Features & the Future Launch

Last but not least, along with today’s card announcements, AMD has also lifted the embargos on the architectural elements of the Vega 10 GPU. While there are no new architecture features to discuss – it turns out AMD was rather thorough with their preview at the start of this year – there’s still quite a bit in the way of details to dig into here. In fact it's more than I can hope to cover in a single day, let alone the few hours I’m getting to write this article ahead of AMD’s embargo. So I want to hit the highlights.

First and foremost, AMD has clarified that they do not consider Vega 10 to be a high performance computing (HPC) chip, unlike Hawaii. In particular, Vega 10’s native FP64 rate is just 1/16, like AMD’s other consumer chips. And the chip, while supporting HBM2’s native ECC mode, doesn’t have ECC on its internal pathways. This doesn’t mean that the chip can’t be placed in a server – its FP16 and INT16/INT8 performance is the cornerstone of its existence as part of the Radeon Instinct family. But that’s still a different market than traditional HPC. When we might see another HPC chip from AMD is anyone’s guess at this moment.

Second of all, we have a formal die size and transistor count for Vega 10. The GPU is officially 486mm2, containing 12.5B transistors therein. That amounts to 3.9B more transistors than Fiji – an especially apt comparison since Fiji is also a 64 CU/64 ROP card – all the while being 112mm2 smaller than Fiji (despite the higher transistor count) thanks to the higher density of GloFo’s 14nm LPP process, which AMD is using to fab the chip.

Talking to AMD’s engineers, what especially surprised me is where the bulk of those transistors went; the single largest consumer of the additional 3.9B transistors was spent on designing the chip to clock much higher than Fiji. Vega 10 can reach 1.7GHz, whereas Fiji couldn’t do much more than 1.05GHz. Additional transistors are needed to add pipeline stages at various points or build in latency hiding mechanisms, as electrons can only move so far on a single clock cycle; this is something we’ve seen in NVIDIA’s Pascal, not to mention countless CPU designs. Still, what it means is that those 3.9B transistors are serving a very important performance purpose: allowing AMD to clock the card high enough to see significant performance gains over Fiji.

Speaking of Fiji, there’s been some question over whether the already shipping Vega FE cards had AMD’s Draw Steam Binning Rasterizer enabled, which is one of the Vega architecture’s new features. The short answer is that no, the DSBR is not enabled in Vega FE’s current drivers. Whereas we have been told to expect it with the RX Vega launch. AMD is being careful not to make too many promises here – the performance and power impact of the DSBR vary wildly with the software used – but it means that the RX Vega will have a bit more going on than the Vega FE at launch.

Fifth, AMD is tweaking their boost clock mechanism and how it's advertised. Up until now, the boost clock was the maximum clock a Radeon card would hit. e.g. a stock Fury X topped out at 1.05GHz, and a stock RX 580 at 1340MHz. Starting with Vega RX, however, the company's definition of the boost clock is changing to the highest clockspeed a card is guaranteed to hit, but not the absolute highest clockspeed it can hit under load. There are still some nuances I want to test once I have the hardware in-hand, but at first glance this sounds closer to NVIDIA's definition, where each and every card is allowed to reach its natural limit (for the stock voltage).

Finally, while AMD didn’t clarify exactly when this change was made, the company has tweaked their Freesync Low Framerate Compensation technology to work with monitors that have a 2x range between min and max refresh rates, as opposed to 2.5x at launch. I’m told that 2.5x (or better) is still the ideal range, but it now works at 2x. This is an important development because the Samsung CF791 monitor that AMD is discounting as part of the Radeon Pack bundles has a 48Hz to 100Hz range; so it wouldn’t have been LFC capable under the old LFC rules.

Overall AMD has given us a lot to chew on – and more that can be covered in a single afternoon – but we still haven’t seen the whole picture. Today’s announcement of the card specifications and prices sets a very important baseline for what we should expect on August 14th, but no hard performance numbers to judge the final product. As I said at the start of this article, AMD’s marketing has done a very good job keeping Vega in the public’s mind. And today’s announcement is the next phase of that plan. But the end is approaching, and RX Vega’s full launch is drawing near, so it will be very interesting to see how everything we’ve learned over the last several months translates into gaming performance, power efficiency, and where AMD’s new place shall be in the realm of high-end video cards.

Buying RX Vega: Prices & Bundles
Comments Locked


View All Comments

  • VulkanMan - Sunday, July 30, 2017 - link

  • ddriver - Monday, July 31, 2017 - link

    $200 Discount: Samsung CF71 34-inch Widescreen Freesync Monitor

    Isn't that CF791?
  • coolhardware - Monday, July 31, 2017 - link

    I was checking on preordering a Vega and had the following question:

    How exactly do the recently unveiled editions at $399 & $499 compare spec wise to the already released Radeon Vega Frontier Edition?

    For reference, product listing for the released Vega at Amazon http://amzn.to/2vXKOsk (url shortended). I looked at the prices and assume the Frontier edition is better, but by how much?

    I kind of long for the simpler days of GPU naming ;-)
  • Amoro - Monday, July 31, 2017 - link

    Vega Frontier is very similar to RX Vega 64 with the same number of shaders, 4096, but has 16GB of HBM2 memory instead. Clockspeeds will differ depending on which version it is, Air versus Liquid.

    Vega 56 $399 will have cut down shaders for a total of 3584 shaders and lower clockspeeds.
  • coolhardware - Monday, July 31, 2017 - link

    Thank you kindly for the info, it is much appreciated! Looks like I should wait for the RX Vega 64 :-)
  • Scabies - Sunday, July 30, 2017 - link

    So where's the Nano?
  • tamalero - Monday, July 31, 2017 - link

    There are photos of it, but no more info than that?
  • tarqsharq - Monday, July 31, 2017 - link

    Is this what you wanted? https://smallformfactor.net/wp-content/uploads/Veg...
  • Mr Perfect - Monday, July 31, 2017 - link

    Yeah, I'm surprise that even these standard size cards are so long. Back when HBM was new, there where diagrams showing how moving the memory from PCB to interposer significantly reduced the PCB space requited. And yet these cards are just as big as traditional cards. What is all the extra PCB space being used for?
  • Threska - Monday, July 31, 2017 - link

    Heatsink? Voltage control? Built-in coffee warmer?

Log in

Don't have an account? Sign up now