Wednesday, May 3rd 2017
AMD "Vega 10" Bears Core-Config Similarities to "Fiji"
A Linux patch for AMD's GPU drivers reveals that its upcoming "Vega 10" graphics processor bears numeric core-configuration similarities to the "Fiji" silicon which drives the enthusiast-segment Radeon R9 Fury series graphics cards. The patch bears configuration values which tell the software how to utilize the resources on the GPU, by spelling them out. The entry "gfx.config.max_shader_engines = 4," for example, indicates that "Vega 10" features four shader engines, like "Fiji." Another entry "Adev-> gfx.config.max_cu_per_sh = 16" signifies the number of GCN compute units (CUs) per shader engine. Assuming the number of stream processors per CU hasn't changed from 64 in the "Vega" architecture, we're looking at a total stream processor count of 4,096. This could also put the TMU count at 256.
At earlier reveals of the "Vega 10" package, you notice a large, somewhat square GPU die neighboring two smaller rectangular memory stack dies, which together sit on a shiny structure, which is the silicon interposer. The presence of just two memory stack dies sparked speculation that "Vega 10" features a narrower 2048-bit memory interface compared to the 4096-bit of "Fiji," but since the memory itself is newer-generation HBM2, which ticks at higher clocks, AMD could run them at double the memory clock as "Fiji" to arrive at the same 512 GB/s bandwidth. The 4,096 stream processors of "Vega 10" are two generations ahead of the ones on "Fiji," which together with 14 nm process-level improvements, could run at much higher GPU clocks, making AMD get back into the high-end graphics segment.
Sources:
aceCrasher (Reddit), ComputerBase.de
At earlier reveals of the "Vega 10" package, you notice a large, somewhat square GPU die neighboring two smaller rectangular memory stack dies, which together sit on a shiny structure, which is the silicon interposer. The presence of just two memory stack dies sparked speculation that "Vega 10" features a narrower 2048-bit memory interface compared to the 4096-bit of "Fiji," but since the memory itself is newer-generation HBM2, which ticks at higher clocks, AMD could run them at double the memory clock as "Fiji" to arrive at the same 512 GB/s bandwidth. The 4,096 stream processors of "Vega 10" are two generations ahead of the ones on "Fiji," which together with 14 nm process-level improvements, could run at much higher GPU clocks, making AMD get back into the high-end graphics segment.
63 Comments on AMD "Vega 10" Bears Core-Config Similarities to "Fiji"
I'm saying that increasing core clock is without a doubt the highest priority, and especially for gaming (which my original comment was about as well). Of course there are gains (Again, already said this), but you are wasting potential core clock and boost clocks by doing so, parameters that WILL increase performance much more.
Would you take a maximum performance boost of 7% or 19%? Again, the 980 is the absolute best example of this since it already has a small bus-width to a powerful GPU.
Well, I mined Dogecoins and random alt-coins at the time. I had every incentive to test. And so far your argument is non-existent, atleast I have presented some actual data instead of pure speculation and guessworksYes, thanks for that proof. I can totally see it now.
Beware of Greeks bearing gifts, is not the same as, Beware of Greeks baring gifts.
You know what would make me buy AMD? Better performance at a lower price. Cause it sure isn't going to be stability and efficiency.
Most cards are built with a decent balance of bandwidth and compute. If you increase the core clock, don't expect much more performance unless you can increase the bandwidth by the same amount.
As for "The 980 being a good example", no it isn't. The 980 has a paltry 5.0 TFlops of compute, and the aggressive memory compression introduced with Maxwell gave it FAR more effective bandwidth than previous 256-bit cards. If you want a good example of the bandwidth issue just look at the 1070/1080 and Polaris cards: They get tiny performance increases from MASSIVE clockspeed increases. In fact a 1070 at 2.1 GHz rarely gets more than a 5-10% performance increase unless you can overclock the memory a ton.
Vega on the other hand will be priced higher like you are saying, but don't expect them to sell it for the same price as Nvidia. The 1080 Ti came out first, and thus Vega will need to be cheaper if it has comparable performance. You can't just sell the same product later and expect success.
My guess is there are really 3 directions for how Vega will pan out:
- AMD manages to get 1550 - 1600 MHz clocks, and Vega turns out to be as big as a leap as Maxwell was for Nvidia. Expect a $600 - $700 card that matches or beats the Titan XP.
- AMD reaches decent 1400 - 1500 MHz clocks, and Vega is a nice new arch (but still not perfect). Expect a $500 - $600 card that trades blows with the 1080 Ti.
- Something goes horribly wrong and AMD can't get above 1200-1300 MHz, and Vega brings a modest 10% IPC increase. The card will cost $400 - $500 and land 15% stronger than the 1080, but 10% weaker than the 1080 Ti.
Still not sure what I would bet on yet, but rumors will probably give us a hint within a couple weeks...That 3% and 4% average increase sure sounds like a lot!
www.sweclockers.com/test/23718-snabbtest-nvidia-geforce-gtx-1080-och-gtx-1060-med-snabbare-minne
www.techpowerup.com/reviews/MSI/GTX_1080_Gaming_X/24.html
www.techpowerup.com/reviews/MSI/GTX_1080_Gaming_X_Plus_11_Gbps/30.html
(The Gaming X 11 Gpbs also maintains a higher core clock. surely that does not matter...)
If Vega has roughly the same performance per TFlop as Polaris, it would require over 14.1 TFlop/s to match GTX 1080 Ti, or ~1730 MHz average boost.
Additionally, Pascal is up to ~80% more energy efficient vs. Polaris. If Vega10 are to match GTX 1080 Ti, it needs to close most of this efficiency gap as well.