[SemiAccurate] Tesla K20 specs: 13 SMX, GeForce probably 12-13 SMX

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
As long as there is no artificial limitation (= missing driver optimizations)...

I'm not an expert, by any means, but I've read that there are no optimizations in the drivers. The OpenCL benchmark bares that out.
 

BenSkywalker

Diamond Member
Oct 9, 1999
9,140
67
91
Consumer cards run OpenCL as well as the workstation equivalent. It's not something that's driver optimized. A 7970 is just as good as a W9000.

7970 doesn't have ECC, it is useless in the K20's market.
 

sontin

Diamond Member
Sep 12, 2011
3,273
149
106
I'm not an expert, by any means, but I've read that there are no optimizations in the drivers. The OpenCL benchmark bares that out.

Sure there are. nVidia is limiting the DP performance, the openGL performance in certain applications and the rasterizing throughput in their consumer products. AMD is not doing it because they have no huge market share in the workstation and HPC market. So it does not hurt them when people buying their consumer products over their FirePro and FireStream cards.
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Sure there are. nVidia is limiting the DP performance, the openGL performance in certain applications and the rasterizing throughput in their consumer products. AMD is not doing it because they have no huge market share in the workstation and HPC market. So it does not hurt them when people buying their consumer products over their FirePro and FireStream cards.

I was just talking OpenCL. Nothing else.

As far as it not hurting AMD to not limit their consumer cards, how do you figure that? Every consumer card that's sold instead of a Firepro is lost revenue. The argument could be made that they would sell more Firepro cards if they crippled their consumer cards. Sorry, don't see the logic in your reasoning (Although I do understand what you are saying. No need to repeat it.). I think it's just a fundamental difference in how the 2 companies conduct themselves.
 

BenSkywalker

Diamond Member
Oct 9, 1999
9,140
67
91
LOL Yes, it's their nonexistent pro GPU market that's killing them.

Profoundly inept management. You have a device you can charge $4000 for, so sell it for $1500 instead? Stupid. No nice way to say it.
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Profoundly inept management. You have a device you can charge $4000 for, so sell it for $1500 instead? Stupid. No nice way to say it.

Who is going to pay $4k for an AMD GPU when they can but one that runs better from nVidia for the same price? AMD has to charge less.
 

BenSkywalker

Diamond Member
Oct 9, 1999
9,140
67
91
Who is going to pay $4k for an AMD GPU when they can but one that runs better from nVidia for the same price? AMD has to charge less.

Then charge $3500. AMD is heading in the direction of going out of business. This is mainly because their margins are pitiful. If you have a device set to make *really good* margins, and you don't, you are a complete and utter idiot. A moron. You shouldn't be trusted to run a maintenance crew, let alone be the executive for a multi billion dollar company.

I would rather not have an Intel/nVidia monopoly in the PC space, if the AMD fans would stop worshiping their idiocy and scream for them to do something smart for a change, maybe they would.
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
1,001
126
Then charge $3500. AMD is heading in the direction of going out of business. This is mainly because their margins are pitiful. If you have a device set to make *really good* margins, and you don't, you are a complete and utter idiot. A moron. You shouldn't be trusted to run a maintenance crew, let alone be the executive for a multi billion dollar company.

I would rather not have an Intel/nVidia monopoly in the PC space, if the AMD fans would stop worshiping their idiocy and scream for them to do something smart for a change, maybe they would.


I think the reason AMD doesn't get the margins is because they are often a follower, not a leader. Nvidia was getting booted out of the chipset business, all that R&D money being flushed down the toilet. So what do they do? They build a GPGPU market.

The computing world is going mobile with ultra lower power parts, not all out performance. Nvidia creates Tegra. AMD says, oh yea, we can make something too... only after ousting Dirk do they head that way.

Usually the smaller company should be more nimble than the giants, but it seems like Intel was able to turn on a dime with the Pentium 4 while AMD kept trucking along with the very mediocore Phenom I and Bulldozer.

To me it seems like AMD doesn't have the margins because they never have the foresight to see an upcoming opportunity. They're always following after a competitor creates the market.

As you said, you have to look towards management for that.
 
Last edited:

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Then charge $3500. AMD is heading in the direction of going out of business. This is mainly because their margins are pitiful. If you have a device set to make *really good* margins, and you don't, you are a complete and utter idiot. A moron. You shouldn't be trusted to run a maintenance crew, let alone be the executive for a multi billion dollar company.

I would rather not have an Intel/nVidia monopoly in the PC space, if the AMD fans would stop worshiping their idiocy and scream for them to do something smart for a change, maybe they would.

Listen, you can leave out the , "AMD fans worshiping their idiocy". You wouldn't pay $3500 for an AMD GPU either. Quit trying to inflame the thread. AMD's financial trouble is their CPU business getting crushed by Intel. It's not their pro GPU business. Trying to go head to head against nVidia's workstation business, which I've already sighted the difficulties with, and CUDA is a losing proposition. They are trying to manipulate it in an open source direction where they have a better chance of competing. You should stick to commenting on something you have some idea about. Throwing around insults isn't going to get this anywhere.
 

lambchops511

Senior member
Apr 12, 2005
659
0
0
Listen, you can leave out the , "AMD fans worshiping their idiocy". You wouldn't pay $3500 for an AMD GPU either. Quit trying to inflame the thread. AMD's financial trouble is their CPU business getting crushed by Intel. It's not their pro GPU business. Trying to go head to head against nVidia's workstation business, which I've already sighted the difficulties with, and CUDA is a losing proposition. They are trying to manipulate it in an open source direction where they have a better chance of competing. You should stick to commenting on something you have some idea about. Throwing around insults isn't going to get this anywhere.

Its not the workstation business that is worth $$, its the HPC market. Titan is buying something just under 20 000 GPUs (you can do the math on revenue), how many workstation / GeForce GPUs do you need to sell to make equivalent revenue (never mind profit).

CUDA is not a losing proposition, the technologies available on CUDA is miles ahead, especially in w. the new Dynamic Parallelism + HyperQ, this almost to many researches / scientists, immediately doubles a GPU value even if it is zero percent faster.

Finally, the customers for these large GPU purchases already invested heavily onto the CUDA codebase (because at the time there was nothing better out there, NVIDIA drove GPGPU computing). The cost of porting our applications (which may not even be possible w. the state of OpenCL) is not worth the cost of $500 per GPU. However, if AMD came up with substantially faster Compute GPUs with decent developer support, then researchers and scientists would be willing to invest their time to do the port.
 

tviceman

Diamond Member
Mar 25, 2008
6,734
514
126
www.facebook.com
18,688 GPU's at $3000 a pop is $56 million in revenue. It's a nice chunk of change, but a drop in the bucket compared to what Intel makes on a daily basis - $150 million dollars of revenue each and every day.

ORNL needs to be merely a starting point for K20, and K20 needs to sell huge for Nvidia to be able to continue to maintain it's dominance over AMD and Intel in 2-4 years.
 

lambchops511

Senior member
Apr 12, 2005
659
0
0
18,688 GPU's at $3000 a pop is $56 million in revenue. It's a nice chunk of change, but a drop in the bucket compared to what Intel makes on a daily basis - $150 million dollars of revenue each and every day.

ORNL needs to be merely a starting point for K20, and K20 needs to sell huge for Nvidia to be able to continue to maintain it's dominance over AMD and Intel in 2-4 years.

Ya its chump change to Intel, maybe thats why Intel (110B) is worth >10x NVIDIA (7.5B) which is worth like 5x AMD (1.5B).
 

BenSkywalker

Diamond Member
Oct 9, 1999
9,140
67
91
You wouldn't pay $3500 for an AMD GPU either. Quit trying to inflame the thread.

I work in distribution. We have over 100 racks used to calculate trends per item spread out across the locations of my employer. These racks came online in 2008 at the cost of $350,000 each. They have half the computing power of AMD's 9000 series GPGPUs. I'm not inflaming the thread, its' just this whole thing with me knowing what I'm talking about and you not having the slightest clue. Would I put in the order for a $3500 part that was faster then the $350,000 part we currently have? Yeah, without hesitation. Would I get a bonus for pulling something like that off? Probably. Would AMD make truckloads of money on me ordering a couple hundred of them? Yep, close to three quarters of a million dollars. So I get increased performance for my job making my life easier, I'm happy, my bosses save ~$50 million dollars so they are dancing a jig, and AMD makes a ton of profit. Downside here?

Trying to go head to head against nVidia's workstation business, which I've already sighted the difficulties with, and CUDA is a losing proposition. They are trying to manipulate it in an open source direction where they have a better chance of competing.

I'm not a brain damaged donkey, so explain how 'open source' is going to help them with anything at all in the GPGPU space. The code I need to run, like almost anyone else that is going to be in a comparable position, is proprietary as hell. We don't allow most of our employees to ever see it any way, not the code base itself, even the program running. You have to have clearance to get into the room where the machine is located(no remote access to it). The financial field and insurance industries are the same way. Open source is so much ignorant fan boy wanking.

You should stick to commenting on something you have some idea about.

When is the next time you are going to be putting a purchase order in for a million dollars worth of computing hardware in this segment? When is the last time you got paid for the work you handled on one? Maybe you should sit back down at the little kids table while the grownups have a discussion :)
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Its not the workstation business that is worth $$, its the HPC market. Titan is buying something just under 20 000 GPUs (you can do the math on revenue), how many workstation / GeForce GPUs do you need to sell to make equivalent revenue (never mind profit).

CUDA is not a losing proposition, the technologies available on CUDA is miles ahead, especially in w. the new Dynamic Parallelism + HyperQ, this almost to many researches / scientists, immediately doubles a GPU value even if it is zero percent faster.

Finally, the customers for these large GPU purchases already invested heavily onto the CUDA codebase (because at the time there was nothing better out there, NVIDIA drove GPGPU computing). The cost of porting our applications (which may not even be possible w. the state of OpenCL) is not worth the cost of $500 per GPU. However, if AMD came up with substantially faster Compute GPUs with decent developer support, then researchers and scientists would be willing to invest their time to do the port.

You misread what I was saying. I said going against nVidia in the workstation and going against CUDA is a losing proposition for the reasons sighted. They're entrenched. Much like ORNL isn't changing from AMD CPU's to Intel, even though they could gain performance, or efficiency, or both.

You obviously know more about this. Feel free to correct me if I'm wrong. I don't believe AMD has the resources to actually go after the HPC market internally. I believe that AMD is more wanting others to do the support for OpenCL and they'll have the hardware that's optimized for it. It might work in the same way Linux has as far as becoming the OS of choice.

I'm not saying this approach is a gimme and is guaranteed to work. I just think, at this point in time, it's AMD's only realistic option.

As far as profit for Tesla GPU's, don't underestimate the cost of the support nVidia supplies. They don't charge more just because they can. It's a very expensive product when you consider the dev costs for both the hardware and the after sale support. Highly profitable? Yes. It's not as simple as selling $500 GPU's for $4k, like some on these boards want to make it seem. The extra 10mos. of intensive development for GK110 cost nVidia a lot of money. Likely many many millions. Money that I doubt AMD has. I believe that if AMD suffered the delays that nVidia has with, first Fermi, and then Kepler, they might have bankrupted themselves. I can only imagine what all of the nVidia proponents here would have been saying about AMD if they executed in a similar fashion. I'll bet more idiocy and moron comments than we see now. ;)
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
I work in distribution. We have over 100 racks used to calculate trends per item spread out across the locations of my employer. These racks came online in 2008 at the cost of $350,000 each. They have half the computing power of AMD's 9000 series GPGPUs. I'm not inflaming the thread, its' just this whole thing with me knowing what I'm talking about and you not having the slightest clue. Would I put in the order for a $3500 part that was faster then the $350,000 part we currently have? Yeah, without hesitation. Would I get a bonus for pulling something like that off? Probably. Would AMD make truckloads of money on me ordering a couple hundred of them? Yep, close to three quarters of a million dollars. So I get increased performance for my job making my life easier, I'm happy, my bosses save ~$50 million dollars so they are dancing a jig, and AMD makes a ton of profit. Downside here?



I'm not a brain damaged donkey, so explain how 'open source' is going to help them with anything at all in the GPGPU space. The code I need to run, like almost anyone else that is going to be in a comparable position, is proprietary as hell. We don't allow most of our employees to ever see it any way, not the code base itself, even the program running. You have to have clearance to get into the room where the machine is located(no remote access to it). The financial field and insurance industries are the same way. Open source is so much ignorant fan boy wanking.



When is the next time you are going to be putting a purchase order in for a million dollars worth of computing hardware in this segment? When is the last time you got paid for the work you handled on one? Maybe you should sit back down at the little kids table while the grownups have a discussion :)

For someone who is so superior, and a professional you have the vocabulary of a teenager on a street corner. You've now earned your way onto my ignore list for being anything but a professional in the way you communicate with people. You're merely arrogant, crude, and rude to deal with.
 

BoFox

Senior member
May 10, 2008
689
0
0
For someone who is so superior, and a professional you have the vocabulary of a teenager on a street corner. You've now earned your way onto my ignore list for being anything but a professional in the way you communicate with people. You're merely arrogant, crude, and rude to deal with.
Looks like it'll be 13 SMX, I think.
 

Jaydip

Diamond Member
Mar 29, 2010
3,691
21
81
I will offer my perspective here.Just because you have good hardware doesn't ensure success, you need to create a ecosystem which AMD severely lacks.When we entered in a contract with NV there were literally no competition from anyone.Organizations feel safe when you have some proprietary stuff as they invest a lot of money and energy in it.Open doesn't necessarily means better.Now if AMD comes with a superior solution now it will still take a long time to forfeit CUDA and go with Opencl. Migration is a pain and it never works smoothly so you need to consider that as well.Moreover NV is very easy to work with as it has been my experience.AMD has almost always been a follower, they need to change that.
 
Last edited:

BenSkywalker

Diamond Member
Oct 9, 1999
9,140
67
91
Migration is a pain and it never works smoothly so you need to consider that as well.

This is actually fairly huge and not to be underestimated. One of the problems moving forward for AMD is going to be the migration period. For companies like mine, we have yet to migrate to GPGPU computing but we will certainly be doing so on the next upgrade cycle. We didn't do it last time because it wasn't viable, this time it will be and the associated costs of porting the code base won't be that bad as we always refine/improve the code when we move to a new platform anyway(always to increase accuracy which higher computing power allows). Long term where it gets tricky for AMD is once companies are already on nVidia, it becomes very easy to stick with them and much more difficult to break into that territory for mass market concerns(speaking of smaller scale machines with larger penetration, not custom setups like Titan where it is much less of a concern).

If they started moving W9000 parts today and dropped every penny of additional revenue into increasing the development platform they could start to build themselves a nice revenue stream. nV's pro division profits more then AMD as a whole. If AMD can start to get decent amounts of black going out of more then just their consumer GPU division perhaps they can mount a successful counter to Intel on the desktop(may be wishful thinking, but it is their most viable option).

ORNL needs to be merely a starting point for K20, and K20 needs to sell huge for Nvidia to be able to continue to maintain it's dominance over AMD and Intel in 2-4 years.

XeonPhi is going to be their biggest competitor. Again using my company as an example, while porting the code base when we swap platforms wouldn't be a major additional expense, it would be an expense and it remains to be seen if that cost is going to be greater then moving to XeonPhi. I'm thinking it is going to come down to how much we are hurt by the non conventional architecture of XeonPhi versus traditional x86 and how well the SIMD units are going to cooperate with our code base. If it works well, then we will almost assuredly end up with Intel over nVidia.
 
Last edited:
Feb 19, 2009
10,457
10
76
Intel is 14x the size of Nvidia but can't compete against Nvidia in the compute arena. Not even with a two node process advantage could they compete. Not even with Knights Corner. So it isn't the behemoth size that makes you superior. It surely makes it easier to "become" superior but the product is key.
Anyway, just thinking out loud.

This is the point. Intel can't even compete vs NV in the GPGPU sector.. AMD has NO chance. They need to shut up shop and focus on other markets (gaming, mobile, smartphones/tablets etc) where they do have a good chance. A poor company bleeding money needs to invest wisely.
 

Keysplayr

Elite Member
Jan 16, 2003
21,209
50
91
This is the point. Intel can't even compete vs NV in the GPGPU sector.. AMD has NO chance. They need to shut up shop and focus on other markets (gaming, mobile, smartphones/tablets etc) where they do have a good chance. A poor company bleeding money needs to invest wisely.

Actually, I feel even AMD would have a better shot at competing against Nvidia in the compute arena than Intel does. If only AMD had the software to back up their massively parallel hardware. That is where their money should have went and they should never have sold snapdragon. Such an awful error.
 

tviceman

Diamond Member
Mar 25, 2008
6,734
514
126
www.facebook.com
Actually, I feel even AMD would have a better shot at competing against Nvidia in the compute arena than Intel does. If only AMD had the software to back up their massively parallel hardware. That is where their money should have went and they should never have sold snapdragon. Such an awful error.

Ironic that snapdragon is the most widely sold SoC today among Android devices.