- Status
The_Trutherizer
Distinguished
- Jul 21, 2008
- 509
- 0
- 18,980
[citation][nom]spaceman2001[/nom]10 years minimum.[/citation]
I'm betting that it could probably run it in 5D Ultra-Mega-Super HD
I'm betting that it could probably run it in 5D Ultra-Mega-Super HD
TheBigTroll
Illustrious
- Apr 15, 2012
- 14,099
- 1
- 42,965
GreaseMonkey_62
Distinguished
- Jul 3, 2009
- 521
- 0
- 18,980
I'm guessing it's a cost per performance factor. They could get more AMD processors/cores. After all it's not like AMD processors are a slouch.
silverblue
Distinguished
- Jul 22, 2009
- 1,199
- 4
- 19,285
[citation][nom]renz496[/nom] given that previous jaguar were using amd cpu i think it is no surprise that when Cray doing the upgrade for the jaguar they will opt for amd future cpu[/citation]
Same socket, so a drop-in upgrade. Far cheaper than getting a brand new motherboard and more expensive Xeon CPU. Also, would you have the same number of memory slots on a Xeon board?
These CPUs would've been paid for a while back, so I doubt we're going to see a positive impact on AMD's financials.
Same socket, so a drop-in upgrade. Far cheaper than getting a brand new motherboard and more expensive Xeon CPU. Also, would you have the same number of memory slots on a Xeon board?
These CPUs would've been paid for a while back, so I doubt we're going to see a positive impact on AMD's financials.
chipwonder
Distinguished
- Jun 28, 2008
- 5
- 0
- 18,510
devBunny
Distinguished
- Jan 22, 2012
- 181
- 0
- 18,690
[citation][nom]ShadowGryphon[/nom]i have always wondered though... what the hell they use these systems for. I know science and simulations but whenever we hear about the latest and greatest super computer, we never hear about the job they are going to perform.. [/citation]
Why not ask them? ...
http://www.ornl.gov/info/press_releases/get_press_release.cfm?ReleaseNumber=mr20121029-00
http://www.ornl.gov/ornlhome/science_technology.shtml
Why not ask them? ...
http://www.ornl.gov/info/press_releases/get_press_release.cfm?ReleaseNumber=mr20121029-00
http://www.ornl.gov/ornlhome/science_technology.shtml
devBunny
Distinguished
- Jan 22, 2012
- 181
- 0
- 18,690
[citation][nom]bawchicawawa[/nom]It's expected that we get the performance of a supercomputer of today in desktop form by 2019.[/citation]
[citation][nom]spaceman2001[/nom]10 years minimum.[/citation]
Given this history .. http://en.wikipedia.org/wiki/History_of_supercomputing#Historical_TOP500_table .. can anyone calculate and compare the FLOPS rating of an i7-3960 with a pair of top of the range graphics cards running CUDA or OpenCL?
[citation][nom]spaceman2001[/nom]10 years minimum.[/citation]
Given this history .. http://en.wikipedia.org/wiki/History_of_supercomputing#Historical_TOP500_table .. can anyone calculate and compare the FLOPS rating of an i7-3960 with a pair of top of the range graphics cards running CUDA or OpenCL?
[citation][nom]ShadowGryphon[/nom]i have always wondered though... what the hell they use these systems for. I know science and simulations but whenever we hear about the latest and greatest super computer, we never hear about the job they are going to perform.. hmmmm.. almost reminds me of Hitchhiker's Guide to the Galaxy, where the computer had to keep building it self over and over until it could finally compute the Answer to the Ultimate Question of Life, The Universe, and Everything.. oh well just a random Monster fueled thought and question. and finally the question that is asked most often of new computers.. can play Crysis???have a great day.[/citation]
It already the answer to the ultimate question, 43. The rebuild was finding the question.
It already the answer to the ultimate question, 43. The rebuild was finding the question.
[citation][nom]socialfox[/nom]All this performance for the wonderful price tag of $1,000,000,000,000![/citation]
Highly doubt it cost $1 trillion, let alone $1 billion, I would wager it cost 10's of millions at most. Which sounds like a lot but its not relative to our budget, even 1 billion dollars is not a lot compared to our budget. Our budget is about $4 trillion dollars,for the kiddies who paid attention in math, that 1/4000 * 100 = .025% of our annual budget, that's how little impact 1 billion dollars is to our budget, it's a fraction of a percent, so say this super computer cost $100 million which including housing, power, etc its .00025% of our budget a fraction of 1/100th of 1%. Ps this roughly the amount of money the government spends on PBS, annually, which is why mittens comment was so absurd, cutting PBS does nothing to ou budget, cutting NPR does nothing, if you cut NASA entire $15 billion budget you again do nothing to our budget. The fact is the us spends very little on education or science but somehow it's th culprit in our debt and deficit? Other things that cost nothing but get blame, welfare, which basically was abolished in 1995, again less than 1%. Contrast that with, military, mm, ss, and debt interest combined they make up 70% of our budget.
It's sad that things to further our knowledge and keep or regain our technological advantage, the thing that ultimately makes any nation more powerful than the next technology is seen as a bad thing. When as a nation did we become so stupid?
Highly doubt it cost $1 trillion, let alone $1 billion, I would wager it cost 10's of millions at most. Which sounds like a lot but its not relative to our budget, even 1 billion dollars is not a lot compared to our budget. Our budget is about $4 trillion dollars,for the kiddies who paid attention in math, that 1/4000 * 100 = .025% of our annual budget, that's how little impact 1 billion dollars is to our budget, it's a fraction of a percent, so say this super computer cost $100 million which including housing, power, etc its .00025% of our budget a fraction of 1/100th of 1%. Ps this roughly the amount of money the government spends on PBS, annually, which is why mittens comment was so absurd, cutting PBS does nothing to ou budget, cutting NPR does nothing, if you cut NASA entire $15 billion budget you again do nothing to our budget. The fact is the us spends very little on education or science but somehow it's th culprit in our debt and deficit? Other things that cost nothing but get blame, welfare, which basically was abolished in 1995, again less than 1%. Contrast that with, military, mm, ss, and debt interest combined they make up 70% of our budget.
It's sad that things to further our knowledge and keep or regain our technological advantage, the thing that ultimately makes any nation more powerful than the next technology is seen as a bad thing. When as a nation did we become so stupid?
dragonsqrrl
Distinguished
- Nov 19, 2009
- 1,280
- 0
- 19,290
[citation][nom]mousseng[/nom]Hm. It says the Titan has 18,688 nodes that each have a K20 - and (15 SMXes * 192 CUDA cores) * 18,688 comes out to 53,821,440 CUDA cores. However, (13 SMXes * 192 CUDA cores) * 18,688 comes out to 46,645,248 CUDA cores - exactly what they listed.So either the K20 doesn't have a full GK110, or two SMX clusters were disabled on each K20 for some reason.[/citation]
[citation][nom]renz496[/nom]i've read somewhere there will be some SMX disabled in GK110 for tesla K20 and the part will clock lower (700mhz+) so it can sit below 225w TDP for each chip. (some says K20 will have 13 SMX and some other says there will be 14 SMX version but if i remember correctly the GK110 in Cray XK7 have will have 13SMX which means 2 SMX disabled from total of 15 SMX in GK110)and about the Opteron that being used in the supercomputer i find this to be funny lol: http://techreport.com/news/23808/n [...] ting-titanbtw given that previous jaguar were using amd cpu i think it is no surprise that when Cray doing the upgrade for the jaguar they will opt for amd future cpu[/citation]
Anandtech recently published an in depth article about Titan, and they've confirmed that the K20 will have 1 SMX disabled, which doesn't match up with the information given in this article.
http://www.anandtech.com/show/6421/inside-the-titan-supercomputer-299k-amd-x86-cores-and-186k-nvidia-gpu-cores
No offense to the Tom's news team, but I've found Anandtech to be a far more reliable source of information then the news articles published here, so I think I'll stick with Anand on this one. According to Anandtech the K20, which is based on gk110, will have 14 out of 15 SMX blocks enabled totaling 2688 CUDA cores clocked at 732 MHz. This translates to 1.3 TFLOPs of DP performance, which means it should offer around 4 TFLOPs SP.
[citation][nom]renz496[/nom]i've read somewhere there will be some SMX disabled in GK110 for tesla K20 and the part will clock lower (700mhz+) so it can sit below 225w TDP for each chip. (some says K20 will have 13 SMX and some other says there will be 14 SMX version but if i remember correctly the GK110 in Cray XK7 have will have 13SMX which means 2 SMX disabled from total of 15 SMX in GK110)and about the Opteron that being used in the supercomputer i find this to be funny lol: http://techreport.com/news/23808/n [...] ting-titanbtw given that previous jaguar were using amd cpu i think it is no surprise that when Cray doing the upgrade for the jaguar they will opt for amd future cpu[/citation]
Anandtech recently published an in depth article about Titan, and they've confirmed that the K20 will have 1 SMX disabled, which doesn't match up with the information given in this article.
http://www.anandtech.com/show/6421/inside-the-titan-supercomputer-299k-amd-x86-cores-and-186k-nvidia-gpu-cores
No offense to the Tom's news team, but I've found Anandtech to be a far more reliable source of information then the news articles published here, so I think I'll stick with Anand on this one. According to Anandtech the K20, which is based on gk110, will have 14 out of 15 SMX blocks enabled totaling 2688 CUDA cores clocked at 732 MHz. This translates to 1.3 TFLOPs of DP performance, which means it should offer around 4 TFLOPs SP.
dragonsqrrl
Distinguished
- Nov 19, 2009
- 1,280
- 0
- 19,290
[citation][nom]ShadowGryphon[/nom]i have always wondered though... what the hell they use these systems for. I know science and simulations but whenever we hear about the latest and greatest super computer, we never hear about the job they are going to perform.. hmmmm.. almost reminds me of Hitchhiker's Guide to the Galaxy, where the computer had to keep building it self over and over until it could finally compute the Answer to the Ultimate Question of Life, The Universe, and Everything.. oh well just a random Monster fueled thought and question. and finally the question that is asked most often of new computers.. can play Crysis???have a great day.[/citation]
Ohhh boy, here you go...
http://www.anandtech.com/show/6421/inside-the-titan-supercomputer-299k-amd-x86-cores-and-186k-nvidia-gpu-cores/3
An excellent overview of the types of workloads that will run on Titan, from the engineers and physicists themselves.
Ohhh boy, here you go...
http://www.anandtech.com/show/6421/inside-the-titan-supercomputer-299k-amd-x86-cores-and-186k-nvidia-gpu-cores/3
An excellent overview of the types of workloads that will run on Titan, from the engineers and physicists themselves.
renz496
Champion
- Jun 16, 2009
- 14,300
- 228
- 53,090
not only toms some other site also reporting that k20 will only have 13 smx enabled resulting in 1.1 TFLOPS DP like this one:
vr-zone.com/articles/nvidia-s-top-end-kepler-unveiled-tesla-k20-comes-with-disappointing-specs-performance/17458.html
so i think the problem might come from the sources.
vr-zone.com/articles/nvidia-s-top-end-kepler-unveiled-tesla-k20-comes-with-disappointing-specs-performance/17458.html
so i think the problem might come from the sources.
dragonsqrrl,
The source AT used is this article:
http://www.hpcwire.com/hpcwire/2012-10-29/titan_sets_high-water_mark_for_gpu_supercomputing.html?page=1
That article states: "Using the 27 peak PF value for Titan, a back-of-the-envelope calculation puts the new K20 at about 1.2-1.3 double precision teraflops"
1.2-1.3 Tflops can be 2,496 or 2,688. 3 separate sources all report 46,645,248 CUDA cores, implying 2,496.
The source AT used is this article:
http://www.hpcwire.com/hpcwire/2012-10-29/titan_sets_high-water_mark_for_gpu_supercomputing.html?page=1
That article states: "Using the 27 peak PF value for Titan, a back-of-the-envelope calculation puts the new K20 at about 1.2-1.3 double precision teraflops"
1.2-1.3 Tflops can be 2,496 or 2,688. 3 separate sources all report 46,645,248 CUDA cores, implying 2,496.
dragonsqrrl
Distinguished
- Nov 19, 2009
- 1,280
- 0
- 19,290
[citation][nom]renz496[/nom]not only toms some other site also reporting that k20 will only have 13 smx enabled resulting in 1.1 TFLOPS DP like this one: vr-zone.com/articles/nvidia-s-top-end-kepler-unveiled-tesla-k20-comes-with-disappointing-specs-performance/17458.html so i think the problem might come from the sources.[/citation]
Unless Titan uses some sort of custom configured K20, with custom binned gk110's, which I highly doubt is the case, then the 'leak' from vr-zone is probably inaccurate. The specs for the K20's I gave in my previous comment are what's being installed in Titan, there's no question about that. Anand went down to Oak Ridge himself and interviewed the technicians, engineers, and scientist involved with the project.
But the problem probably does come down to sources, I agree with you on that.
Unless Titan uses some sort of custom configured K20, with custom binned gk110's, which I highly doubt is the case, then the 'leak' from vr-zone is probably inaccurate. The specs for the K20's I gave in my previous comment are what's being installed in Titan, there's no question about that. Anand went down to Oak Ridge himself and interviewed the technicians, engineers, and scientist involved with the project.
But the problem probably does come down to sources, I agree with you on that.
dragonsqrrl
Distinguished
- Nov 19, 2009
- 1,280
- 0
- 19,290
[citation][nom]BestJinjo[/nom]dragonsqrrl,The source AT used is this article:http://www.hpcwire.com/hpcwire/201 [...] tml?page=1That article states: "Using the 27 peak PF value for Titan, a back-of-the-envelope calculation puts the new K20 at about 1.2-1.3 double precision teraflops"1.2-1.3 Tflops can be 2,496 or 2,688. 3 separate sources all report 46,645,248 CUDA cores, implying 2,496.[/citation]
Ryan Smith, an editor at Anandtech, confirmed the 14 SMX spec and the 1.3 GFLOP figure. So unless he, Anand, and the other editors at Anandtech are either somehow mistaken, or outright lying (I have no idea why they would) then that information is probably accurate.
edit: and thanks for posting on Anand (just to double confirm my original question), it's an interesting discrepancy and I'd like to know what Ryan/Anand has to say.
And correct me if I'm wrong, but it's also interesting to note that I don't see that "46,645,248 CUDA core" figure given anywhere in the Oak Ridge article Gruener sourced. In fact, out of the 3 sources I'm assuming you're referring to, Tom's is the only one that gives that figure. So really the whole basis for this dispute is coming from this news article here on Tom's. Also you're correct in that the hpcwire article doesn't dispute Anandtech's figures, it's simply a little less specific. And then there is the vr-zone article, who's leaked source I would call into question, and no where is 46,645,248 mentioned. I'm not trying to say Anandtech is incapable of error, it's just that the sources and cross referencing seems to be in their favor.
Ryan Smith, an editor at Anandtech, confirmed the 14 SMX spec and the 1.3 GFLOP figure. So unless he, Anand, and the other editors at Anandtech are either somehow mistaken, or outright lying (I have no idea why they would) then that information is probably accurate.
edit: and thanks for posting on Anand (just to double confirm my original question), it's an interesting discrepancy and I'd like to know what Ryan/Anand has to say.
And correct me if I'm wrong, but it's also interesting to note that I don't see that "46,645,248 CUDA core" figure given anywhere in the Oak Ridge article Gruener sourced. In fact, out of the 3 sources I'm assuming you're referring to, Tom's is the only one that gives that figure. So really the whole basis for this dispute is coming from this news article here on Tom's. Also you're correct in that the hpcwire article doesn't dispute Anandtech's figures, it's simply a little less specific. And then there is the vr-zone article, who's leaked source I would call into question, and no where is 46,645,248 mentioned. I'm not trying to say Anandtech is incapable of error, it's just that the sources and cross referencing seems to be in their favor.
dragonsqrrl
Distinguished
- Nov 19, 2009
- 1,280
- 0
- 19,290
[citation][nom]renz496[/nom]not only toms some other site also reporting that k20 will only have 13 smx enabled resulting in 1.1 TFLOPS DP like this one: vr-zone.com/articles/nvidia-s-top-end-kepler-unveiled-tesla-k20-comes-with-disappointing-specs-performance/17458.html so i think the problem might come from the sources.[/citation]
[citation][nom]BestJinjo[/nom]dragonsqrrl,The source AT used is this article:http://www.hpcwire.com/hpcwire/201 [...] tml?page=1That article states: "Using the 27 peak PF value for Titan, a back-of-the-envelope calculation puts the new K20 at about 1.2-1.3 double precision teraflops"1.2-1.3 Tflops can be 2,496 or 2,688. 3 separate sources all report 46,645,248 CUDA cores, implying 2,496.[/citation]
And we have a confirmation...
http://www.anandtech.com/show/6447/titan-takes-1-spot-on-top500-list-with-1759-petaflops-in-linpack
The system packs 50,233,344 CUDA cores, not 46,645,248, with each Tesla K20X carrying 2688 CUDA cores. I honestly don't know where Gruener got his core count, or how he came publish it as fact, but I'm guessing he probably made a series of assumptions about the specs of each individual card, and simply multiplied that by the total number of nodes published by Oak Ridge (sort of like what you guys did).
And as usual we probably won't get any sort of an acknowledgement or retraction of this mistake. As a matter of fact I'm kind of surprised this article hasn't mysteriously disappeared yet. But this is what I was talking about in my previous comment. The news articles here on Tom's are not a reliable source of information, and I think anyone who's read these articles for any length of time would realize this. It's yellow journalism at its worst.
[citation][nom]BestJinjo[/nom]dragonsqrrl,The source AT used is this article:http://www.hpcwire.com/hpcwire/201 [...] tml?page=1That article states: "Using the 27 peak PF value for Titan, a back-of-the-envelope calculation puts the new K20 at about 1.2-1.3 double precision teraflops"1.2-1.3 Tflops can be 2,496 or 2,688. 3 separate sources all report 46,645,248 CUDA cores, implying 2,496.[/citation]
And we have a confirmation...
http://www.anandtech.com/show/6447/titan-takes-1-spot-on-top500-list-with-1759-petaflops-in-linpack
The system packs 50,233,344 CUDA cores, not 46,645,248, with each Tesla K20X carrying 2688 CUDA cores. I honestly don't know where Gruener got his core count, or how he came publish it as fact, but I'm guessing he probably made a series of assumptions about the specs of each individual card, and simply multiplied that by the total number of nodes published by Oak Ridge (sort of like what you guys did).
And as usual we probably won't get any sort of an acknowledgement or retraction of this mistake. As a matter of fact I'm kind of surprised this article hasn't mysteriously disappeared yet. But this is what I was talking about in my previous comment. The news articles here on Tom's are not a reliable source of information, and I think anyone who's read these articles for any length of time would realize this. It's yellow journalism at its worst.
mrstarware
Honorable
- Apr 19, 2013
- 1
- 0
- 10,510
- Status
TRENDING THREADS
-
News AMD says Intel's 'horrible product' is causing Ryzen 9 9800X3D shortages- Started by Admin
- Replies: 83
-
-
-
-
Discussion PSU recommendations and power supply discussion thread - Tom's hardware- Started by Darkbreeze
- Replies: 2K
-
Discussion What's your favourite video game you've been playing?- Started by amdfangirl
- Replies: 4K
Latest posts
-
-
Question New PC wont post with GPU plugged in - Please help this noob- Latest: pleasehelpimclueless
-
Question What CPU + GPU combo for QHD gaming ~$1500 budget for the system?- Latest: logainofhades
Facebook
Twitter
Instagram