(WCCF) Nvidia ‘Big Daddy’ Maxwell GM200 GPU Alleged Specs Leaked

[desprado]

There some specs leaked about GTX 980 Ti/Titan X or II.

• Die Size: 551mm^2
• SMMs/CUDA Cores: 20-22/ 2560-2816
• Memory Bus: 384 Bit
• Performance: ~50% Boost over GTX Titan Black
• Launch: Q4 2014

http://wccftech.com/nvidia-geforce-gtx-980-ti-gtx-titan-x-coming/#ixzz3EHspvYa2
​

 

[tviceman]

Sounds believable, but who knows for sure if this is accurate. I guessed that the Titan II version (fully unlocked and loaded GM200) would be 35% faster than GTX980, while the GTX 980 TI version (cut down GM200) would be 20% faster than GTX980.

 

[alcoholbob]

Would make sense the Halo Card would be the Titan II and they could do a cut down version to sell for $650-700 as the 980Ti or 990.

 

[f1sherman]

yeah fat enough to be believable, but this is coming from a really poor leak

http://pctuning.tyden.cz/component/...-geforce-gtx-990-a-nejspis-i-titan-x-na-ceste

 

[Fallen Kell]

I think they will do exactly what the last paragraph speculates. If AMD comes out with something that has a performance improvement, Nvidia will release it. If AMD doesn't, then Nvidia may just sit on this and wait for 20nm to be ready.

 

[Silverforce11]

Maxwell is pretty epic, if they make a big one with the same die area, certainly 35-50% is possible depending on aggressive they clock it.

It's almost unheard of to get such an improvement, at the same node & die area, that's how good Maxwell is.

 

[Cookie Monster]

The die size seems a bit too small no?

Anyway, they'd want to release in the form of Tesla/Quadros where all the margins are at anyway

 

[RussianSensation]

Sounds believable, but who knows for sure if this is accurate. I guessed that the Titan II version (fully unlocked and loaded GM200) would be 35% faster than GTX980, while the GTX 980 TI version (cut down GM200) would be 20% faster than GTX980.

Why such a huge difference between Titan II and 980TI? First, 780Ti > Titan in performance. Second, an after-market Titan Black is barely faster than an after-market 780TI which makes me think an after-market gaming focused 980Ti shouldn't be 15% worse than a Titan II.

Also, your estimate of being only 20% faster than a 980 seems overly conservative.

Using TPU:

980 is 7.5% faster than 780Ti at 4K and 1080p
980 is 6.4% faster than 780Ti at multimonitor
980 is 5.3% faster than 780Ti at 2560x1600

On average a 980 is only 6% faster than a 780Ti, which is inline with Computerbase's 5-6% faster and 3DCenter's 7.5% faster (570% vs. 530%).

Based on the ~50% faster than Titan Black claim in the article and based on the projected specs of 2816 CUDA core GM200, it should beat 980 by 30-40%.

I don't see NV releasing GM200 that's 30-40% faster than 980 at $699 though as that would make 980 at $549 a dead card. If NV releases the 980Ti GM200 gaming card this year, the price would have to be > $699 or NV will delay this chip to 2015 while milking the mid-range 980 for as long as possible. Also, it would look great for NV if in 5-6 months AMD launched a 390X which beat 980 but then NV trumps that in 1 week with GM200 and takes the crown back. NV likes to keep an ace in its hand.

 

[escrow4]

*** Throws hundreds at Nvidia ***

Day 1 purchase for me. Although pffft should be 512 bit.

 

[RussianSensation]

I think they will do exactly what the last paragraph speculates. If AMD comes out with something that has a performance improvement, Nvidia will release it. If AMD doesn't, then Nvidia may just sit on this and wait for 20nm to be ready.

The other possibility is 2nd gen Maxwell = 28nm. Then 3rd gen Maxwell in late 2015/early 2016 = 20nm. Then late 2016 low end Pascal card like 750Ti replacement and mid-range and high-end Pascal in 2017. Pascal in 2016 is a very vague timeline since even releasing 1 Pascal chip meets that deadline. There is a lot of time between now and Pascal flagship for NV to pull off both 28nm and 20nm shrink for a refresh of Maxwell. Not sure if they'll do it, but it's possible.

 

[USER8000]

The die size seems a bit too small no?

Anyway, they'd want to release in the form of Tesla/Quadros where all the margins are at anyway

The Czech leak said 10% to 15% larger than a GK110 and 40% to 50% faster than a Titan Black.

 

[ShintaiDK]

The die size seems a bit too small no?

Anyway, they'd want to release in the form of Tesla/Quadros where all the margins are at anyway

550-600mm2 sounds about right. But whatever these specs are real or not is a whole other matter.

 

[raghu78]

Why such a huge difference between Titan II and 980TI? First, 780Ti > Titan in performance. Second, an after-market Titan Black is barely faster than an after-market 780TI which makes me think an after-market gaming focused 980Ti shouldn't be 15% worse than a Titan II.

I am going with the assumption of a 550 - 600 sq mm 28nm GM200. If Nvidia wants to sell GM200 as flagship Quadro and Tesla products they will have to pay a die area cost to enable Quadro / Tesla specific features for workstation/ HPC market and higher double precision performance. GTX 980 has lower DP performance than GTX 580.

http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/20

a GM200 at 28nm with all the HPC and DP features will not have the same perf / sq mm as GM204. btw I would expect a GM200 to sport 2816 - 3072 cuda cores but at slightly lower clocks to fit within a 250W TDP.

Also for compute/HPC GM200 will have to be clocked significantly lower as those workloads tax the GPU much more consistently than gaming

http://www.tomshardware.com/reviews/nvidia-geforce-gtx-980-970-maxwell,3941-12.html

I don't see NV releasing GM200 that's 30-40% faster than 980 at $699 though as that would make 980 at $549 a dead card. If NV releases the 980Ti GM200 gaming card this year, the price would have to be > $699 or NV will delay this chip to 2015 while milking the mid-range 980 for as long as possible. Also, it would look great for NV if in 5-6 months AMD launched a 390X which beat 980 but then NV trumps that in 1 week with GM200 and takes the crown back. NV likes to keep an ace in its hand.

NV will milk GM204 for as long as it can and when AMD has a R9 390X released NV will release desktop GM200. NV would also want to release GM200 in the HPC/workstation market first where it can sell it for much higher prices and margins. When AMD has a R9 390X NV can launch GM200 and can easily trim prices across the stack at that time. Ideally the GTX 980 can come down to USD 449 while the GTX 970 can remain at USD 329.

GTX Titan II (GM200) - 999 (successor to GTX Titan Black.)
GTX 980 Ti (GM200) - 649 (successor to GTX 780. )
GTX 980 (GM204) - 449 (successor to GTX 770)
GTX 970 (GM204) - 329 (successor to GTX 760)

AMD needs to reply with a strong GCN 2.0 architecture otherwise Nvidia will rule for the next 2 years and AMD will be further marginalized in the GPU market.

 

[CakeMonster]

Anyone could have guessed specs in that ballpark. The real question here is release date and price.

Many who are debating whether to upgrade right now (980) would like to know if the pricing will be as insane as we've seen before. Might as well take the plunge now if this one will be $800. I would be rather annoyed if I waited until December then finding out it was out of my price range anyway and I could have had a 980 all along...

 

[Erenhardt]

Anyone could have guessed specs in that ballpark. The real question here is release date and price.

Many who are debating whether to upgrade right now (980) would like to know if the pricing will be as insane as we've seen before. Might as well take the plunge now if this one will be $800. I would be rather annoyed if I waited until December then finding out it was out of my price range anyway and I could have had a 980 all along...

Then vote if your wallet. If you buy gtx980, you let them know that you accept those high-end prices for mid-range graphics. If you then upgrade from 980 to 990ti (or titan2 or whatever it will be called) you let them know that you will pay two times for the same architecture.

Soon we will be introduced to a single tier of new generation at the time. Down from a little GX107. After everyone upgraded to GX107, they will release GX104. When people upgraded from GX107 to GX104, they will release GX100. By the time they go through all tiers, new generation will be knocking our door. Rinse, repeat.

The only one responsible for the situation we have is consumer, no one else

 

[ShintaiDK]

Then vote if your wallet. If you buy gtx980, you let them know that you accept those high-end prices for mid-range graphics. If you then upgrade from 980 to 990ti (or titan2 or whatever it will be called) you let them know that you will pay two times for the same architecture.

Soon we will be introduced to a single tier of new generation at the time. Down from a little GX107. After everyone upgraded to GX107, they will release GX104. When people upgraded from GX107 to GX104, they will release GX100. By the time they go through all tiers, new generation will be knocking our door. Rinse, repeat.

The only one responsible for the situation we have is consumer, no one else

Its also a matter of cashflow. You can see AMD doing the exact same thing for the exact same reasons.

Without affordable node progress, this is simply unavoidable. And prices on GPUs will only increase with time as volume shrinks while R&D and node costs only goes up.

 

[hawtdawg]

I was gonna jump from 780's to 980's, but now I think im definitely gonna wait.

 

[Genx87]

So this looks like a 28nm part then with a possible shrink to 20nm in 2015? I guess that seems believable. Unfortunately it also appears if true 20nm parts are further off than the Spring

 

[therealnickdanger]

I just want DP 1.3 and 120 Hz 4K monitor with G- or F-Sync. Whether that's AMD 3xx or Nvidia GM200, I'll give them my money.

 

[scooterlibby]

I'm hoping for something like this. The 980 is tempting, but this could be a true single GPU replacement for my 690. I wouldn't mind seeing two 970's on a single PCB either.

 

[Subyman]

The price stack does get weird when you factor in the 970's. GTX 980 price barely makes sense. A TI version for 15% more performance over the 980 for $650-700 and a Titan II at 25-30% more performance over a 980 for $1000 is hard to get excited about. That's 3-way SLI on 970's, or simply 2-way would beat it for much less. Obviously, some balk at the negatives of SLI and there is no touching SLI of the Titan II. However, that's only like 100 dudes in the world that want to spend $2000 on GPUs.

I'd like to think nvidia would hold them back for 20nm and really make a leap unless AMD responds and makes it necessary.

 

[el etro]

These specs are speculation specs by the 3dcenter moderator Ailuros.

 

[tviceman]

Why such a huge difference between Titan II and 980TI? First, 780Ti > Titan in performance. Second, an after-market Titan Black is barely faster than an after-market 780TI which makes me think an after-market gaming focused 980Ti shouldn't be 15% worse than a Titan II.

I don't think GM200 can have as big a leap over GM204 as GK110 over GK104 because there isn't enough die space left moving up from 398mm^2 to accommodate all the compute-specific transistors on top of more memory controllers, ROP's, and SMM's. GK110 had 87.5% more cores than GK104 and a 250mm^2 larger die size. In my mind (before this thread was started), I was thinking GM200 could maybe squeeze out 50% more cores than GM204, adding up to 24 SMM's and 3072 cores. I still think that is a possible outcome. 22 SMM's make no sense because nothing adequately divides into 22 for balance purposes, and since 20 is a rather small (25%) increase over GM204's 16, 24 makes the most sense to me.

But 50% more cores, ROP's, and bandwidth != 50% more performance. I think 35%, maybe 40% max more performance than GM204 with a real-world power usage slightly lower than GK110 is both feasible and reasonable for a fully functional GM204 chip. On top of that, I think GM200 is going to come before GCN 2.0, so I think Nvidia will slow roll GM200 just like they did with GM204. If it is to have 24 total SMM's, I think Nvidia would release a gtx 980 TI with 20 SMM's active and DP castrated. Titan II would have 24 SMM's and full functional DP operation. Then later on Nvidia will release a full functional, DP castrated GTX 985 (or whatever) to combat a new high end AMD chip, or if no chip exists, just to get more sales at a higher margin than 980 TI.

TL;DR I think GM204 on 20nm will have a slightly lower TDP than GK110, but will not be quite as much a jump over GM204 as GK110 is over GK104. I also think GM200 will have some serious OC headroom just like GM204 has, and I think Nvidia will slow roll GM200 (introducing harvested dies first) like they did with GK110.

 

[cbrunny]

I think they will do exactly what the last paragraph speculates. If AMD comes out with something that has a performance improvement, Nvidia will release it. If AMD doesn't, then Nvidia may just sit on this and wait for 20nm to be ready.

I'm not sure I understand. Is the assumption that this card would not turn a profit? And that this countering AMD's performance improvement will simply maintain market share?

 

[badb0y]

How likely is it that this lands before The Witcher 3?