Nvidia Maxwell GM200 pictured

[n0x1ous]

Usman calculates 632mm2.

http://wccftech.com/nvidia-maxwell-gm200-die-shot-analysis/

Beastly - gotta be the largest thing TSMC has ever fabbed

 

[Enigmoid]

I really think NV will repeat mid-range --> Titan --> flagship strategy. All things are pointing to this. This allows them to maximize profits on a 980 and Titan while holding back 1080Ti until they can launch it with even higher clocks by Spring. With 16nm FinFets having delays to late 2016, I just don't see NV launching 1080Ti in the next 2 months. We'll see.

Sure looks this way.

Other thing that it looks like is that is is a huge chip (600 mm^2). Prices are going to be high and remain high which may really hurt nvidia against a possible 390x (if AMD does use 20 nm).

 

[Arachnotronic]

Nice!

 

[Lepton87]

Agreed. We're getting a Titan X (or whatever) first. A DP-castrated Geforce card will launch later on. Unfortunately this is the new norm. I think for any future upgrades, I'm going to be upgrading mid-cycle or whenever price drops + bundles occur.

Nah, this chip won't have any DP capabilities beyond that of GM204. GM204 is at 400mm2 and 50% more units and you are at 600mm2, where do you fit that DP performance?
ps. I'm quite surprised I didn't know about GK210, it's very unusual and never been done before by NV. I think they wouldn't have done that if they were to release a new DP chip so close in time IMHO GM200 is focused on SP performance.
ps. And I just bought water blocks for my Titans.... eh, that just makes two of them obsolete. Oh, well I'll wait for a node shrink with 6GB I'm good to go. Maybe after the release of this monster the orginal Titan will get really cheap and I'll buy another to hold me over until next node. I wonder if this will be 20 or 16ff/14

 

[Vesku]

Well they are bringing this chip in as the next top end Quadro so it probably has some DP specific goodness inside that extra die area.

 

[alcoholbob]

Photo analysis isn't precise. Based on the leaks of GM204 people estimated it was a 432mm^2 chip, it ended up being 398mm^2.

 

[sontin]

Well they are bringing this chip in as the next top end Quadro so it probably has some DP specific goodness inside that extra die area.

Yeah, this talk about "none" DP performance is beyond ridiculous.
GM200 is 10-20% bigger than GK110 but will only have 50% more Cluster and 6% more Cuda cores.

GM204 is 33% bigger than GK104 with 100% more cluster and 33% more Cuda cores.

GM200 will have a 2:1 ratio between FP32 and FP64 because Maxwell only has 128 instead of 192SP units.

 

[tolis626]

LOL I tried doing that when I was in med school and it was fine for the first two years because I was in the library all day. Once clinical rotations started, I switched to gaming laptops and then eventually a desktop my last year. Just get the dgpu now and enjoy some BF4 during study breaks.

Well holy crap, I'm in med school too! Coincidence or what?

Anyway, I think you understand how things go. I try to squeeze everything in my schedule, but with exams it's just not possible. At least I'm a year and a half away from going to 100% clinical training, so I got time at home.

Well, if I have the money at some point no one is forcing me to keep the 970. I can always sell it and get whatever the best thing is at that moment. Or, you know, I can wait. But I don't like waiting.

Anyway... Regarding the DP/no DP discussion. I don't think this will be an SP card. It just doesn't make sense for NVidia to manufacture such a big chip with only SP in mind. Sure, 28nm is a mature process and all, but we're talking about a jumbo-sized chip here. It may end up being 580sq.mm or 650sq.mm for all I care. It's still huge, it WILL be expensive to manufacture and NVidia isn't a charity, they WILL pass down the cost to the buyers. If professionals (Don't most pros need DP?) don't pay for most of it, I dread to think about how much a potential GeForce card will cost. 1500$ single-GPU card anyone? Yikes...

OTOH, I am very curious to see how these things will overclock. With such huge dies, overclocking potential gets... Interesting. Sure, it probably won't sip power like the 980 and 970 do, but that heat will have to be dissipated from a huge (Comparatively speaking) surface area. So it will definitely be interesting to see how water/LN2/DICE cooling guys will do. And with Maxwell's overclocking in general, I can hope for awesome things to happen.

With all that said, I'm even more interested to see how AMD will do this round. But that's a completely different discussion.

 

[nitromullet]

Definitely excited to see this. I had planned on getting a GTX 970, but got a pretty good deal on a GTX 780. I had mixed feelings about getting another 780, but with GM200 and Fiji/Bermuda XT looking more likely in the nearer future I think the GTX 780 will suffice for the time being.

 

[HurleyBird]

Well they are bringing this chip in as the next top end Quadro so it probably has some DP specific goodness inside that extra die area.

If anything that points to little emphasis on DP. While there is some overlap between Geforce/Quadro/Tesla the distinction is Geforce for gaming, Quadro for rendering, and Tesla for compute. GK110 began life as Tesla, and now we have GK210 that as far as I know is only available as Tesla. Tesla also tends to sell at a premium over Quadro, so why would GM200 begin life as a Quadro product if it weren't for a lack of DP or other compute features?

 

[Vesku]

If anything that points to little emphasis on DP. While there is some overlap between Geforce/Quadro/Tesla the distinction is Geforce for gaming, Quadro for rendering, and Tesla for compute. GK110 began life as Tesla, and now we have GK210 that as far as I know is only available as Tesla. Tesla also tends to sell at a premium over Quadro, so why would GM200 begin life as a Quadro product if it weren't for a lack of DP or other compute features?

While the first GK110s introduced were Tesla cards the first FULL GK110 was launched as a Quadro.

http://www.geeks3d.com/20130724/nvidia-quadro-k6000-first-graphics-card-with-full-gk110-enabled/

 

[HurleyBird]

While the first GK110s introduced were Tesla cards the first FULL GK110 was launched as a Quadro.

http://www.geeks3d.com/20130724/nvidia-quadro-k6000-first-graphics-card-with-full-gk110-enabled/

You're cherry picking the one point that fits your argument, but it's not very relevant. The K6000 was released at that point in time to line up with SIGGRAPH for PR reasons, and the fully enabled K40 followed right after.

If GM200 has better DP than GK210 Nvidia should release it as a Tesla product first to maximize their profits.

 

[Red Hawk]

I cant see any relation to AMD. nVidia wants to sell new cards. Waiting for whatever AMD does doesnt sell new cards. If anything, AMD will have to speed up. The leader doesnt wait for the runner up.

Regarding this, Nvidia did seem to keep the fully enabled GK110 in the tank for gaming until AMD released a competitor with Hawaii. Nvidia has no pressing need on the current market to replace their high-end when they're already ahead of AMD by a significant margin with the 980 over the 290X.

 

[KaRLiToS]

No competition = high prices

Me =

 

[ShintaiDK]

Regarding this, Nvidia did seem to keep the fully enabled GK110 in the tank for gaming until AMD released a competitor with Hawaii. Nvidia has no pressing need on the current market to replace their high-end when they're already ahead of AMD by a significant margin with the 980 over the 290X.

Are you saying nVidia isnt interested in selling cards ? the notion that companies delay products on purpose exist only in the minds of people.

 

[CP5670]

This card looks like a beast, but that means it will most likely be released as a $1000 Titan. I would not expect much of a price drop in the 980 due to this, maybe down to $450-500. It will be interesting to see how the 390 compares to this.

 

[Red Hawk]

Are you saying nVidia isnt interested in selling cards ? the notion that companies delay products on purpose exist only in the minds of people.

They're already selling cards. The question is if releasing GM200 is really necessary to sell more cards than GM204 right now and if it would reduce sales later on. Good companies don't just think of the here and now.

 

[Arachnotronic]

They're already selling cards. The question is if releasing GM200 is really necessary to sell more cards than GM204 right now and if it would reduce sales later on. Good companies don't just think of the here and now.

Price GM200 as a premium option. You don't think they'd sell GM200 at the same price as the GM204, do you?

 

[Silverforce11]

Price GM200 as a premium option. You don't think they'd sell GM200 at the same price as the GM204, do you?

The 980 is already a premium option. Its priced much higher than the 970 than its performance would justify.

NV & partners are making a killing selling a mid-range chip & low complexity PCBs for what was once top-dog pricing. If they release GM200, it would have to be a Titan 2 at $1,000. Well beyond the market niche of 980 so its sales & price won't be affected negatively. It works, we had it already.

 

[ShintaiDK]

A 799$ GM200 and up and GTX980 would still sell like hotcakes as it does today. But you also get all those GTX780/Titan owners to upgrade.

 

[Grooveriding]

Once it launches in the pro line my guess is we see a geforce about 4 months after. I don't think we will see this in a geforce card until at least the spring. These chips sell for something like $3000-$5000 in pro cards, no reason not to sell the initial ones there and keep pumping out 980s There is nothing faster than the 980 right now.

I'm also thinking they may see less success the second time around with a Titan card. Some will hold off for the inevitable 980ti or whatever its called after seeing what happened last time with the 780 coming so quickly after the $1000 Titan.

 

[ocre]

Perhaps to some extent.

But it ultimately depends on what AMD can and will do. They can really make a huge difference here.

I have been thinking though..
If the gm200 is 150% a gtx 980, which it appears to be if the leaked Cuba cores are correct, then a full gm200 is looking to be a beastly card.

Is 150% a reasonable performance guess? A full gm200 at +50% over the gtx980?

That seems like a large jump. Is that the what people are expecting?

 

[3DVagabond]

Perhaps to some extent.

But it ultimately depends on what AMD can and will do. They can really make a huge difference here.

I have been thinking though..
If the gm200 is 150% a gtx 980, which it appears to be if the leaked Cuba cores are correct, then a full gm200 is looking to be a beastly card.

Is 150% a reasonable performance guess? A full gm200 at +50% over the gtx980?

That seems like a large jump. Is that the what people are expecting?

I think the 1st part of your post would be the reason that nVidia would be hesitant to launch GM200 if they have it ready now. The only problem nVidia could face would be launching GM200 @ $750 (or $1000 as a so called Titan II) so it doesn't cannibalize GM104 sales and have AMD release the 390X also @ $550 and have it be faster. Maybe appreciably so. nVidia has suffered this "black eye" before and I really think there is a saturation point with the public where they'll get put off by it and just think nVidia is really putting it on their appreciation for their product. The safe move is to continue selling 970's and 980's at a considerable profit and wait to see what AMD has up their sleeve.

 

[stahlhart]

Thread cleaned.

The next time I see someone here derail an otherwise productive discussion to troll for a flamewar, that individual is going to have all sorts of time to appreciate the superior gaming performance of their beloved company's GPU, because their posting privileges in this forum aren't going to be working any longer.

Figure out how to compare the products of competing companies without turning it into a personal confrontation, and do so quickly, because I'm starting to lose my patience with several of you.

-- stahlhart

 

[ocre]

I think the 1st part of your post would be the reason that nVidia would be hesitant to launch GM200 if they have it ready now. The only problem nVidia could face would be launching GM200 @ $750 (or $1000 as a so called Titan II) so it doesn't cannibalize GM104 sales and have AMD release the 390X also @ $550 and have it be faster. Maybe appreciably so. nVidia has suffered this "black eye" before and I really think there is a saturation point with the public where they'll get put off by it and just think nVidia is really putting it on their appreciation for their product. The safe move is to continue selling 970's and 980's at a considerable profit and wait to see what AMD has up their sleeve.

I really hope things get heated. Honestly, we need the 390x to be really great. I want it to be exciting.

I got a gtx970 and my plan is to go SLI if i need to. I rarely buy the top tier but i still love to watch PC technology as it unfolds.

My post above was the first time i crunched the gm200 numbers. I usually do not pay attention to rumors at all but i do read forums. I have been seeing people predicting the 390x will be faster than the gm200. I think it stems from the terrible leak that had came out a little while ago. I call it terrible because it had too many unreleased cards on it. No one would have access to all those cards much less drivers that work on them all. Not back then. I really think it is a total fake.

But regardless, people have taken to believe the 390x will surpass the gm200. But, looking at the specs. The supposed specs, the gm200 is 150% the 980. As i said, it is the first time i crunched it. But, if these specs are correct, and the leaked 4000 sp for the 390x is correct........

The gm200 should come out on top. And i wasnt really thinking that it added up like that. We are talking 150% a 980 vs 145% a 290x.

Thats not taken in account clock speeds or HBM.....anything else really.

I think things can actually be really really close this round. At least i hope they are. I just assumed the gm200 was at a disadvantage spec wise. Maybe it was the way 4096sp looked, i mean thats a huge number. But crunching the figures, the gm200 might have a much better chance with its 3072 cuda cores.

idk, for all we know these cards can be months and months out.