while we all like to spend a lot of time talking about the halo cards, they don't actually move much volume).
Said another way, GK110 production seems to already be at the volume that would be needed for a desktop launch, by a comfortable margin actually. If they were going to launch in January at that production rate they would have tens of thousands of boards ready at launch, which would be quite a bit higher then we have seen for any top tier SKU for a while now(while we all like to spend a lot of time talking about the halo cards, they don't actually move much volume).
blastingcap
Diamond Member
- Sep 16, 2010
- 6,654
- 5
- 76
No idea where you are getting those numbers but I think it is likely that for such a big die like GK110, we'll see someone getting the full dies (Tesla) and someone getting the harvested dies (GeForce?), in which case you can't just look at K20 numbers because harvested GPUs may be produced (or identified rather?) at a lower rate.
It says it uses 2090's. Those are dual GK104 cards. Sorry, have I missed something?
They don't want to cure disease. They want to come up with treatments that allow you to live with it. Medication for the rest of your life with more pills that counter the side effects is far more profitable.
SirPauly
Diamond Member
- Apr 28, 2009
- 5,187
- 1
- 0
2090s are single GPU parts-
http://en.wikipedia.org/wiki/Nvidia_Tesla
K10 are dual GK104 parts.
Crazy thing called math
The revenue numbers are the ones to think about. The numbers I put down are increased by 900% for every BigK chip that is sold as a Tesla instead of a GeForce. I don't think you'll find too many people that are going to believe nVidia is going to pull down billions in revenue from one SKU any time soon
NIGELG
Senior member
- Nov 4, 2009
- 852
- 31
- 91
Wow cynic much?
.I understand what you're saying .I just hope that a cure can be found for these diseases.
There are also many scientific problems a supercomputer might help solve in many fields...Spaceflight,Engineering,Genetics etc etc....
2090s are single GPU parts-
http://en.wikipedia.org/wiki/Nvidia_Tesla
K10 are dual GK104 parts.
Crazy thing called math
The revenue numbers are the ones to think about. The numbers I put down are increased by 900% for every BigK chip that is sold as a Tesla instead of a GeForce. I don't think you'll find too many people that are going to believe nVidia is going to pull down billions in revenue from one SKU any time soon
So they are. I misread the source I went to original. They are GF110 Fermi cards. Thanks for pointing that out for me.
Cookie Monster
Diamond Member
- May 7, 2005
- 5,161
- 32
- 86
Id think that due to the power consumption which is quite important for what Oak Ridge labs are doing, the tesla versions might only use 13/14 SMX versions of the GK110 (the 13SMX ones were on sale from a shop while looks like the K20s used in this project is the 14SMXs or a mixture I suppose). So this means that the desktop counterpart could use the fully enabled ones thanks to there being no hardcap on the power spec (possibly up to 300W).
It could be as you say, or it might be like GF100 with no fully functioning chips. Time will tell.
Yeah, this is definitely a 'wait and see' issue, until we get an official announcement from Nvidia.
brightside of news is reporting that Titan has 46,645,248 cuda cores, which when divided by 18,688 GPU's equals 2,496 cuda cores per GPU - 13 SMX's. So it appears there is conflicting reports as to exactly how many SMX's are active, what operational speed the GPU's are functioning at, OR what GK110's peak theoretical tflops is.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter