News - NVIDIA Tesla K20 and K20X GPU accelerators officialy released

[HEXUS]

NVIDIA's Tesla line-up is at-last complete.

Read more.

[scaryjim]

the K20X features an amazing 2,688 CUDA cores on a single die (we assume 16 SMX units)

If so that would make 168 cores per SMX, giving K20 14.857... SMX More likely the top end part is 14 SMX (== 192 cores per SMX), which gives K20 13 SMX. AFAIK the full die was meant to be 16 SMX, so these are heavily disabled parts. It does wrap up a lot of the conflicting specs we saw a couple of weeks ago though.

I strongly suspect that the K20X won't actually be seen outside Titan: if they're having to chop out 3 SMX to get a saleable K20, I suspect 18,866 good K20X cards is all they've been able to make

[maverik-sg1]

hehe - We always read a lot more about Nvidia yield issues compared to others, so there's some truth in that, I wonder what the issues are an how they work with TSMC at resolving them?

I say that because in the SOC market place if they get too 'clever' with the GPU design on the replacement for Tegra3, I doubt there's room for yield issues.... and these issues seem to have started before Fermi and continued into Kepla.

No denying these GPGPU's are mighty and given the 'kudos' of being selected for the most powerful supercomputer there's even more to them than that.....2688cores ffs!!!

Because of this there are such high hopes for the GTX780, lets hope people won't be disappointed in one way or another.

I have to say, pricing on graphics cards is massively OTT right now, if these prices are maintained then it will slow down peoples upgrade path. I like shinies and I always enjoyed upgrading, but I don't game at 2560x1440 or have a multi-monitor set-up, so my gtx570 has more than enough life left in it... I can't justify the current prices for a minor speed bump.

[Hicks12]

Lets not get ahead of ourselves, this means very little for the consumer enthusiast market as far as I can tell as the yields would be so poor they will be mighty expensive and only sold in relatively small quantities, AMD is certainly the best bet for GPGPU in the consumer market as they are priced similarly to Nvida small die "gaming only" cards but offer similar performance and then in the gpgpu space they decimate Nvidia.

Nvidia have a long way to go in getting those chips down to acceptable yields to combat AMDs current strength.

[CAT-THE-FIFTH]

Lets not get ahead of ourselves, this means very little for the consumer enthusiast market as far as I can tell as the yields would be so poor they will be mighty expensive and only sold in relatively small quantities, AMD is certainly the best bet for GPGPU in the consumer market as they are priced similarly to Nvida small die "gaming only" cards but offer similar performance and then in the gpgpu space they decimate Nvidia.

Nvidia have a long way to go in getting those chips down to acceptable yields to combat AMDs current strength.

Both the K20 and K20X have the core and the shaders running at between 700MHZ to 750MHZ. The consumer cards have the core and shaders running at over 900MHZ,with boost probably running them at anything between 1100MHZ to 1300MHZ,dependent on the model,conditions and production variations in the GPUs themselves.

The K20X does have 2688 shaders as opposed to the 1536 the GTX680 has,and has more memory bandwidth too. However,the 50% to 70% realworld increase in clockspeeds of the GTX680,negates the shader advantage IMHO. It is probably the improvement in memory bandwidth and the increase in ROPs, which probably would be the main reason for any performance increase over the GTX680.

[Terbinator]

If so that would make 168 cores per SMX, giving K20 14.857... SMX More likely the top end part is 14 SMX (== 192 cores per SMX), which gives K20 13 SMX. AFAIK the full die was meant to be 16 SMX, so these are heavily disabled parts. It does wrap up a lot of the conflicting specs we saw a couple of weeks ago though.

I strongly suspect that the K20X won't actually be seen outside Titan: if they're having to chop out 3 SMX to get a saleable K20, I suspect 18,866 good K20X cards is all they've been able to make

From AT:

Interestingly NVIDIA tells us that their yields are terrific – a statement backed up in their latest financial statement – so the problem NVIDIA is facing appears to be demand and allocation rather than manufacturing.

Of course yield is most likely into relation the 14 cluster downwards, but maybe that was there initial target anyway?

Worth a read:http://www.anandtech.com/show/6446/n...rrives-at-last

[Terbinator]

The K20X does have 2688 shaders as opposed to the 1536 the GTX680 has,and has more memory bandwidth too. However,the 50% to 70% realworld increase in clockspeeds of the GTX680,negates the shader advantage IMHO. It is probably the improvement in memory bandwidth and the increase in ROPs, which probably would be the main reason for any performance increase over the GTX680.

Having 8x the FP64 will help for it's targeted workloads.

[CAT-THE-FIFTH]

Having 8x the FP64 will help for it's targeted workloads.

Sorry,I did not mean for compute purposes but gaming purposes,as I was replying to what Hicks12 said. I changed the post to make it clearer.

I expect the increase in memory bandwith and ROPs to be the main reasons any GK110 based gaming card will be noticeably be faster than a GK104, considering the clockspeed differences, and the fact the K20X is probably using the highest binned GK110 parts.

From AT:


Of course yield is most likely into relation the 14 cluster downwards, but maybe that was there initial target anyway?

Worth a read:http://www.anandtech.com/show/6446/n...rrives-at-last

Of course they will say that,why wouldn't they??

If yields were brilliant,you would be having a full 15 SMX GPU in the K20X or the GK100 would have been released already. Marketing and PR arms of companies exist for a reason.

[ZaO]

are these kind of chips good for gaming or are they better for other tasks? not that I wanna buy one or anything cos I'm sure the price is crazy...

[scaryjim]

I'd make a comment about expecting nvidia to have planned a 16 SMX main chip, but given they've done a 2.5 SMX consumer Kepler I'm not too sure on that...

You don't make a bigger die planning on only being able to yield partially functioning silicon - that'd be ridiculous even for nvidia. I'm sure they planned on releasing a full 2880 (or more? we'd need a die shot to be sure) core card, but I'm equally sure they simulated performance, ran multiple clock speed evaluations, and worked out the appropriate price-points for shipping 14 SMX, 13 SMX, and probably even 12 SMX variants to account for how many dies they needed to get in each bin and what tolerances they had in the manufacturing process.

If you can price highly enough for a 13 SMX die that a 1% yield will cover the manufacturing expenses, then a 10% yield to that bin - whilst actually being terrible - is terrific in a business sense...

The real question for the consumer is how much can you chop out of GK110 and still get a working GPU? Even if they're conservative with binning, an 11 SMX GK110 would have ~ 38% more shaders than GK104. Even at lower clock speeds that should still run faster than a GTX680, at which point its all about TDP and pricing...

are these kind of chips good for gaming or are they better for other tasks? not that I wanna buy one or anything cos I'm sure the price is crazy...

GK110 has a lot more compute power than the lower end chips, particularly in double precision maths. These specific cards, which support ECC memory, and a lot of enterprise/HPC features, are more geared towards heavy compute workloads and tend to have drivers focussed on stability, rather than gaming performance. The underlying GPU, however, could potentially be put into a consumer class card. I suspect though (as mentioned above) that if they do make a consumer card from them we'll see parts with even more shaders fused off.

[CAT-THE-FIFTH]

Even if they're conservative with binning, an 11 SMX GK110 would have ~ 38% more shaders than GK104. Even at lower clock speeds that should still run faster than a GTX680, at which point its all about TDP and pricing...

The main issue is going to be clockspeeds. People underestimate how much the core boost on Nvidia cards does help them and how agressive it is.

[Terbinator]

The main issue is going to be clockspeeds. People underestimate how much the core boost on Nvidia cards does help them and how agressive it is.

Since release the boost appears to have been all but nerfed. I'm lucky to see anything above 1058MHz - and there are no TDP/cooling limitations in play.

[CAT-THE-FIFTH]

Since release the boost appears to have been all but nerfed. I'm lucky to see anything above 1058MHz - and there are no TDP/cooling limitations in play.

The review samples don't appear to show that though?! Even,Tarinder made a point of how the GTX670 boosts very agressively.

[Terbinator]

The review samples don't appear to show that though?! Even,Tarinder made a point of how the GTX670 boost very agressively.

.

When i first got mine (around May, i think) I'd see the card hit north of 11xx in BF3(!). Now i'm lucky to see it go above 1058 when left at stock in nearly every game. The card never hits it's 100% TDP limit even under OC conditions for the most part and it's under water so heat isn't an issue. They just seem to be nerfing it with software with every driver release.

[CAT-THE-FIFTH]

.

When i first got mine (around May, i think) I'd see the card hit north of 11xx in BF3(!). Now i'm lucky to see it go above 1058 when left at stock in nearly every game. The card never hits it's 100% TDP limit even under OC conditions for the most part and it's under water so heat isn't an issue. They just seem to be nerfing it with software with every driver release.

That would piss me off TBH. So basically the reviews are not a totally accurate state of affairs then??

This seems to be going hand in hand with the new limits on overclocking. It makes me think on whether they want the next generation to look better than it is??

[Terbinator]

That would piss me off TBH. So basically the reviews are not a totally accurate state of affairs then??

This seems to be going hand in hand with the new limits on overclocking. It makes me think on whether they want the next generation to look better than it is??

It's hard to say concretely, but every new thread about beta driver is someone saying they've had to lower their clocks by about 20MHz to stay stable. Of course, it's a beta so perhaps there are issues but there must be a reason why, R304.xx for example, are the best drivers for 3DMark11 runs.

i suppose i could go more indepth myself but i cba doing a load of benchmarks on different drivers