Skip to main content

AMD Vega Details Leak Ahead Of Official Launch

AMD Vega Frontier Edition

The long-awaited unveiling of AMD's Vega is scheduled for July 30 at the company's Capsaicin shindig, just prior to SIGGRAPH, but a few details are leaking out ahead of time. We have a few of our own sources who are telling us that there will be four different reference design models, including two versions at the top end (the XTX), a presumably lesser-resourced XT model, and then the XL at the lower end.

CUThermal (Reference Design)Board PowerASIC Power
Vega XTX (OC)64Water Cooling375W300W
Air Cooling285W220W
Vega XT64Air Cooling285W220W
Vega XL56Air Cooling285W220W

We're told the XTX will come as either an air-cooled design with a blower fan or liquid-cooled, with the latter at 375W TDP and the former at 285W TDP. The XT and XL will both be air cooled and operate at 285W TDP. The already-shipping Vega Frontier Edition also comes in air- or liquid-cooled variants, with the air cooled version hitting 300W TDP.

All cards will come equipped with HBM2 memory, of course. The only information we are able to suss out is that the XTX will likely start out with 8GB of HBM2. The Frontier Edition uses 16GB of HBM2, or "high-bandwidth cache," as AMD has taken to calling it.

The only other aspect we've learned is that the top-end models will come with 64 compute units, much like Vega Frontier Edition, whereas the low-end XL will have 56. Of course, we weren't made privy to shader counts, texture units, cache, and other resources and efficiency enhancements. As a reminder, the Radeon RX 480 had 36 CUs, and the Radeon R9 390 had 40. Vega Frontier edition's CUs have 4,096 stream processors. We know much more about Frontier Edition's floating point performance as well, but we'll have to wait--either for more leaks or for July 30.

  • Nomnom_136
    The High bandwidth cache controller is a feature, HBCC, they don't call the memory high bandwidth cache..why are you writing an article about something you know nothing about?
    Reply
  • nitrium
    Seems a quite a high TDP - about the same as an R9 390 (which actually melted one of my Seasonic's PSU 8-pin plugs the other day). The burning question is how much faster will it be?
    Reply
  • mp_4__
    How much faster than what? Previous AMD architectures? Current performance king? You need to be a liitle more specific when asking questions. We already know that Vega Frontier Edition is at the level of GTX 1080 so a watercooled Vega would be in between 1080 and 1080 Ti. The big question here must be the pricing : A 1000$ product that cannot beat another at 700 is not even worth considering, so the price of the RX Vega needs to be adjusted according to performance, otherwise will be a big flop.
    Reply
  • PaulAlcorn
    19935523 said:
    The High bandwidth cache controller is a feature, HBCC, they don't call the memory high bandwidth cache..why are you writing an article about something you know nothing about?

    Yes, they do call it high bandwidth cache. From previous coverage.
    It’s no secret that Vega makes use of HBM 2; that information was on roadmaps through 2016. But we now know that AMD wants to call this pool of on-package memory—previously the frame buffer—a high-bandwidth cache. Got it? HBM2 equals high-bandwidth cache now. Why? Because AMD says so.

    http://www.tomshardware.com/news/amd-vega-architecture-new-features,33265.html


    Reply
  • JamesSneed
    19935625 said:
    How much faster than what? Previous AMD architectures? Current performance king? You need to be a liitle more specific when asking questions. We already know that Vega Frontier Edition is at the level of GTX 1080 so a watercooled Vega would be in between 1080 and 1080 Ti. The big question here must be the pricing : A 1000$ product that cannot beat another at 700 is not even worth considering, so the price of the RX Vega needs to be adjusted according to performance, otherwise will be a big flop.

    I do agree those are the expectations. Although I would like to see Vega reviews first because driver optimizations could easily eek out another 10% performance which would lean more into the 1080ti arena. The key like you said is pricing. Another odd factor is how good this is at mining, well since that is effecting pricing these days, sadly.
    Reply
  • Martell1977
    This isn't as bad as it looks, considering the 1080ti (ASUS Strix) averages 283w and peaked at 293w.
    https://www.techpowerup.com/reviews/ASUS/GTX_1080_Ti_Strix_OC/27.html
    So while not as efficient, it's not massively more then it's competition.
    Reply
  • bit_user
    Vega Frontier edition's CUs have 4,096 stream processors.
    So does R9 Nano and R9 Fury X (both Fiji XT).
    Reply
  • bit_user
    19935782 said:
    very high TDP ... I m very disappointed.
    After Polaris, you're surprised?
    Reply
  • InvalidError
    19935590 said:
    Seems a quite a high TDP - about the same as an R9 390 (which actually melted one of my Seasonic's PSU 8-pin plugs the other day).
    If the connector got hot enough to melt, then you had a defective connector, crimp or solder job either on the GPU card or PCIe cable. MiniFit Jr pins are rated for 9A each even for the "low spec" variant, which would be enough to safely pass ~300W over the PCIe power cables' three 12V wires.
    Reply
  • TJ Hooker
    19935854 said:
    but look at the bright side , those are not good for Crypto mining at that TDP ... they will not be overpriced nor out of stock.
    Eh, not necessarily. It's all about hash/dollar and hash/watt. Also, ETH is memory bound, so you aren't pinning the GPU at 100% and hitting those TDPs (can even undervolt/underclock without really affecting performance). In theory ETH should love the bandwidth of HBM2, although it remains to be seen whether the existing algorithms will work with that kind of memory. I know they don't perform particularly well with GDDR5X.
    Reply