The LinkedIn leak said 23-something.
Just the old rumour of it being 232mm2 - http://videocardz.com/58237/amds-project-f-is-232mm2-discrete-gpu-made-in-14lpp-process
That size fits well with AMD's recent historical midrange ie Pitcairn and Barts. If Samsung's 14LPP is around 9% denser than TSMC's 16FF+ and 232mm2 should be effectively around 250mm2 compared to the 294mm2 of GP104 (assuming GP104 is 294mm2...)
With what kind of transistor density exactly for the first? Because I'm slowly starting to admire the art of reading out of a single sterile number a ton of GPU characteristics.
True. But I have the moot feeling that technology has the nasty habit for having upper boundaries for possibilities too.
I would go on historical precedence where we know that AMD generally does more with less die space (with Maxwell being a spectacular exception).
Yes. For example AMD's mindshare is nowhere near Nvidia's, and history proves that people will wait even for something like Fermi no matter how good the competition was. It doesn't matter how much better Polaris is, Nvidia will outsell AMD anyway. They have a long-term plan for dealing with that, but I'm not going into it here.
D
Deleted member 2197
Guest
Duplicate post. Pls delete.
For the tech wizzards interested in the presentations at GTC 2016
After you have registered you will be able to access all video recordings.
After you have registered you will be able to access all video recordings.
I think perf/watt is going to be close and also performance close as well to the 1070 that will be the 294mm2 Polaris.
I would go on historical precedence where we know that AMD generally does more with less die space (with Maxwell being a spectacular exception).
Yes. For example AMD's mindshare is nowhere near Nvidia's, and history proves that people will wait even for something like Fermi no matter how good the competition was. It doesn't matter how much better Polaris is, Nvidia will outsell AMD anyway. They have a long-term plan for dealing with that, but I'm not going into it here.
More with less in die space, doesn't seem like that is going to happen this round at least no where near to the degree of 2x the ALU's to not see those ALU's being that effective in over all performance. All leaks to date seems like this round both GPU manufactures are targeting better utilization of the ALU units. Now going by historic relevancy, nV should have an upper hand when doing things like this, and also the ability to clock higher as they have shown time and time again, they are capable of doing this......
AMD if they have decent products that compete with nV will automatically gain marketing share in their given segments. The only reason they lost marketshare before was because they didn't have any cards out for 9 months.
Last edited:
A higher transistor density doesn't necessarily mean lower power consumption, being just one example.
Yes. For example AMD's mindshare is nowhere near Nvidia's, and history proves that people will wait even for something like Fermi no matter how good the competition was. It doesn't matter how much better Polaris is, Nvidia will outsell AMD anyway. They have a long-term plan for dealing with that, but I'm not going into it here.
The problem isn't to find a PR alike statement like the above, but to answer the question which was more like how you manage to come to N amount of conclusions with one sterile number that doesn't even have to do anything with reality. Apart from marketing drivel from AMD and an up to now meaningless specsheet of a HPC oriented monster chip from the green side I haven't seen anything substantial yet to draw any such conclusions from. Indications so far point that AMD has made a few probably significant architectural changes vs. relatively small ones for NV, which unfortunately renders at least for the first even more useless even if that 232mm2 figure would be real.
Last edited:
Well are you expecting something out of the ordinary? If so it must be from the AMD side as you clearly believe that Nvidia hasn't done a whole lot different from Maxwell, right?
So that leaves us with an AMD failure or an AMD success. If they've failed then we can expect Polaris 10 to be much slower than the 1070, fair enough? But how can you come to that conclusion based on what we've already seen from Polaris 11 compared to the 950 in the December demonstration? We've also seen Polaris 10 playing Hitman at higher framerate than Fury X. Do you expect the 1070 to be much faster than the 980 Ti?
So if we take all of this information I guess you must believe Polaris will be a runaway success instead? Therefore the 232mm2 Polaris 10 will easily sail past the cut 294mm2 GTX 1070? Isn't it a lot more likely that they'll be pretty close in performance?
Up to now Maxwell was ahead in terms of perf/W and AMD claims to have improved for the coming generation by up to 2.5x times. Without even a millimeter of viable data how much really things have changed on the red side and not a single guarantee yet that Maxwell and Pascal are on the exact same level without a single persentage improvement you expect a reasonable answer to your gut feeling? It's your gut feeling and nothing else.
Not really and I wan't even thinking or pointing in that direction. It's just plain and simple idiotic to repeat a meaningless number that doesn't mean anything without any further data.
It seems that that hypothetical 232 number has started some new mantra if not religion as if one chants 232 on 24/7 the world will spin faster....
Answer your own questions yourself, since you obviously don't seam to understand what others are trying to say.
Last edited:
Assuming GP104 goes light on fp64 (1/32 FP32) what would a 300mm^2 Pascal look like?
Random speculation cause I like round numbers :
4 GPCs
20 TPCs
40 SMs
2560 ALUs
@~1500Mhz
That should be good for 20% improvement on the 980 Ti.
Not seeing a lot of price movement on current parts. Possible $650 launch price?
Random speculation cause I like round numbers :
4 GPCs
20 TPCs
40 SMs
2560 ALUs
@~1500Mhz
That should be good for 20% improvement on the 980 Ti.
Not seeing a lot of price movement on current parts. Possible $650 launch price?
Considering sweoverlockers mentions 3 hypothetical replacements for 980Ti, 980 & 970 later on it sounds a bit weird to have 2 salvage parts almost from the beginning: https://translate.google.gr/transla...ttare-i-maj-gtx-970-avloses-i-juni&edit-text=
Could it be that there's something like the fabled "GP102" also at play and they've re-arranged completely how they address each market segment with different chips?
Could it be that there's something like the fabled "GP102" also at play and they've re-arranged completely how they address each market segment with different chips?
I'm thinking that Nvidia could strike with another mid-sized GPU dressed as a high-performance part, a.k.a. GTX680, if the power/clock wall of the 16FF process allows for significantly higher turbo boost (1600+ MHz). Also, the consumer Pascal has a lot of dead weight logic to loose from its HPC sibling (GP100).
A maximum of 1,920 FP64-ALUs (1,680 more realistically) and 4 NVLink Interfaces. Doing the math, it seems more likely that Gamer-Pascal will get 1/16th DPFP-rate, since maxwell's 1/32th would lead to only 2 DPFP-Units per SM. Other than that: They could've opted for pure GDDR5X interface, not designed to go to extreme clock speeds and thus a bit smaller than HBM2. IIRC, AMD claimed their HBM-MC on Fiji was a bit smaller actually than the GDDR5 controller in HawaiI which in turn was a bit smaller than Tahitit's higher clocked PHYs.
Now with HBM2 and it's increased data rate per pin, a piece of the area improvement will likely be gone for good again.
I'd fancy the idea of a dedicated GP102 for gaming, even though GP100 could have better thermal characteristics with all those powergated or fused off DPFP-units acting as thermal spacers.
Now with HBM2 and it's increased data rate per pin, a piece of the area improvement will likely be gone for good again.
I'd fancy the idea of a dedicated GP102 for gaming, even though GP100 could have better thermal characteristics with all those powergated or fused off DPFP-units acting as thermal spacers.
Last edited:
A similar rumour to Sweclockers - http://www.bitsandchips.it/52-engli...ill-have-2-pcbs-base-gddr5-and-premium-gddr5x
For them, both the 1080 and 1070 are cut-down with GDDR5 and later a full Ti version comes with GDDR5X. For the record I don't like the rumour, but I do believe they are a good source.
For them, both the 1080 and 1070 are cut-down with GDDR5 and later a full Ti version comes with GDDR5X. For the record I don't like the rumour, but I do believe they are a good source.
asdd
I do not know if HWbattle is using same source or they have another "solid" one.
http://www.hwbattle.com/bbs/board.php?bo_table=news&wr_id=18732
Cheers
I thought recently Sweclockers (which I linked earlier) are suggesting there will be replacement for 980 and 980ti initially, followed up a little bit later with the replacement to 970 - I have reservations about this rumour for reasons I mentioned earlier.
I do not know if HWbattle is using same source or they have another "solid" one.
http://www.hwbattle.com/bbs/board.php?bo_table=news&wr_id=18732
Cheers
I have a problem with statements like this Adored. First off Polaris 10 was shown playing Hitman in DX12 at 1440p, at unknown settings, and it was stated it played at 60fp all the time (with frame cap), in a limited part of the game, we don't know what FuryX was doing in that part of the game with a similar settings. Fury X does get around 60 FPS at that resolution at the max settings for the entire game. So we can't draw any conclusions from that demonstration with Polaris 10 because we don't know the settings.
hmm maybe they are getting mixed up with a lower end card.... I don't think a 980ti replacement is coming this soon.
