AMD Vega Hardware Reviews

They explained it. The memory speed was changing even when they only touched the core.
Ah - and did they try independent OC of Core and Memory - iow only mem oc - yet?

Ah well, I guess Overclocking is for gamers.
 
Last edited:
Good to see the GamerNexus gave AMD and NVidia each the choice of one game to bench.
We communicated with both AMD and nVidia about the new titles on the bench, and gave each company the opportunity to ‘vote’ for a title they’d like to see us add. We figure this will help even out some of the game biases that exist.

  • Ghost Recon: Wildlands (built-in bench, Very High; recommended by nVidia)
  • Sniper Elite 4 (High, Async, Dx12; recommended by AMD)
  • For Honor (Extreme, manual bench as built-in is unrealistically abusive)
  • Ashes of the Singularity (GPU-focused, High, Dx12)
  • DOOM (Vulkan, Ultra, 0xAA, Async)
 
The Chip itself can warp space and time with its top-tier features, that's why area measurement results are so different across the board:yep2:

Don't forget the on die flux capacitor alone adds a ton of area.. people seem to be forgetting. Advanced Asynchronous Time and Space-warp for unparalleled out of this world gfx! :mrgreen:
 
Yeah, there's a dedicated PISS (Path Integral Solution Solver) that modulates quantum effects to encode clocking and timing information in sub-cycle timespans which allows it to overclock for brief amounts of time (< one cycle).

Do note that you are also able to overclock the PISS, but beware that by doing so, you risk overmodulating the quantum effects, causing the GPU to start rendering frames asynchronously to the natural flow of time, which in extreme cases, results in your GPU sporadically rendering frames of games you haven't even bought yet.
 
The Chip itself can warp space and time with its top-tier features, that's why area measurement results are so different across the board:yep2:

Don't forget the on die flux capacitor alone adds a ton of area.. people seem to be forgetting. Advanced Asynchronous Time and Space-warp for unparalleled out of this world gfx! :mrgreen:

Could we please not do this and keep the discussion on point. Yes, it's funny that people have a hard time using calipers, digital or not. Haha.

I would certainly take Raja Koduri's written words at face value. Wonky measurements or not.

What really have left me flabbergasted is the lack of apparent performance gains from all the improvements they have made to GCN and the increased transistor count compared to Fiji. Right now a lot of us are thinking a full node shrink of Fiji would have yielded the same or better performance with a very tight and lean R&D budget. It's just confusing at best right now but I suppose we will know more when RX Vega is released.
 
But it would not have had the performance in FP16 and INT8 of Vega.

If you look at the size of GP100 and GV100, it seems like dual rate FP16 might come with a price in die size.
 
What really have left me flabbergasted is the lack of apparent performance gains from all the improvements they have made to GCN and the increased transistor count compared to Fiji. Right now a lot of us are thinking a full node shrink of Fiji would have yielded the same or better performance with a very tight and lean R&D budget. It's just confusing at best right now but I suppose we will know more when RX Vega is released.
Everything seen so far suggests the Vega FE is currently acting just like it was Fiji, instead of having all the fancy new features and whatnot enabled.
It's been suggested that the current FE-drivers are in fact still "Fiji-drivers", which is supported by the fact that Doom performance is pretty much spot on with the Doom performance late last year when it was running Fiji-drivers (confirmed by AMD) and that the RasterBin(or-what-that-raster-pattern-test-program-was-called? ) output is identical to that of Fiji
 
So, it is binary compatible to Fiji?
¯\_(ツ)_/¯
AMD did indeed say at the time that those early Doom demos where using the current (at the time) Fiji drivers... now why would it still be the case more than 6 months later.... that's another mystery..
 
Why would it be using "Fiji" drivers and not even Polaris-derived ones?

Can please actually LINK to a statement from AMD that gives credence to the "Fiji" driver theory? I mean, we are on the Internet, where URLs are a thing.
 
Why would it be using "Fiji" drivers and not even Polaris-derived ones?

Can please actually LINK to a statement from AMD that gives credence to the "Fiji" driver theory? I mean, we are on the Internet, where URLs are a thing.
And why marketing people wouldn´t clarify this instead of letting people believe Vega is a steep back architecture?.
 
Back
Top