[keritto]

Originally Posted by CarstenS

Exactly. Nvidia shouldn't have brought Fermi over to the 600 series any more than AMD should have VLIW-parts in their HD7000 line-up.

Exact. They both make this strategy since on desktop market there arent enough demand for 2-3 gen old parts that they could easily get rid off in more consumerism oriented notebooks.

But two things are definitely weird. Bringing pretty consumable GF114 part into laptops, probable some DTR (desktop replacement) if there's still such market. As most of oldish rebranded AMD products are still budget mainstream line and can be implemented to support crappish Sandy Bridge based CPU

And more weird thing is that OEMs like freakin' DELL, HP when the deploy Fusion A10 (Piledriver core) which already have more than decent integrated GPU they also tend to join them with 7750M/7850M GPU (CapeVerde-GCN1 Pro/XT) and to claim weaseley CFX support while most of their customers could get much more competitively priced laptop (for 100-150USD less) against oldish Sandy Bridges if they just stick with damn Fusion Gen2 APU.

btw i never thought that MX parts could be more powerful than originally branded parts (referring to MX420/440/MX4000 based on GeForce2 competing against GF4/GF5 class cards a decade ago)

[keritto]

Originally Posted by homerdog

So now we know where all the GK110s went

You really think that those inside are GK110 chips? Its just too optimistic to say at least.

Originally Posted by UniversalTruth

They, of course, might release it for desktop but basically they make huge margins on GK104, and without any performance pressure from the other party, then they will simply settle as much as they can with this situation. Even indirectly (or directly, who knows), both parties agree to keep the situation as it is.

Another opportunist You really think envy have a working GK110 silicon already and yet they suppress themselves from claiming their bragging rights :rofl:

It was never mentioned what kind of FP performance do they offer in their Cray XK7 on that link envydia was so kind to provide to us http://nvidianews.nvidia.com/Releases/NVIDIA-Powers-Titan-World-s-Fastest-Supercomputer-For-Open-Scientific-Research-8a0.aspx

Originally Posted by Gipsel

http://www.hpcwire.com/hpcwire/2012-10-29/titan_sets_high-water_mark_for_gpu_supercomputing.html

(...) What I find interesting, is that the XK6 nodes obviously gets upgraded to XK7 ones in the process. Originally it was planned that the Tesla cards are just drop in extensions for the new XK6 nodes (it should have worked as Jaguar just got upgraded from XT5 to XK6). What is strange, is that HPCWire claims a max power consumption of 12.7 MW (up from prior 10.8 MW afaik), while the specs from Cray say only 54.1 kW per rack, same as with XK6. No idea what to make out of that.

You could put that the whole real life story is a pure wild estimation. Just like we're doing it right here just not claiming any expert proofs.

If it's really fully functional GK110 (well with SMX disabled according to your numbers) why then we shouldn't hear any specs yet?

[trinibwoy]

Originally Posted by keritto

You really think that those inside are GK110 chips? Its just too optimistic to say at least.

You're joking right?

[keritto]

Originally Posted by ninelven

Yes, the 128bit bus is just odd... I guess they felt it would be too close to the 660 otherwise, but if that was the case, I think they would have been better off just using lower clocks on the 650 Ti. As for competing with the 7850, that isn't really its job. Nvidia's 4th tier card is the 660. The pricing reflects Nvidia's belief they can charge a premium for their name. For the high end, that may be true, but I doubt they will be so lucky in the value segment.

They already binned this parts by working SMx units. If they gone for another binning by speed that would require additional cost. I guess they placed some GTX650Ti into space just to mess around with AMDs sales and not to gives us best performance that chip could offer running at 192-bit bus. And btw all those GTX660/GTX650Ti are way overpriced just judging with DAMN SHORT PCB they use ... they could be a far cheaper parts than even HD7770 BOTH
But envy aint worried about some dumb customer that would walk into store and just bought a card judging by manufacturers shiny box and not by real performance. And those would be soon (which is GREAT9 replaced with better performing card hopefully again from same manufacturer (envy9 because people are just suckers for the branding.

Unfortunately this is how PC market works :rofl:

Originally Posted by trinibwoy

You're joking right?

You're joking right? (Not that you're funny to me, but at least you find yourself to be amusing)

[ninelven]

I'm sorry I don't understand your post, or even why you are quoting me.

And yes, that is GK110 (AKA K20) in Titan.

http://www.cray.com/Products/XK/Specifications.aspx

http://investors.cray.com/phoenix.zh...cle&ID=1750839

Or are you suggesting that Cray is lying?

[Gipsel]

Originally Posted by trinibwoy

You're joking right?

My guess would be the day was a bit too long and he is now recreating by trying the role of a troll on a rampage.

[keritto]

Originally Posted by ninelven

I'm sorry I don't understand your post, or even why you are quoting me.

And yes, that is GK110 (AKA K20) in Titan.

http://www.cray.com/Products/XK/Specifications.aspx

http://investors.cray.com/phoenix.zh...cle&ID=1750839

Or are you suggesting that Cray is lying?

I quote you about GTX650Ti being only 128bit and its marketing position its more than evident if you try and read.

And when its about Cray they never stated they use GK110 but it could be concluded from "Peak performance: 100+ Tflops per system cabinet -- NVIDIA® Tesla® K20 GPU Accelerators, up to 96 per cabinet" relation because GK104 itself have FP32 performance of 3Tflops so the numbers should be higher. Anyway they still didnt brag nothing about GK110 and that they was meant to be envydia not cray.

[ninelven]

They explicitly state the GPU is K20 (GK110) in both the specs and press release.

But perhaps you should inform the Department of Energy they have been had.

Originally Posted by keritto

I quote you about GTX650Ti being only 128bit and its marketing position its more than evident if you try and read.

I'm afraid I still do not understand, and quite frankly, I don't think I ever will. Additionally, given your irrational response on the Titan matter, I am inclined to disregard whatever else you might say. Sorry.

[rpg.314]

Originally Posted by Gipsel

Yes, quite a bit. Basically all tools, compilers, and libraries would need to be exchanged.

Or more realistically, real tools, compilers and libraries would have to exist for AMD.

[Gipsel]

Originally Posted by rpg.314

Or more realistically, real tools, compilers and libraries would have to exist for AMD.

They created a lot of them for Titan (or the XK6/7 clusters in general), so one would need to repeat that, albeit from a lower base.

[boxleitnerb]

It is 14 SMX, now it is definitive and confirmed by Anandtech:

NVIDIA's K20 is the server/HPC version of GK110, a part that never had a need to go to battle in the consumer space. The K20 features 2688 CUDA cores, totaling 7.1 billion transistors per GPU built using TSMC's 28nm process.

http://www.anandtech.com/show/6421/i...idia-gpu-cores

[lanek]

Originally Posted by boxleitnerb

It is 14 SMX, now it is definitive and confirmed by Anandtech:



http://www.anandtech.com/show/6421/i...idia-gpu-cores

Excellent article..

2688CC ...

Im a bit worry about the clock speed for make they running.. ( 1536CC @1-1.1ghz/195W vs 2688CC @ ? )

I believe the Tesla card ( not like on the titan, but retail ), should not go over 225W. So with certainly a core speed around 800mhz max.
I really doubt after the good return on the consumer and press side about the TDP of the GK104, Nvidia want release a 250W+ gamers card.

( but at the same time, AMD have the same problem with their new series who will stay on 28nm process, if they have not improve dramatically the efficiency here and there, they will need more SP and so more TDP. )

[boxleitnerb]

Take away 3 GB GDDR5 and add the turbo that should keep GK110 within its TDP and I believe 800-850 MHz are possible. 250W TDP are okay as long as performance increases accordingly.

GK104 has about 165W consumption (card only) in games. GK110 should add 50% performance over GK104, then 250W are not a problem.

[iMacmatician]

Originally Posted by lanek

Im a bit worry about the clock speed for make they running.. ( 1536CC @1-1.1ghz/195W vs 2688CC @ ? )

According to HPCwire it's 732 MHz.

[lanek]

Originally Posted by iMacmatician

According to HPCwire it's 732 MHz.

I know, but i speak mostly if they do a "gamer high end card " based on it.

[boxleitnerb]

I wouldn't bet on more than 850 MHz.

[iMacmatician]

I'm guessing around 800 MHz (14 SMXes) at 250 W, which would give 10% better FLOPS/W than the 13 SMX K20.

Also, I'm wondering if there will be a second revision of the K20, at least for Teslas, in late 2013 to 2014 with 15 SMXes and faster memory.

[boxleitnerb]

2013 (rather 2014) will see Maxwell. A second Kepler-Tesla revision then wouldn't make much sense.

How much do 3 GB GDDR5@1250 MHz consume?

[tviceman]

Originally Posted by boxleitnerb

I wouldn't bet on more than 850 MHz.

Agreed, 800-850mhz is going to probably be what GK110 ends up running at in it's highest end Geforce form.

Originally Posted by boxleitnerb

How much do 3 GB GDDR5@1250 MHz consume?

I can't answer that, but I know I've seen power consumption numbers in reviews all over the web with gtx670 and gtx680's 4 gig cards that aren't really consuming that much more power than their 2gb counterparts.

http://www.xbitlabs.com/articles/gra...b_5.html#sect1 - the same power draw
http://www.pcper.com/reviews/Graphic...ption-and-Temp - pcper gtx670 4gb is running at a higher clockspeed, drawing 10 more total watts under load

I doubt the removal or addition of 3gb of vram affects the vendor's final TDP numbers. We are probably talking less than 5 watts total (guess).

GTX680 is bandwidth constrained. It probably needs 6.6ghz vram before the bottleneck is mostly (or entirely) alleviated. That said, a 14 SMX GK110 with 1500mhz ram and 800mhz clock speed, would have 40% more core performance and 50% more ROP's and memory bandwidth. The gtx580 had 25% more core performance and 25% more bandwidth (50% more ROP's) than the gtx560ti and ended up consistently ~40% faster. It will come down to TDP, but the potential for huge performance out of GK110 is there.

[kalelovil]

Originally Posted by tviceman

I can't answer that, but I know I've seen power consumption numbers in reviews all over the web with gtx670 and gtx680's 4 gig cards that aren't really consuming that much more power than their 2gb counterparts.

http://www.xbitlabs.com/articles/gra...b_5.html#sect1 - the same power draw
http://www.pcper.com/reviews/Graphic...ption-and-Temp - pcper gtx670 4gb is running at a higher clockspeed, drawing 10 more total watts under load

I doubt the removal or addition of 3gb of vram affects the vendor's final TDP numbers. We are probably talking less than 5 watts total (guess).

It depends if that extra RAM is actually in use. In those game benchmarks on a GK104 it will presumably be idle most of the time doing little reading or writing, while the opposite will be the case on a K20 running optimised code.
The TDP difference will be more significant than those benchmarks show since TDP measures theoretical peak power.

[boxleitnerb]

Good point! One should benchmark with 8xMSAA, record all the power consumed over the course of the benchmark, see to it that it is repeatable value and then compare 2GB vs 4GB cards. To be as accurate as possible, one should also disable any boost (if one is active on the card).

[iMacmatician]

Originally Posted by boxleitnerb

2013 (rather 2014) will see Maxwell. A second Kepler-Tesla revision then wouldn't make much sense.

For my statement above I was assuming that the first Maxwell chips would be like GK104, so not a "real" successor to GK110. So there would be a place for a GK110 revision, but they may not do that in any case.

Originally Posted by tviceman

GTX680 is bandwidth constrained. It probably needs 6.6ghz vram before the bottleneck is mostly (or entirely) alleviated.

How would a ~1100 MHz 8 SMX GK114 with 7 Gbps memory do in terms of bandwidth constraints? Could it be ~15% faster than the 680?

[rpg.314]

Originally Posted by Gipsel

They created a lot of them for Titan (or the XK6/7 clusters in general), so one would need to repeat that, albeit from a lower base.

My guess is a lot of sw support for the GPUs came from NV. I dont see AMD providing anything more polished beyond their DX11 driver.

[Psycho]

Originally Posted by tviceman

GTX680 is bandwidth constrained. It probably needs 6.6ghz vram before the bottleneck is mostly (or entirely) alleviated. That said, a 14 SMX GK110 with 1500mhz ram and 800mhz clock speed, would have 40% more core performance and 50% more ROP's and memory bandwidth. The gtx580 had 25% more core performance and 25% more bandwidth (50% more ROP's) than the gtx560ti and ended up consistently ~40% faster. It will come down to TDP, but the potential for huge performance out of GK110 is there.

A 14 SMX GK110 with 800 mhz core(*) and 1500mhz memory, compared to a (average review clock) 1080mhz 680:
+29% ALU/TEX
+50% BW
+11% ROP
-8% setup(!)

580 to 560ti:
+25% ALU (but higher efficiency)
+50% BW
+40% ROP (but not export)
+100% setup
and yes, around +40% game performance
But shader/texture units are very different between gf104 and gf110, so it's hard to compare the numbers to the GK110 situation, where it looks like register space (doubt it will mean much/anything for usual graphics) and L2 cache is the only graphics related difference.

But claiming as much as 50% game performance seems overly optimistic from the above numbers..

(*) I would probably expect more like 850mhz usual boost, but ofcourse depends on yields and binning of those tesla cards. If the 225w 725mhz teslas (double memory, but no fan) is from the creme-de-la-creme bin, 850mhz with 250-300w can be hard on the "cheap" geforces. The teslas are clocked for maximum efficiency, while the geforces will be for maximum performance on a acceptable power budget, ie just before the power-curve goes completely wild.

[tviceman]

Thanks for the info, I didn't realize the memory controllers were different on GK110. What is the decrease in the setup?