AMD: Speculation, Rumors, and Discussion (Archive)

Status
Not open for further replies.
Yep, free dGPU... and yet I'm complaining anyways because I would have rather seen the power budget go to the APU). How ungrateful is that ??

I'm running Linux on everything (bought the laptop to test the HSA stack on what would be available to customers) so Crossfire isn't a big draw for me. For someone running Windows it could be OK though, IIRC the APU and dGPU are configured with 6 CU's each so performance should be comparable. Haven't picked through the logs to see if it's running DDR3 or GDDR5 though.

If I get free time I guess I might try to bring the Boltzmann stack up on the Topaz dGPU :)
 
The Boltzmann runs on Tonga and Fiji today (although Tonga testing has been limited to kernel driver unit tests since our Fiji cards arrived), so hopefully supporting Iceland/Topaz should be pretty straightforward. The main task IIRC will be tweaking the topology code to properly support APU + dGPU -- so far testing has focused on "APU only" and "bunch of dGPUs" scenarios.

It would be nice to have something else to do with that dGPU, since I'm unlikely to bother using it for graphics other than maybe playing with the existing <render on dGPU display on APU> code in the all-open stack.
 
At MWC, Lenovo just announced a bunch of laptops, and one of them - Yoga 510 - is bringing an optional "R7 M460 2GB":

http://news.lenovo.com/news-release...-ready-windows-10-tablet-and-yoga-laptops.htm

EDIT: Here's what Ryan had to say about it yesterday:
We don't have any specs at this moment, but it's a rebrand. The next-gen FinFET products have not yet been announced, and AMD will want to do that themselves.

I also found this slide over at wccftech, but it could be made up by them:

OlLsCsq.jpg
 
Rebrand or Oland/Iceland upgraded to newer GCN (but not 4.0/Polaris)

edit: 460 might be too high number for oland/iceland though, but too low for Tonga probably? Maybe Bonaire or updated Bonaire

Anything starting with R7 is going to be low-end. Even the R7 M380 is only Cape Verde+DDR3. It's probably just Iceland.
 

I didn't read the entire thread, but we got a couple decent blurbs.



I've seen speculation of a possible far-future arrangement where the inherent difficulty of making big chips on super-small processes is combated by having one or two different designs of very small chip combined in great numbers on an interposer - could that actually become a reality? I vaguely remember reading a white paper that touched on the matter. Is there much you can say that isn't sensitive?

Yes, it is absolutely possible that one future for the chip design industry is breaking out very large chips into multiple small and discrete packages mounted to an interposer. We're a long ways off from that as an industry, but it's definitely an interesting way to approach the problem of complexity and the expenses of monolithic chips.

I'm thinking Robert may have dropped some info without knowing that he did it:

How are you guys going to cool the FuryX2? Is it going to be a blower like the dev kits Roy Taylor has been posting or is it going to be a liquid cooling loop like the 295x2? Also any idea of a price point?

With all due respect to Tizaki, he was erroneous in placing dual Fury on the list of kosher topics. I am not in a position where I can discuss that product.

Robert calls it "dual Fury". While it's clear that he's talking about Gemini, I think his phrasing points to two possibilities:

  • He was talking about a product with dual R9 Fury GPUs.
  • He was talking about a product with dual Fiji GPUs (e.g. Fury X, Fury or Nano, but not specifically any of them).

But no matter which, he was careful to not say "dual Fury X" despite the "Fury X2" or "dual Fury X" phrasing being used many many times by others in that thread (even Tizaki's post in question used "Fury X2" just for the record).

His wording could be innocent, but I'm wondering if it supports the previous theory that Gemini will end up being effectively 2x Nanos.

We've all seen the old info that Gemini will fit in a Tiki, when combined with the fact that the Tiki cannot handle the 500W 295X2 suggests that Gemini will fall well below that (likely 375W to fit into the PCIe spec). Two 175W Nanos would pretty cleanly fit under a 375W budget and would pretty closely match Gemini's specified 12TFLOPS performance.

And then there's the other old 2x Nano speculation because 2x Nano shows up in the benchmarks on Tech Report's SteamVR article. The idea was that Crossfire Nanos is one of the silliest dual GPU setups in history (if you have dual GPUs, you at least have a microATX mobo and effectively ammo microATX cases are big enough for longer GPUs) and yet it's the only dual GPU setup in the list despite dual GPU setups being considered very VR-friendly. Unless 2x Nano is really Gemini (or it is 2x Nano, but meant to emulate Gemini).

dvo2e8r.png
 
His wording could be innocent, but I'm wondering if it supports the previous theory that Gemini will end up being effectively 2x Nanos.
AMD has already stated in VRLA that Fiji Gemini is at 12 TFLOPS, which puts it slower than 2x Nano would be (2x full Fiji @ 12 TFLOPS = 732MHz, Nanos work around 8xxMHz most of the time IIRC)
 
AMD Radeon Technology Group group will hold GDC webcast on the 14th where they will talk about their nextgen GPUs, software products and game partnerships. Raja Koduri will host it.
http://wccftech.com/amd-capsaicin-gdc-2016-live-webcast/ [link also includes list of Radeon-related GDC talks]

Some more information here:
http://gpuopen.com/gdc-2016-presentations/?utm_source=silverpop&utm_medium=email&utm_campaign=24919071&utm_term=btn-learnmore&utm_content=p-global-developer-hcnewsflash-march-2016 (1):&spMailingID=24919071&spUserID=MjA2MzI4NTg0NDc5S0&spJobID=761337456&spReportId=NzYxMzM3NDU2S0

Not related but AMD Compress 2.2 is too out http://developer.amd.com/community/...0&spJobID=761337456&spReportId=NzYxMzM3NDU2S0
 
Status
Not open for further replies.
Back
Top