I get the AMD lack of resources thing, I don't expect miracles but Fury X was the 980TI competition in summer 2015. The 980TI is roughly equivalent to the 1070. So if VEGA (heaven forbid) is at 1070 levels of performance then the entire idiocracy of AMD's GPU division has been almost back pedalling for two years. OK they went forward a tiny bit but it would be such a puny pathetic incremental improvement that if they are really GPU engineers they need to find another job as they lost their way.
When you put it that way, I find it hard to imagine that even the most jaded AMD hater (you know who you are! ) would honestly believe that AMD is actually going to release a card that performs at the same level of their ~2 year-old flagship. It makes no sense....(OK, let's ignore Bulldozer. )
I agree Raghu but additionally I have money screaming to go into AMD's pockets. I want seamless quadfire as in proper driver support with loads of memory so i can have 2 x2Vega in my machine. I don't even need quadfire yet or ever BUT I need better performance without the SLI/xfire overhead this year. FFS I have 1600W PSU I ran 6990 Quadfire, 295x2 quadfire and I need about 1.5-2x what I got with 2xTitan XP(~1080TI SLI equivalent). I don't give a fig/flip/frick how I get there I'll pay £1500 ($1700) this August for it but without that I'll be going Volta next and fk em (as in AMD). Clearly I'm
using the term 'need' loosely as I need beer food and water so it's more of a 'I would like if you cretins can pull your fingers out and get around to selling me this' sort of thing.
Chip of this core count will be around 750mm2 die size.CF/SLI will have issues with scaling and smoothness and there is no replacement for a single GPU which can provide the best performance. I think GV102 could easily be 50% faster than GP102. GV102 is most likely going to be a 5376 cores GPU with 5120 enabled at launch. With IPC improvements and 42.8% more cores than 1080 Ti and 40% more cores on a full GV102 vs GP102 its not so hard.
Not necessarily. GP102 has 1.5x the cores of GP104, along with a 1.5x wider memory bus, and the listed die size is exactly 1.5x that of GP104. A straight doubling of GP104 would give 5120 cores, a 512 bit bus and a 628mm² die size.Chip of this core count will be around 750mm2 die size.
Especially when you take into account that we should not expect Nvidia to use HBM2 on consumer products.
Expect also Q2 launch of Volta products, next year, rather than Early 2018.
It would probably be in the 600mm^2 range.Chip of this core count will be around 750mm2 die size.
Especially when you take into account that we should not expect Nvidia to use HBM2 on consumer products.
Expect also Q2 launch of Volta products, next year, rather than Early 2018.
Chip of this core count will be around 750mm2 die size.
Especially when you take into account that we should not expect Nvidia to use HBM2 on consumer products.
Expect also Q2 launch of Volta products, next year, rather than Early 2018.
It would probably be in the 600mm^2 range.
GV100's Tensor cores take a massive % of the die if you look at diagrams. Then there's DP, FP16, etc' etc'
There's plenty to tweak from GV100 that wouldn't matter to gamers.
It wont be 750mm2. Not sure where you get this estimation from but if we look at the downscaling of GP100 to the consumer orientated GP102, we could expect a hypothetical ~627mm2 GV102 with 5376CC (64CC per SM/14SM per GPC/6GPC) with a 384bit GDDR5X/6 memory system. This is probably sometime next year.
Using the same method (GP102 -> GP104), GV104 which could get released this year is in the vincinity of ~418mm2 with 3584CC (64CC per SM/14SM per GPC/4GPC) and 256bit GDDR5/5X/6 memory system. All this assumes that the volta based geforce variants shares some similarities to its compute version (like how SMs are partitioned etc). If its Pascal on steroids, then it may also be different. Thinking its more of the former and would be interesting to know how they achieved a 50% in efficiency in their SMs with Volta over Pascal (This would be hugely beneficial given they are stuck on 16nm or 12nm)
GV104 could very well be possible this year using GDDR5X memory and perhaps their 3rd or is it 4th(?) memory compression technology.
Anyway, the VEGA demo was abit underwhelming to say the least. If history tells us something, theres a lot of signs that it will underperform compared to expectation mainly due to the delays. Or else we would be seeing clear benchmark figures by now, something along the lines of the P100 vs Vega FE deepbench graphs. Doesn't bode too well imo.
It would probably be in the 600mm^2 range.
GV100's Tensor cores take a massive % of the die if you look at diagrams. Then there's DP, FP16, etc' etc'
There's plenty to tweak from GV100 that wouldn't matter to gamers.
Raja Koduri ""It was running comfortably about 60 and comfortably above any single gpu and we aren't done yet with Vega rx optimizations for prey""
http://imgur.com/a/aqvL7
Ugh I can't believe we have 2 more months of waiting and forum crap.
my failing R9 290
RMA it? You'll probably get Polaris or a 390 in return
That's a good point, we'll at least have some performance points in a month from now.Surely the Internet will get hold of the AMD Radeon RX Vega Frontier Edition when it releases, and beta-test and post results of it, in preparation for Vega Gaming Edition.
As crazy as trying to compete with Intel on the cpu manner given how many generations behind they were and yet they surpassed our expectations (the reasonable ones not the people who are expecting amd to compete directly with Intel with just 1 cpu drop).It just seems a bit crazy to think that AMD could skip competing vs Pascal and then come out and competitively go after Volta. Although they did throw a bit of smack talk with the "poor volta" thing. I didn't personally think they were being serious, but you have a strong technical understanding so perhaps they were being serious. Very interesting. Your insight is always much appreciated.
Rofl.... Please leave that nationalism out of here. I'm as pro my own country as it gets but I'm crazy enough to believe china can't develop good tech.Both Polaris and Vega were developed in China, which may explain why they are so underwhelming.
The problem is that if it takes two Vega GPUs to beat a single top-tier Nvidia card, then Vega is still coming up short.
No one wants to screw around with all the drawbacks of multi-GPU if they can help it. Unless AMD has rolled out some secret sauce that fixes things and causes two GPUs to transparently appear as one big GPU in 100% of applications, then this is basically admitting a shortfall.
If one Vega could beat "any single GPU" then Raja would have said so.
He failed, and should be fired. So should the Chinese development team, and R&D brought back to the USA.
If you go up a bit, you'll see the convo and explanation on why people think it was running at close to ~120 fps.Came across the Twitter convo Raja posted his comment about performance, and it's not very reassuring.
https://twitter.com/GFXChipTweeter/status/869895814545514497
Unless someone has a follow up, the question is regarding CFX Vega, and he states it ran above 60 FPS. That is beyond vague, but considering the convo. CFX Vega running above 60 FPS is expected.
Really glad I stopped waiting on AMD to deliver. But I still prefer a two person fight. At least then I had better options and price tags.