![]() |
|
|
#12 |
|
Feb 2016
UK
26×7 Posts |
I only just saw the leaked slides claiming that. Not much to go on. Target 2H 2018, 7nm process, 1/2 DP compared to Vega 10's 1/16, same number of CU. Assuming similar clock rates, that could put Vega 20 around 6 DP TFLOPS. That could do some damage! But 1.5 to 2 years wait to find out...
Vega 10 in 1H 2017 is claimed to be about 750 DP GFLOPS, so somewhat reversing the recent trend. I would speculate, given the trend to offer increased FLOPS at reduced precision, maybe they will have a way to do the same the other way with good efficiency? 1/2 rate is pretty much the best case. Or maybe they don't have a good way and will rely on 7nm process to give them the transistor budget to squeeze it all in. |
|
|
|
|
|
#13 |
|
(loop (#_fork))
Feb 2006
Cambridge, England
2·7·461 Posts |
ECM results: 32 threads on dual-E5-2650 box do 1280 stage-1 ECM curves on a C194 in about 128000 seconds and 26.8 megajoules; one GTX1080 does 1280 stage-1 ECM curves on the same C194 in about 60000 seconds and 7.6 megajoules.
(stage-2 curves take about 706 thread-seconds on the E5-2650, so one box can keep up with two GPUs) Last fiddled with by fivemack on 2017-01-11 at 23:59 Reason: Replace numbers with correct numbers |
|
|
|
![]() |
Similar Threads
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| GPU brand | kracker | GPU Computing | 12 | 2014-04-20 15:56 |
| Brand New to Prime95 | bmorgan | PrimeNet | 5 | 2013-02-20 22:06 |
| NFS sqrt by hand | henryzz | Factoring | 18 | 2010-09-26 00:55 |
| Zeta function by hand | Damian | Math | 0 | 2006-07-27 14:43 |
| What CPU brand do you prefer? | eepiccolo | Hardware | 29 | 2003-05-11 05:57 |