Normally, I pose these sorts of rumors as “Well, here you go, and here's a grain of salt.” This one I'm fairly sure is bogus, at least to some extent. I could be wrong, but especially the GP100 aspects of it just doesn't make sense.
Before I get to that, the rumor is that NVIDIA will announce a GeForce GTX Titan P at Gamescom in Germany. The event occurs mid-August (17th – 21st) and it has been basically Europe's E3 in terms of gaming announcements. It also overlaps with Europe's Game Developers Conference (GDC), which occurs in March for us. The rumor says that it will use GP100 (!?!) with either 12GB of VRAM, 16GB of VRAM, or two variants as we've seen with the Tesla P100 accelerator.
The rumor also acknowledges the previously rumored GP102 die, claims that it will be for the GTX 1080 Ti, and suggests that it will have up to 3840 CUDA cores. This is the same number of CUDA cores as the GP100, which is where I get confused. This would mean that NVIDIA made a special die, which other rumors claim is ~450mm2, for just the GeForce GTX 1080 Ti.
I mean, it's possible that NVIDIA would split the GTX 1080 Ti and the next Titan by similar gaming performance, just with better half- and double-precision performance and faster memory for GPGPU developers. That would be a very weird to me, though, developing two different GPU dies for the consumer market with probably the same gaming performance.
And they would be announcing the Titan P first???
The harder to yield one???
When the Tesla version isn't even expected until Q4???
I can see it happening, but I seriously doubt it. Something may be announced, but I'd have to believe it will be at least slightly different from the rumors that we are hearing now.
It may have the same number
It may have the same number of Cuda cores as the GP100, but a lot of those cores will be defective and have been fused off and some will be purposefully disabled to arrive at a targeted amount of Cuda cores for the TI’s performance/price range. Even the Tesla P100 is a binned part with that GPU hardware over provisioning that all the GPU makers use to get the die/wafer yields up to manageable amounts.
I got a feeling this is gonna
I got a feeling this is gonna cost 1000 USD
That would make sense for a
That would make sense for a Titan product… but, if this is true, it might be more.
Good idea to not even
Good idea to not even consider buying a gtx 1080 until we see the announcement. We know nvidia is selling their mid range cards as the 1070 and 1080 right now. Just like they did with the 680/670 and 980/970.
There is a big Titan/ti version coming for Pascal and it will destroy the 1080s performance, making it look like a joke of card.
With this possible announcement the 1080 is a definite no buy until the new card is unveiled.
>.> My GTX 1080 is currently
>.> My GTX 1080 is currently on courier. That said, it was at the top end of my budget anyway, and I play 1080p, albeit high frame rate, so I guess it's fine.
1080p? That is a sad
1080p? That is a sad resolution for a GTX 1080.I moved from an 290X to a Fury X at 1080p and games like Dead Island The Definitve Edition, Hard Reset etc hits a ceiling around 120-150fps, with minimal gains over the 290X, once I use VSR or Super Sampling, then the difference is notable. ANy GPU faster than the 290 is overkill and that comes from a guy with an i7 3770K OCED to 4.50GHz, so no CPU bottleneck here. Games are more CPU bound than GPU bound and the DX11 API does not help either, plus quite a lot of games aren’t taxing enough to take advantage of the additional execution resources. Even a GTX 980 would give you a similar gaming experience at such lame resolution.
Heh. 1080p gaming is not the
Heh. 1080p gaming is not the only thing I do with the card. I do GPGPU, which would be better with AMD but what can you do, and (the main reason for a GeForce card) rendering work with Cycles in Blender.
Also, it's 1080p with 3DVision, and I might pick up a VR headset in the future.
I can think of 4 reason why
I can think of 4 reason why Nvidia would have a GP102 die.
1) Don’t want to sell a GP100 die with all that Double Percision performance in that price bracket when they can put out a Titan product.
2) They don’t want to use HBM2 for a consumer part
3) They want to trim off all that die space for extra maximum cores and/or a higher frequency.
4) Cost savings. Going from 551mm2 to ~450mm2 is nothing to sneeze at. And just on the potential of better yields would not be a bad plan.
They would announce the Titan P first because a) they already have GP100 dies to use for it and b) Titans are super low volume because of the price. We are talking about $1000+ parts while the GTX1080Ti would re-adjust their product stack downward and start at $750-800. Something they do not even need to happen as Vega looks to be a like GCN4.0 would place it at the same performance as a standard 1080.
you seriously think this is
you seriously think this is going to be $1000 its 1500-3000 due to the supply of hmb2
The GPU pricing has gone to
The GPU pricing has gone to hell this year. Aib partners have forgotten that the GTX 1080 is the direct replacement for the 980, yet we are seeing 1080 models go for about the same price as the 980 ti when it launched.
I agree, it doesn’t add up.
I agree, it doesn’t add up. If the Ti was based off of an easier to make die, it would make sense to get them to market first. But I have a hard time believing they would release the 1080Ti this close to the 1080.
Maybe the Titan will come first but has something crazy we haven’t thought of like 32GB of HBM?
HBM can only go up to 4GB.
HBM can only go up to 4GB. You mean HBM2
The timing seems a little
The timing seems a little off. I expected to see big Pascal Titan in Q1 2017 and big Pascal 1080Ti some 3 or 4 months later.
I’m due to buy a 1080 sometime soon. I’m in no rush, so I might just hold off and see what develops.
With 1080 prices a little
With 1080 prices a little higher than they should, the only way this card could help Nvidia is if it costs at $1500-2000 minimum. In that case it would help in the marketing department, making charts look funny with a really long green bar on top of them, and also move a few more 1080’s because there are plenty out there waiting for the Ti model.
But if they see that the Titan P comes at $1500+ for the 12GB version, that would probably mean over $1000 for the Ti model, $1200 for example. So, people might be persuaded to buy a 1080 today, even if it sells at higher price than the MSRP and stop hoping for a Ti at a reasonable, under $1000, price.
I don’t mind the price. I
I don’t mind the price. I want the best of the best and I know many enthusiasts in the same boat as me.
Exactly
I want the best of
Exactly
I want the best of the best and I don’t understand why anyone looking at the highest end cards would quibble over an extra 3 or 4 hundred bucks.
Look, I know people are often on a tight budget but these ARE the top end we are talking about. I’m not rich but I don’t spend $500 buck a month on booze and cigarettes either.
Just get me the best and fastest is all I want…price be damned…
The rumour is accompanied
The rumour is accompanied with the idea of an extra-long card, and of end-situated PEG connectors (rather than on top as is normal for consumer cards). As the Tesla P100 PCIe version is not an extra long card (10.5″, same as the K40) and end-mount connectors are only used for the Tesla/Quadro cards intended to be used in 3U height0-limited chassis, methinks this rumour is the result of someone getting confused.
Doesnt seem entirely far
Doesnt seem entirely far fetched to me.
I think people who bought the 900 series titan were a little pissed off when the 980Ti released and had near identical performance.
They may just be trying to avoid that situation happening again by using different silicon for the titan and having a bigger performance gap to justify the huge price tag.
Doesn’t seem too odd to me.
Doesn’t seem too odd to me. Nvidia hasn’t exactly stuck to any pattern with the titan and ti cards. For example, the titan black had the same cuda core count as the 780ti so that’s pretty much in line with the 1080ti having the same count as the titan p.
People made similar claims
People made similar claims regarding the 980 ti based on previous “patterns” in nvidia’s now obviously random naming scheme. A lot of people claimed that since the 980 was a fully unlocked chip and the titan x was a fully unlocked chip and up until then “ti” denoted cards with fully unlocked chips the 980ti couldn’t be a thing since there was no chip between gm104 and gm110. At the end of the day they aren’t gonna thrive as a company by following the same pattern every year, they need to dynamically change their strategies to compete in the marketplace. This does make me think that the rumour is fake though, since they wouldn’t release another high end card too soon since they’d just be competing with themselves and cannibalising their profits. I’d guess a q4 2016 release or maybe q1 2016.
Actually this
Actually this August-September release of the GP102 makes PERFECT sense. As of now Nvidia is enjoying TOTAL monopoly of the high end market. AMD has literally NOTHING to compete with their GTX 1080 AND the GTX 1070. So they can lay back and sell 1080/1070 at high price all they want, and release the 1060 at their leisure to compete with the gimped and mediocre imo RX 480.
BUT, once October rolls around AMD is releasing VEGA, which will have up to 4,096 stream processors with the same 14nm and GCN 3.0 design on the Polaris cards. This means that Vega will roughly be equivalent to a GTX 1080 or so, give or take a few percent depending on the memory capacity/bandwidth and a few other things like clock speed etc..etc..
So Nvidia HAS to do something if they want to keep total market dominance and make AMD look foolish by always being beat to the punch with top end cards. So…they simply release a GP102 “1080 TI” or TITAN card with ~3,584 – 3,840 cuda cores and 384 bit G5X (480gb/s) or maybe even HBM2 on 3072 bit bus (768gb/s) giving them a VERY distinct advantage over AMD in the enthusiast high end market. And the fact that they are releasing on paper in AUGUST, and likely having the cards fully available in SEPTEMBER, means that they will beat AMD’s Vega release by ONE MONTH! You think that’s a coincidence?….i guarantee you it isn’t. They’re doing it on purpose to be the first out with a high end card.
Actually it DOESN’T have the
Actually it DOESN’T have the same amount of cuda cores. The GP100 has MORE than 3,840 cuda cores; its just that nobody realized that it has 1,920 DOUBLE PRECISION dedicated cuda cores on top of the 3,840 single precision dedicated cores. This means that GP100 has a total of 5,760 cuda cores.
This is how GP102 can have up to 3,840 cuda cores. HOWEVER, i’m guessing that GP102 will actually have 3,584 cuda cores, which is a 40% increase over the 2,560 on the 1080, as if you take the 314mm2 die size of GP104 and multiply by 1.4x you get ~440mm2, which is just shy of the ~470mm2 estimate most have been saying for GP102; and i assume you always need a little “extra” die space on top of that added for more cores (i.e. if you have 40% more cores you probably need ~45-50% more die space to accomodate those cores and still have room for other necessary additions to the die etc..)
Whereas the full GP100 with it’s ~600mm2 die size can have up to 5,760 cuda cores; i imagine if they go the TITAN X route and forego double precision entirely that they will simply make GP100 have something along the lines of ~4,480 cuda cores which is less than the massive 5,760 on the Tesla P100 GP100 card which means that yields won’t be abysmal since they’re still new to 16nm; but at the same time is still a sizeable 35% increase in cores over GP102, and a whopping 75% increase in cores over GP104. Coincidentally, if you take the ~470mm2 estimated die size for Gp102 and multiply THAT by 1.35x (35% added) you get to precisely 600mm2. So the math PERFECTLY adds up.
My estimates are this:
May 2016 = GP104:
GTX 1080: 2,560 cores – 8GB G5X @ 256 bit – 320gb/s bandwidth
Late 2016 = GP102:
GTX 1080 TI – 3,584 cores – 12GB G5X @ 384 bit – 480gb/s bandwidth
Mid 2017 = GP100:
GTX 1180: 3,840 cores – ~12GB HBM2 @ 3072 bit – 768gb/s bandwidth
GTX 1180 TI: 4,480 cores – ~12GB HBM2 @ 4,096 bit – 1024gb/s bandwidth (1tb/s)
> This means that GP100 has a
> This means that GP100 has a total of 5,760 cuda cores.
Nope – that’s wrong 😉
Why on earth would they
Why on earth would they release a GTX 1080 Ti on GP100? That would be a bad marketing move and we all know that NVIDIA’s marketing is formidable. I’m guessing that a GP102 would be more efficient for gaming (Ti), whereas the Titan P released on the GP100 would be suitable to both games and HPC, albeit at a premium price.
We, enthusiasts, know what we are looking for in a card (gaming, HPC, streaming, etc.). I’m more interested in gaming.