vacancies advertise contact news tip The Vault
facebook rss twitter

AMD Radeon Vega Frontier Edition arrives in June

by Mark Tyson on 17 May 2017, 11:01

Tags: AMD (NYSE:AMD)

Quick Link: HEXUS.net/qadhjs

Add to My Vault: x

Here we have the first graphics card based upon the high-end Radeon Vega architecture. The AMD Radeon Vega Frontier Edition will be available to purchase from "late June". There's some less-good news; the new card is aimed at "data scientists, immersion engineers, and product designers," not consumer-side PC enthusiasts or gamers. However, the presentation last night surely whets ones appetite for the consumer cards that will come.

Before describing the Radeon Vega Frontier Edition and its capabilities further let's look at a specifications summary. As you can see in the slide below, this graphics card packs in 64 next-gen compute units providing 4096 shader processors, 16GB HBM2, supports 8K displays, and offers 12.5TFLOPS FP32 performance or 25TFLOPS of FP16 performance.

In various product shots you can see 2x 8-pin power connectors. In another factoid AMD will be producing a gold liquid cooled version of the Vega Frontier Edition.

At the presentation Raja Koduri shared some comparative performance slides comparing the new AMD Radeon Vega Frontier Edition against older AMD tech such as the Radeon Fury X and against arch rival Nvidia's latest and greatest Tesla P100 accelerators. Check out the comparison slide below and you can see AMD's  deep learning achievement.

Unfortunately for PC enthusiasts AMD's presentation looked at some gaming enhancement tech but not directly at gaming performance. For example we heard about the High Bandwidth Cache Controller helping make the most of memory constraints. Furthermore Vega was designed to deliver smooth 4K gameplay at 60fps+.

Rather than gaming AMD sees the Radeon Vega Frontier Edition cards being used as follows:

  • Machine learning – via AMD's ROCm open software platform. See Deepbench benchmark above
  • Advanced visualization - Vega can outrun Titan Xp by an average of 42 per cent in some pro apps, says AMD.
  • VR workloads
  • Game design workflows

Last but not least a Radeon Instinct accelerator based on the same GPU, dubbed the MI25, will be released.

Future graphics

We haven't been tipped about consumer Radeon Vega graphics cards. Hopefully that announcement will come at Computex. With Vega-based APUs delivered in H2 2017, the enthusiast desktop cards are likely to arrive earlier in that release window.

At the financial analyst presentation AMD touched a little on the future of graphics up to 2020. "AMD plans to follow its 'Vega' architecture with the introduction of 'Navi' and its subsequent next-gen architecture, both of which are planned to be built using 7nm process technology," stated AMD.



HEXUS Forums :: 13 Comments

Login with Forum Account

Don't have an account? Register today!
Correct me if I'm wrong:

The fans were expecting the 12.5tflop card to be the big Vega competing against the 1080ti.

I know gamer/enthusiast cards tend to be clocked higher than workstation/computational cards, but still, it doesn't look too good for them, not unless the gamer card clock is cranked even harder.
On the financial day, it makes sense to push the high margin enterprise products. Especially if that allows you to gloss over another month delay on the consumer parts.
So Vega is Q2, but not the Vega that everyone actually wants? :/
daddacool
So Vega is Q2, but not the Vega that everyone actually wants? :/


Who is this ‘everyone’ of which you speak? :O_o1:
daddacool
So Vega is Q2, but not the Vega that everyone actually wants? :/

Well, F's Edition and a Titan like release. Guess AMD could use some those crazy margins but while NV are really good at getting people to spend big, I doubt AMD will have as much success there.

Of course, if the chip was ready ages (demos back in December etc.) then aside from the Ryzen launch (rightly) taking all their resources (well any which can be shared between Radeon Group and CPU), there are really only two possibilities for the delay I can think of:

  1. HBM2 shortages
  2. Having to go back for revision

Still hope, they used the time to have polished launch drivers.