AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread
Remember this? http://www.extremetech.com/computin...y-with-tsmc-claims-22nm-essentially-worthless
If Nvidia is so unhappy about it you can imagine that AMD won't exactly be thrilled either. Who can really afford it so early on the process? Not AMD, that's for sure. They've gone on record to state that they won't be moving immediately on to new processes, and that 28nm is going to last longer than previous nodes.
If we see anything on 20nm before the end of this year/beginning of next it'll surely be from Apple or Qualcomm who can afford it. I maintain that we're more likely to see 20nm shrinks of Tegra and Temash in mid 2014 before we see new GeForce's or Radeons at the end.
If Nvidia is so unhappy about it you can imagine that AMD won't exactly be thrilled either. Who can really afford it so early on the process? Not AMD, that's for sure. They've gone on record to state that they won't be moving immediately on to new processes, and that 28nm is going to last longer than previous nodes.
If we see anything on 20nm before the end of this year/beginning of next it'll surely be from Apple or Qualcomm who can afford it. I maintain that we're more likely to see 20nm shrinks of Tegra and Temash in mid 2014 before we see new GeForce's or Radeons at the end.
UniversalTruth
Veteran
Hmm , TPU rewrites the story. Why?
AMD's Answer to GeForce GTX 700 Series: Volcanic Islands
the new GPU family sees AMD rearranging component-hierarchy within the GPU, in a big way
20 nm silicon fab process
4096 stream processors
16 serial processor cores
4 geometry engines
256 TMUs
64 ROPs
512-bit GDDR5 memory interface
http://news.softpedia.com/news/AMD-...c-Islands-GPUs-Set-for-Late-2013-351659.shtml
So, they will ditch the 8000 line and use it only for OEM/portables, and introduce 20nm Radeon 9970 in late 2013/early 2014?
Is there ANY way of predicting how faster will this monster be from 7970?
So, they will ditch the 8000 line and use it only for OEM/portables, and introduce 20nm Radeon 9970 in late 2013/early 2014?
Is there ANY way of predicting how faster will this monster be from 7970?Reposting the same stuff as in the OP, just reposted by different sites several times over, doesn't make it make any more sense. And that diagram surely does not make any sense at all for a GPU to begin with: "Serial Processing Modules", "Unified South Bridge", "FPU", "Security Co-processor (Arm Cortex A5)", ...
If anything it's an APU diagram, and I'm guessing some old next-gen console fake.
If anything it's an APU diagram, and I'm guessing some old next-gen console fake.
UniversalTruth
Veteran
Perhaps by taking the performance jumps between top solutions from the past. And statistically estimating the mean...
dagnastyep
Newcomer
The diagram lists the CU arrays as CUs, 16 CUs is 4096sp, if the next xbox was using this and has 12 CUs, then it would have 3072sp (7970 has 2048). That would be a serious monster chip...it would probably put it close to Titan power wise, not sure if we can expect something that powerfull.
Slow down folks. IF nv is rebadging the 28 nm chips, there is no way in hell AMD is ready to put out 20 nm monsters. Screams fake.
Readable image
Looks like 8 Bulldozer modules, and 16 CU's
Yeah, maybe a gpgpu optimised apu for supercomputers. Not a gpu, that's for sure.
It's Kryptos, the APU for Durango! (Just kidding)
If MS could manage a 16 core Steamroller design this year they deserve to win!
EDIT: Just to be clear, as of the recent hUMA slides "serial processing module" is the new AMD nomenclature for CPU cores, while "parallel compute modules" are the new designation for the GPU in an APU.
EDIT: Just to be clear, as of the recent hUMA slides "serial processing module" is the new AMD nomenclature for CPU cores, while "parallel compute modules" are the new designation for the GPU in an APU.
I wouldn't get too excited about the diagram, guys. It's apparently someone's personal mockup from 2012.
NextGen AMD GPU to have octa-precision FP?
http://www.hpcwire.com/hpcwire/2012...=2&featured=top&adclass=FrontPage TopFeatured
No, it's full ludicrous precision actually, you heard it here first! Now, let's draw a line in the sand and remember that this is Beyond3D. So no more clearly silly things please (ludicrous precision is one of them, the mock-up that gets thrown around and that is just a dude's drawing being another).
If John Gustafson mentioned it, I don't think we should dismiss the possibility of some nominal support for 128-256-bit precision, although it would most likely be no more than a hack using regular FMA units, and at a very substantial performance cost.
Still, it should be much less costly than doing it all in software. So if there's actual demand for it (something I'm admittedly not aware of) why not?