NVIDIA GeForce GTX 980 Ti Silicon Marked "GM200-310"

[btarunr]

NVIDIA's upcoming high-end single-GPU graphics card, based on the GM200 silicon, which debuted with the GTX TITAN X, will feature a silicon marked "GM200-310." The SKU will be named GeForce GTX 980 Ti, and is more likely to be priced around the $600-650 mark, than replacing the $550 GTX 980 off the shelves. Going by the way NVIDIA re-positioned the GTX 780 to $499 with the introduction of the GTX 780 Ti, we imagine something similar could happen to the GTX 980. From what we gathered so far, the GTX 980 Ti will be based on the GM200 silicon. Its CUDA core count is unknown, but it wouldn't surprise us if it's unchanged from the GTX TITAN X. Its different SKU numbering shouldn't be an indication of its CUDA core count. GTX 780 Ti and GTX TITAN Black had different numbering, but the same CUDA core counts of 2,880.

The card will feature 6 GB of GDDR5 memory across the chip's 384-bit wide memory interface. It will feature five display outputs, similar to that of the GTX 980. Unlike with the GTX TITAN X, NVIDIA partners will have the freedom to launch custom-design GTX 980 Ti products from day-one. There are two theories doing rounds on when NVIDIA plans to launch this card. One suggests that it could launch in mere weeks from now, probably even on the sidelines of Computex. The other suggests that it will launch towards the end of Summer, as NVIDIA wants to make the most cash from its existing GTX 980 inventory.

View at TechPowerUp Main Site

 

[Mathragh]

So if this is also a full chip, what will they do with the many defective chips that will very likely be there as a result of the massive die area? Make yet another 9xx big-maxwell based card with a part of the cuda cores disabled? (Or part of the L2/memory interface for 5,25GB of effective fast RAM hehehehehehe)

 

[Lionheart]

So if this is also a full chip, what will they do with the many defective chips that will very likely be there as a result of the massive die area? Make yet another 9xx big-maxwell based card with a part of the cuda cores disabled? (Or part of the L2/memory interface for 5,25GB of effective fast RAM hehehehehehe)

GTX 970 Ti

 

[wachuwey]

GTX 970 Ti

That's what I think, but, a 970Ti with a better gpu than the 980?? At what price?

 

[Luka KLLP]

I was planning to upgrade to a system with a GTX980 in late June/early July... I hope this card isn't launched too late after that, since I'll have to wait longer (a price drop on the GTX980 seems likely)

 

[Parn]

Well even though 780Ti and Titan Black had the same CUDA core count, Titan Black had the full double precision performance of the GK110 whereas 780Ti didn't. That explains the different numbering.

Now GM200 lacks strong DP performance by default, so the lower numbering on 980Ti might actually indicate it is not a fully fledged GM200 compared to Titan X.

 

[rodneyhchef]

I would expect ~2560 cores

As the previous poster said, there were more factors that separated the 780 range from the original titan/titan black.

Also, The Titan X hasn't currently got a lesser binned counterpart: The orginal Titan/Black had the vanilla 780, with significantly reduced core count

I hope this means the 980 price will be lowered, it is far too expensive as it is compared to the 970

 

[64K]

Well even though 780Ti and Titan Black had the same CUDA core count, Titan Black had the full double precision performance of the GK110 whereas 780Ti didn't. That explains the different numbering.

Now GM200 lacks strong DP performance by default, so the lower numbering on 980Ti might actually indicate it is not a fully fledged GM200 compared to Titan X.

That was my first thought too when I saw that on VideoCardz yesterday. The expected launch of Pascal is sometime around the end of this year or early next year and it could get delayed by another TSMC fail. If we say Pascal at the end of this year then that's 8 months away. Plenty of time to release a gimped GM200 and call it GTX 980 Ti for $650 and then later release the full GM200 for ~$700. That's what they did with the GK110. They may have had enough partially defective GM200 chips to do the same as they did with the Kepler GTX 780.

Nvidia could continue to milk the GTX 980 for a while longer and get rid of their partially defective GM200s too. We'll see.

 

[Parn]

That was my first thought too when I saw that on VideoCardz yesterday. The expected launch of Pascal is sometime around the end of this year or early next year and it could get delayed by another TSMC fail. If we say Pascal at the end of this year then that's 8 months away. Plenty of time to release a gimped GM200 and call it GTX 980 Ti for $650 and then later release the full GM200 for ~$700. That's what they did with the GK110. They may have had enough partially defective GM200 chips to do the same as they did with the Kepler GTX 780.

Nvidia could continue to milk the GTX 980 for a while longer and get rid of their partially defective GM200s too. We'll see.

I heard NV had been approaching Samsung for some additional fab capacity in order to prevent another TSMC supply failure. So Pascal may well be on time.

Now if 980Ti does turn out to be a gimped version of GM200, it makes me think that 390X might not be as powerful as we have hoped (speculations are 390X could be faster than Titan X while costing much less). That's why NV have no problem releasing a cut-down GM200 while still remaining competitive.

 

[GhostRyder]

I don't know, this is quite an oddity this round in terms of GPU's mostly because they are almost name locked in what they can call the GPU since they decided not to have the Titan X and other full fledged GM200 cores (Or cut down) on the next numbering scheme (IE 1080, 1080ti etc).

To be honest, ill probably be shocked if they release the GTX 980ti with the full amount of cores enabled. I believe because of how the Titan X is this round they do not want to immediately make a card that performs as well or better that fast and butcher its sales. My guess is that they will probably release the GTX 980ti as a cut down model this round from the Titan X with higher core clocks (Maybe even better overclockability) and that will be the first before they release another card after the R9 390X hits the market (probably also waiting to see the competition). My guess is that later they will release a card with the full cores unlocked called something like the GTX 980ti Boost that will be overclocked and ready for higher clocking later down the line. But that is all speculation based on the little bit of information we have.

 

[alwayssts]

I don't know, this is quite an oddity this round in terms of GPU's mostly because they are almost name locked in what they can call the GPU since they decided not to have the Titan X and other full fledged GM200 cores (Or cut down) on the next numbering scheme (IE 1080, 1080ti etc).

To be honest, ill probably be shocked if they release the GTX 980ti with the full amount of cores enabled. I believe because of how the Titan X is this round they do not want to immediately make a card that performs as well or better that fast and butcher its sales. My guess is that they will probably release the GTX 980ti as a cut down model this round from the Titan X with higher core clocks (Maybe even better overclockability) and that will be the first before they release another card after the R9 390X hits the market (probably also waiting to see the competition). My guess is that later they will release a card with the full cores unlocked called something like the GTX 980ti Boost that will be overclocked and ready for higher clocking later down the line. But that is all speculation based on the little bit of information we have.

Sure, it's speculation, but I think that totally makes sense in terms of what people could/should expect.

What you're saying, in essence and I agree, is that nvidia needs to compete on performance and value within some kind of closer margin, with most pertinent markets first.

By all indications 390x will be similar to Titan-X, but cheaper. I still think the larger threat is a 390 out-performing 980 somewhere around $500 though, not unlike what was the original plan for the 290 (compete with 770; the change to 780 was a literally last-second decision). nVIDIA needs a much cheaper part that is faster than 980. AMD could price 390 so that 980 has nowhere to go in terms of cost vs asp (margin). It could end up being a faster card at a (relatively) better price. I can't see 980 dropping further than ~$450 right now, nor could I imagine 390 more than $550, which would seem relatively proportionate, in theory.

The smart play imho is for nvidia to release a part at stock that beats 390, and can overclock to a similar level as a stock 390x for less money. Bonus points (meaning higher msrp) if it overclocks better than 390.

Just as a guesstimate, but figure a 390's core is clocked similar to the avg performance of a 980, say 3584 shaders at ~925mhz (versus [2048+512]*1291.5), but bandwidth brings performance closer to it actually performing at 1033mhz. The smart call is for nvidia to release a part that performs slightly better compared to a 1033/3584sp AMD part, and can overclock closer to 4096 at (normalized) 1106mhz.

What you end up with is something like 21SMM at 1100mhz that can clock to 1350mhz (perhaps slightly higher on both counts for breathing room, or lower if they count slight performance from extra bandwidth which would be a few percent). That is just one example, and it could be configured differently, but you get my point; the smart play for nvidia is a cheap (for them) part that can act more-or-less as a practical magic bullet against Fiji as a whole. Cheaper than 390x, faster than 390 at stock, similarish to 390x stock when overclocked. I agree a full-fat 6GB card probably will come later when all the sheen has thoroughly been rubbed (sales drop) from Titan X...but this should be their priority. If there is a spot for such a card, it probably won't be known until we see how 390x 4/8GB perform, and how the market accepts them. It probably wouldn't hurt nvidia to have a little more time to bin high-clocking full chips, either.

We rarely know how these things are actually binned, but 310/400 (77.5%) could be something like 87.5% binned at 13% higher voltage or lower clocks per voltage, which essentially pans out to exactly what I postulated above.

TLDR: GM200 needs a '970' vs Fiji.

 

[net2007]

6gb is nice, but I think 8 would be better. AMD will launch with 8.

 

[arbiter]

Now if 980Ti does turn out to be a gimped version of GM200, it makes me think that 390X might not be as powerful as we have hoped (speculations are 390X could be faster than Titan X while costing much less). That's why NV have no problem releasing a cut-down GM200 while still remaining competitive.

All the speculation of 390x performance has been hype by amd fans. If you go by AMD's own slides its performance is around same performance of titan X. But problem for it will be price since supply of HBM memory will make it a high one.

6gb is nice, but I think 8 would be better. AMD will launch with 8.

There will be a 4gb and 8gb version of card. that 8gb version likely carry a very high premium due to very limited supply of HBM chip's. if 4gb starts at 700$ which is rumor for it, the 8gb card could very well be 1000$ if you can get retail card that isn't jacked up in price cause demand. the GDDR5 290x 4/8gb cards are 120$ premium. so 2-300$ would be easy for limited supply card.

Keywords from the AMD slide, "Up to 8gb"
http://www.techpowerup.com/img/15-03-16/78a.jpg

 

[Casecutter]

I don't know, this is quite an oddity this round in terms of GPU's mostly because they are almost name locked in what they can call the GPU since they decided not to have the Titan X and other full fledged GM200 cores (Or cut down) on the next numbering scheme (IE 1080, 1080ti etc).

To be honest, ill probably be shocked if they release the GTX 980ti with the full amount of cores enabled. I believe because of how the Titan X is this round they do not want to immediately make a card that performs as well or better that fast and butcher its sales. My guess is that they will probably release the GTX 980ti as a cut down model this round from the Titan X with higher core clocks (Maybe even better overclockability) and that will be the first before they release another card after the R9 390X hits the market (probably also waiting to see the competition). My guess is that later they will release a card with the full cores unlocked called something like the GTX 980ti Boost that will be overclocked and ready for higher clocking later down the line. But that is all speculation based on the little bit of information we have.

I'd say the full-fledge GM200 generous clocks, no memory allocation they bring back to life the "ULTRA" moniker as in a "980Ti Ultra 6Gb" and priced at $850. Then 980Ti with a missing shader or so, but no memory cut-back $700. Finally they resurrect the "SE" designation as " GTX980 SE" with gelded shaders and other L2/ROP missing perhap even with 4Gb at $550, and reduce the GTX 980 to $475.

 

[Captain_Tom]

I heard NV had been approaching Samsung for some additional fab capacity in order to prevent another TSMC supply failure. So Pascal may well be on time.

Now if 980Ti does turn out to be a gimped version of GM200, it makes me think that 390X might not be as powerful as we have hoped (speculations are 390X could be faster than Titan X while costing much less). That's why NV have no problem releasing a cut-down GM200 while still remaining competitive.

Yeah because it is crazy talk to think that AMD will be releasing something stronger than anything Nvidia has. Except that is nothing new...

 

[GhostRyder]

I'd say the full-fledge GM200 generous clocks, no memory allocation they bring back to life the "ULTRA" moniker as in a "980Ti Ultra 6Gb" and priced at $850. Then 980Ti with a missing shader or so, but no memory cut-back $700. Finally they resurrect the "SE" designation as " GTX980 SE" with gelded shaders and other L2/ROP missing perhap even with 4Gb at $550, and reduce the GTX 980 to $475.

Good point, there are a lot of monikers they can use for these cards. Question will be what will they use and how will the designate differences if this is the route they choose.

Sure, it's speculation, but I think that totally makes sense in terms of what people could/should expect.

What you're saying, in essence and I agree, is that nvidia needs to compete on performance and value within some kind of closer margin, with most pertinent markets first.

By all indications 390x will be similar to Titan-X, but cheaper. I still think the larger threat is a 390 out-performing 980 somewhere around $500 though, not unlike what was the original plan for the 290 (compete with 770; the change to 780 was a literally last-second decision). nVIDIA needs a much cheaper part that is faster than 980. AMD will price 390 so that 980 has nowhere to go in terms of cost vs asp (margin). It should end up being a faster card at a (relatively) better price.

The smart play imho is for nvidia to release a part at stock that beats 390, and can overclock to a similar level as a stock 390x for less money. Bonus points (meaning higher msrp) if it overclocks better than 390.

Just as a guesstimate, but figure a 390's core is clocked similar to the avg performance of a 980, say 3584 shaders at ~925mhz (versus [2048+512]*1291.5), but bandwidth brings performance closer to it actually performing at 1033mhz. The smart call is for nvidia to release a part that performs slightly better compared to a 1033/3584sp AMD part, and can overclock closer to 4096 at (normalized) 1106mhz.

What you end up with is something like 21SMM at 1100mhz that can clock to 1350mhz. That is just one example, and it could be configured differently, but you get my point; the smart play for nvidia is a cheap (for them) part that can act more-or-less as a practical magic bullet against Fiji as a whole. Cheaper than 390x, faster than 390 at stock, similarish to 390x stock when overclocked. I agree a full-fat 6GB card probably will come later when all the sheen has thoroughly been rubbed (sales drop) from Titan X...but this should be their priority. If there is a spot for such a card, it probably won't be known until we see how 390x 4/8GB perform, and how the market accepts them. It probably wouldn't hurt nvidia to have a little more time to bin high-clocking full chips, either.

We rarely know how these things are actually binned, but 310/400 (77.5%) could be something like 87.5% binned at 13% higher voltage or lower clocks per voltage.

TLDR: GM200 needs a '970' vs Fiji.

I agree, I think this round has already been a bit interesting as far as chips go. They have to release some sort of lower core chip because I doubt they would just write that off (I mean they could but that would not be the best decision) so were going to have to see where and what they put out. With the speculation/rumored performance of the 390X (Especially is leaks are to be believed) it could be very competitive in the market (Especially in high resolution scenarios).

All the speculation of 390x performance has been hype by amd fans. If you go by AMD's own slides its performance is around same performance of titan X. But problem for it will be price since supply of HBM memory will make it a high one.

There will be a 4gb and 8gb version of card. that 8gb version likely carry a very high premium due to very limited supply of HBM chip's. if 4gb starts at 700$ which is rumor for it, the 8gb card could very well be 1000$ if you can get retail card that isn't jacked up in price cause demand. the GDDR5 290x 4/8gb cards are 120$ premium. so 2-300$ would be easy for limited supply card.

Keywords from the AMD slide, "Up to 8gb"
http://www.techpowerup.com/img/15-03-16/78a.jpg

It was hyped do to supposed leaks not just people screaming stuff like "Its going to beat Titan X". That all still speculation so we can only choose to believe the leaks or not at this point and go based on what specs are out there. Its all going to come down to base performance and how much of a difference at 1440p+ the HBM memory makes.

GTX 980ti will probably be a slightly cut down chip off the line, won't be a bad thing but that is depending on where its priced and how well it overclocks because I expect its going to clock significantly better than the GTX Titan X.

 

[Captain_Tom]

I like how everyone says that he 980 Ti will be "Cut down without the memory problem." Everyone here does realize that the architecture of Maxwell makes that inherently impossible right?

The memory and cores are directly linked. Disable any of the cores and a proportional amount of the memory will need to be disabled as well - which besides being wasteful, isn't that bad as long as they don't lie about it.

 

[the54thvoid]

Leaked to seed doubt prior to a 390x announcement? Make people think that, if 390x is faster than Titan X (looking less likely to me, alas) then people hold off?
I dunno. Any further gm200 full chip release requires higher clocks to dislodge TitanX for what? A 5-10% perf boost?
If this is below Titan X performance, it looks like they know AMD's upcoming card isn't as strong as expected.
Remember the GTX680 riposte to the HD7970? Maybe Nvidia know something....

 

[HumanSmoke]

Leaked to seed doubt prior to a 390x announcement? Make people think that, if 390x is faster than Titan X (looking less likely to me, alas) then people hold off?

If that were the case, wouldn't a more valid strategy be to point out that Fiji's launch date might be on the slide (pun intended)? The launch 390X rumours have been circulating for months - reaching a height coincidentally at the time of Titan-X's own launch, yet the Fiji and HBM official unveiling is now slated for August 25th, at the Hot Chips Symposium

I dunno. Any further gm200 full chip release requires higher clocks to dislodge TitanX for what? A 5-10% perf boost?

Probably, although I think you'll see the usual suspects from Asus, Gigabyte, EVGA, MSI, and GALAX pushing the envelope out a little further....at a price.

If this is below Titan X performance, it looks like they know AMD's upcoming card isn't as strong as expected.

Or two bites of the cherry. If Nvidia know that the 390X is still some way off, they have an opportunity to cash in on those people wanting to upgrade but unwilling to pay Titan-X prices. Most people in the enthusiast bracket would welcome both a price cut, and more importantly, variation from the stock reference design. If Nvidia launch a 980Ti at $649-699, it could persuade a significant number of those waiting to stump up for the card. When the 390X arrives, it's potential customer base will be less than it would have been even if it equals/shades the 980Ti, and those that bought the 980Ti wouldn't likely upgrade a second time in short order even if the 390X equals or shades the 980Ti- who upgrades for 5-10% performance improvement? In this case, the 390X could very well be the definition of a pyrrhic victory, benchmark success at the cost of a high bill of materials and a much reduced potential customer base.
From my viewpoint, GM200 and Fiji, look increasingly like stepping stones and stopgap measures before the real fun starts with Arctic Islands and Pascal.

Remember the GTX680 riposte to the HD7970? Maybe Nvidia know something....

If that is indeed the case, the AMD are in serious trouble I think. If the rest of the 300 series are warmed over silicon from previous generations, and the halo GPU cannot deliver an unconditional performance advantage over GM200, I don't see the company clawing back market share. They really need a shining star at the top of product stack allied with a very solid midrange GPU that can grab OEM sales off GM204's desktop and mobile markets.

 

[arbiter]

If that is indeed the case, the AMD are in serious trouble I think. If the rest of the 300 series are warmed over silicon from previous generations, and the halo GPU cannot deliver an unconditional performance advantage over GM200, I don't see the company clawing back market share. They really need a shining star at the top of product stack allied with a very solid midrange GPU that can grab OEM sales off GM204's desktop and mobile markets.

they need to keep what they got for now and maybe gain a little. But not sure that is gonna happen.

 

[TheinsanegamerN]

they need to keep what they got for now and maybe gain a little. But not sure that is gonna happen.

They need to gain a lot more than just "a little". what they have now is sinking the company.

 

[johnspack]

I've said it before, when will this thing drop? We need much cheaper 970s soon.....

 

[INSTG8R]

If that were the case, wouldn't a more valid strategy be to point out that Fiji's launch date might be on the slide (pun intended)? The launch 390X rumours have been circulating for months - reaching a height coincidentally at the time of Titan-X's own launch, yet the Fiji and HBM official unveiling is now slated for August 25th, at the Hot Chips Symposium

While I always appreciate Smokes data. I think your August date is gonna be a bit late for the fact of this:

http://www.pcgamingshow.com/

AMD are the big sponsor for this and I'm pretty sure this is where they are gonna drop the new cards

 

[HumanSmoke]

While I always appreciate Smokes data. I think your August date is gonna be a bit late for the fact of this:
http://www.pcgamingshow.com/
AMD are the big sponsor for this and I'm pretty sure this is where they are gonna drop the new cards

Well, I hope you're right. The Hot Chips Symposium could well be a post-launch whitepaper dissection, although the 16th June is a little early if you subscribe to the exactness of supposed insider knowledge. If the 16th is the drop date then we should be seeing reliable leaks (as opposed to anonymous unverified benchmark charts) fairly soon.

 

[INSTG8R]

Well I expect at least a "reveal" I mean why host a "PC Gaming Show" If your not gonna bring some new toys to play with. They certainly can't be promoting the 290X...