[tviceman]

Originally Posted by iMacmatician

From a post by sontin on AnandTech forums:

(I can't find the seller at the moment. EDIT: http://www.cadnetwork.de/konfigurato..._g13/system=88, thanks Ailuros!)

If true, that's 13% more theoretical SP FP performance and (if 6 Gbps memory is used) 25% more bandwidth than the 680. If a GeForce version is even further cut down, it would seem that there's almost no point in releasing it unless they are planning to pull what Blazkowicz mentioned and put it at or below the 780.

On the other hand, the Tesla/Quadro cards this generation are clocked a lot lower than the corresponding GeForce cards with the same CCs, so a GK110 GeForce may be clocked considerably higher than 700 MHz, even if it is cut down more in terms of units.

On another note, this seems to contradict rumors of Tesla K20 memory size of a multiple of 6 GB, unless there will be multiple K20 variants (in terms of SMXes, bus width, etc. not just memory). EDIT (the same edit): The link shows this K20 is 5 GB.

First gen Fermi Tesla cards had gtx470 spec a, i.e. lower than the gtx480. If GK110 isn't coming as a geforce card until March, they have plenty of time to do another respin and get higher yields with more functional units.

[mczak]

Originally Posted by lanek

Outside that, for the 200gb/s, it look normal taking in account the memory is ECC ( so performance on real bandwith are lowered )

No that figure is without taking ECC into account so it is indeed quite low (25% lower than FirePro W9000 and way lower than Xeon Phi will be) since it is apparently only using 5 of the 6 memory channels. Effective memory bandwidth (with ECC) is lower than that of a GTX 680 (without ECC).

[lanek]

Originally Posted by mczak

No that figure is without taking ECC into account so it is indeed quite low (25% lower than FirePro W9000 and way lower than Xeon Phi will be) since it is apparently only using 5 of the 6 memory channels. Effective memory bandwidth (with ECC) is lower than that of a GTX 680 (without ECC).

I was outline the words on the article about " 200GB's lie, the bandwith is lower than that in reality " .. saying this is normal, ECC lower performance, and the 200gb/s will never been reached with ECC .

Outside than that, this card dont look to be the "High end Tesla " ... 320bits memory controller ( should be 384bits on full model ), 5Gb ( should be 6gb ), and 15SMX annonced ( 13 SMX here ).
( but at the same time, 200mhz more ( ~ 900mhz ) look to be enough for reach the Tflops/SP annonced for the K20 ) . ( maybe my calculation is bad )

Now.. maybe this explain why we see this card enter the market by some backdoor.. ( Cad material configurator ).. I dont say the full model will be like that, but i still dont remember seeing Nvidia let this type of things happend with the " big models ".

[iMacmatician]

Just as an FYI, the CADNetwork page no longer lists the specs for the Kepler K20 (just the name only now).

[lanek]

Originally Posted by iMacmatician

Just as an FYI, the CADNetwork page no longer lists the specs for the Kepler K20 (just the name only now).

Damn, i was know we should not speak so much about it ..

[UniversalTruth]

View nVidia Kepler 2013 refresh generation

[tviceman]

Yeah I see GK110's fully unlocked Geforce card starting out at $599. Possibly even $649. I don't think that graph has GK114's initial MSRP correct, though. I don't see Nvidia pricing it $200 lower than the $500 MSRP it had in March. I think GK114 will have 10% more performance than GK104, lower power consumption a tiny bit, and start out at $399 for the top sku.

[LittleJ]

If AMD Oland HD 8870 comes in at ~270mm2 and similar performance to GK104(114) I doubt they could command 399$ for a mid range(at the time) GPU.

[sonen]

Originally Posted by tviceman

Yeah I see GK110's fully unlocked Geforce card starting out at $599. Possibly even $649. I don't think that graph has GK114's initial MSRP correct, though. I don't see Nvidia pricing it $200 lower than the $500 MSRP it had in March. I think GK114 will have 10% more performance than GK104, lower power consumption a tiny bit, and start out at $399 for the top sku.

There are already $350 GK104 out there.

[Ailuros]

Originally Posted by sonen

There are already $350 GK104 out there.

Inevitably will a salvage part from N core have a lower price for obvious reasons; it won't make much difference for future GK114 respective salvage parts.

The point is that it highly depends in which configuration and at what MSRP NV might release a GK110. If the price would be as low as $499 (whereby "low" is relative; that was the exact MSRP for the 480 and 580) and you'd consider that it'll have at least one salvage part, anything above ~$300 for a GK114 wouldn't make sense. If the GK110 price ends up higher than it'll keep a similar analogy from top to bottom in SKU pricing, with higher prices overall.

[iMacmatician]

So the K20 apparently has 5 Gbps memory, the same as the GK104 K10, does that point towards GeForce GK110s (or higher end K20s if they exist) and GeForce GK114s topping out at 6 Gbps (instead of, say, 7 Gbps), just like with the GK104?

[tviceman]

Originally Posted by iMacmatician

So the K20 apparently has 5 Gbps memory, the same as the GK104 K10, does that point towards GeForce GK110s (or higher end K20s if they exist) and GeForce GK114s topping out at 6 Gbps (instead of, say, 7 Gbps), just like with the GK104?

Probably. I think the more memory controllers there are, the avg. max speed vram can run at will be a little lower (since each memory controller has it's own transistors and chances of minor defects).

Hopefully GK114 pushes it's reference vram speed up to 6.6ghz or even slightly higher. It will need all the bandwidth improvement it can get.

[Ailuros]

Originally Posted by iMacmatician

So the K20 apparently has 5 Gbps memory, the same as the GK104 K10, does that point towards GeForce GK110s (or higher end K20s if they exist) and GeForce GK114s topping out at 6 Gbps (instead of, say, 7 Gbps), just like with the GK104?

Both the K10 as well as that supposed 13 SMX/K20 have a TDP of 225W. The GTX690 has each core clocked at 915MHz (+turbo) vs. 745MHz (no turbo) on K10 and has 6.0Gbps ram instead of 5.0Gbps on K10. The GTX690 has also a TDP of 300W, which would be quite on the steep side IMHO for a desktop single chip high end SKU. For mGPU 290-300W (if not even somewhat more) isn't that much of an issue.

If those 13SMX K20 specs are true, NV clearly needs at least one good metal spin to get the GK110s in better shape if they intend to go for 14/15SMX cores with a 800-850MHz core frequency, 6.0 Gbps ram and a power consumption that wouldn't be all that much higher than 250W.

[tviceman]

Originally Posted by Ailuros

Both the K10 as well as that supposed 13 SMX/K20 have a TDP of 225W. The GTX690 has each core clocked at 915MHz (+turbo) vs. 745MHz (no turbo) on K10 and has 6.0Gbps ram instead of 5.0Gbps on K10. The GTX690 has also a TDP of 300W, which would be quite on the steep side IMHO for a desktop single chip high end SKU. For mGPU 290-300W (if not even somewhat more) isn't that much of an issue.

If those 13SMX K20 specs are true, NV clearly needs at least one good metal spin to get the GK110s in better shape if they intend to go for 14/15SMX cores with a 800-850MHz core frequency, 6.0 Gbps ram and a power consumption that wouldn't be all that much higher than 250W.

I think few other things people aren't taking into consideration is 1) the improvements made by a more mature node process and 2) Nvidia isn't done tweaking Kepler, hence the reasons we are getting refreshes. Despite their really good performance per watt this time around, there is still probably room to tweak and squeeze out a little bit more performance AND / OR draw a little less power. Add up these points, and it's not unreasonable to see GK110 performing ~50% faster than GK104 @ 250 watt TDP.

[UniversalTruth]

Originally Posted by tviceman

Add up these points, and it's not unreasonable to see GK110 performing ~50% faster than GK104 @ 250 watt TDP

I really hope this becomes reality because it is the only possible option to see low prcies from AMD, if we see significantly faster geforces, otherwise both will charge as much as they wish.

[jimbo75]

Originally Posted by tviceman

I think few other things people aren't taking into consideration is 1) the improvements made by a more mature node process and 2) Nvidia isn't done tweaking Kepler, hence the reasons we are getting refreshes. Despite their really good performance per watt this time around, there is still probably room to tweak and squeeze out a little bit more performance AND / OR draw a little less power. Add up these points, and it's not unreasonable to see GK110 performing ~50% faster than GK104 @ 250 watt TDP.

Isn't the assumption that the perf/watt of Kepler is due in a large part to it's castrated compute?

[jimbo75]

Originally Posted by UniversalTruth

I really hope this becomes reality because it is the only possible option to see low prcies from AMD, if we see significantly faster geforces, otherwise both will charge as much as they wish.

AMD cards are very low priced, honestly I don't know how much lower you expect them to go.

[UniversalTruth]

Originally Posted by jimbo75

Isn't the assumption that the perf/watt of Kepler is due in a large part to it's castrated compute?

How much compute castrated do you think GF114 is?

Originally Posted by jimbo75

AMD cards are very low priced, honestly I don't know how much lower you expect them to go.

250$ for the 7970, otherwise they lose me as a customer.

[jimbo75]

Originally Posted by UniversalTruth

250$ for the 7970, otherwise they lose me as a customer.

So what are you gonna buy - one of those great value Nvidia cards?

You won't see 7970's at $250 because the 8870 is likely to be faster and cheaper before then. The days when AMD's top tier card dropped so far in price are over I think.

[UniversalTruth]

Originally Posted by jimbo75

So what are you gonna buy - one of those great value Nvidia cards?

Of course not. I will stay as long as I can with my oldie Barts. For several more years. And instead of giving my money for revenue, I will be in sleepy position waiting for the inevitable...

[jimbo75]

Originally Posted by UniversalTruth

Of course not. I will stay as long as I can with my oldie Barts. For several more years. And instead of giving my money for revenue, I will be in sleepy position waiting for the inevitable...

You need to get used to the reality of the situation - the desktop market is imploding and enthusiast parts will command enthusiast prices else there is no point designing and fabbing them. There is simply no point in AMD offering a card like the 7970 for as low as $250.

[Ailuros]

Implosion won't get any better if prices stay as high as they are, rather the contrary. Simple market rules: the higher the price the lower the demand.

[jimbo75]

AMD just lost $200 million in Q3 at these current "high" prices.

It should be obvious to everyone what AMD's strategy at the high end is. They realise that no matter how much faster they are, people will still pay more for Nvidia cards. Tahiti probably has about 1/5th to 1/10th the wafer starts of GK104. Those claims that the 680 outsold the 7970 by 4-1? Undoubtedly true.

If you're all waiting on AMD gifting you cheap high-end cards again, you'll be waiting a while. Those wafers are better spent on the midrange and below where AMD can actually command a fair price.

Expect the situation to get worse - assuming they don't just pull out of the high-end altogether. What would you do if you were them and your customers were constantly waiting on price drops before buying?

[tviceman]

Originally Posted by jimbo75

Isn't the assumption that the perf/watt of Kepler is due in a large part to it's castrated compute?

That probably has some effect on the big improvements Nvidia made. Getting rid of the hotclocks probably had a larger effect, though. The "compute" parts of the chip don't do much of anything when gaming, hence those transistors aren't going to be burning through too much power when playing BF3 or Crysis 3 or whatever.

[Ailuros]

Originally Posted by jimbo75

AMD just lost $200 million in Q3 at these current "high" prices.

I rarely follow IHVs financials but there's anything to substantiate the entire loss can be attributed just to that? Considering they're not selling only GPUs if there's nothing to back it up, it sounds quite an unfair conclusion to me; especially since their CPU offerings are quite a bit behind those from Intel.

It should be obvious to everyone what AMD's strategy at the high end is. They realise that no matter how much faster they are, people will still pay more for Nvidia cards. Tahiti probably has about 1/5th to 1/10th the wafer starts of GK104. Those claims that the 680 outsold the 7970 by 4-1? Undoubtedly true.

If you're all waiting on AMD gifting you cheap high-end cards again, you'll be waiting a while. Those wafers are better spent on the midrange and below where AMD can actually command a fair price.

Hold on a second. That strategy is NOT something that is holding on for years, but rather the exact opposite. And it's also quite debatable what "high end" really is in the end. A performance chip with a high end price definitely yes.

If any IHV watches its direct competition by an as high margin as you're claiming, then the only other sensible solution would be to further reduce prices. I'm not even buying that 4:1 analogy unless there's some serious statistic to back it up.

Expect the situation to get worse - assuming they don't just pull out of the high-end altogether. What would you do if you were them and your customers were constantly waiting on price drops before buying?

Any sort of detailed armchair analysis considering AMD's strategic decisions in total for the past few years won't help especially not in this thread, but let's just say that I disagree with a LOT of them from where I stand.

What I can tell you though for sure is that I as a user refuse to pay 400-500+ price tages for performance chips irrelevant from which IHV.