The Vega Architecture: AMD’s Brightest Day

From an architectural standpoint, AMD’s engineers consider the Vega architecture to be their most sweeping architectural change in five years. And looking over everything that has been added to the architecture, it’s easy to see why. In terms of core graphics/compute features, Vega introduces more than any other iteration of GCN before it.

Speaking of GCN, before getting too deep here, it’s interesting to note that at least publicly, AMD is shying away from the Graphics Core Next name. GCN doesn’t appear anywhere in AMD’s whitepaper, while in programmers’ documents such as the shader ISA, the name is still present. But at least for the purposes of public discussion, rather than using the term GCN 5, AMD is consistently calling it the Vega architecture. Though make no mistake, this is still very much GCN, so AMD’s basic GPU execution model remains.

So what does Vega bring to the table? Back in January we got what has turned out to be a fairly extensive high-level overview of Vega’s main architectural improvements. In a nutshell, Vega is:

  • Higher clocks
  • Double rate FP16 math (Rapid Packed Math)
  • HBM2
  • New memory page management for the high-bandwidth cache controller
  • Tiled rasterization (Draw Stream Binning Rasterizer)
  • Increased ROP efficiency via L2 cache
  • Improved geometry engine
  • Primitive shading for even faster triangle culling
  • Direct3D feature level 12_1 graphics features
  • Improved display controllers

The interesting thing is that even with this significant number of changes, the Vega ISA is not a complete departure from the GCN4 ISA. AMD has added a number of new instructions – mostly for FP16 operations – along with some additional instructions that they expect to improve performance for video processing and some 8-bit integer operations, but nothing that radically upends Vega from earlier ISAs. So in terms of compute, Vega is still very comparable to Polaris and Fiji in terms of how data moves through the GPU.

Consequently, the burning question I think many will ask is if the effective compute IPC is significantly higher than Fiji, and the answer is no. AMD has actually taken significant pains to keep the throughput latency of a CU at 4 cycles (4 stages deep), however strictly speaking, existing code isn’t going to run any faster on Vega than earlier architectures. In order to wring the most out of Vega’s new CUs, you need to take advantage of the new compute features. Note that this doesn’t mean that compilers can’t take advantage of them on their own, but especially with the datatype matters, it’s important that code be designed for lower precision datatypes to begin with.

Vega 10: Fiji of the Stars Rapid Packed Math: Fast FP16 Comes to Consumer Cards
POST A COMMENT

215 Comments

View All Comments

  • Threska - Thursday, April 04, 2019 - link

    Hello. I'm writing from the future and I bring important news about Google Stadia.

    " To make it possible on its servers, Google has combined an x86 processor (likely an Intel one) with hyperthreading that runs at 2.7GHz, with 16GB of RAM, and a custom AMD graphics chip. It’s said to uses HBM 2 and has 56 compute units, delivering enough raw horsepower for 10.7 TFlops.

    That sounds like a modified Vega 56, although it’s equally possible that it’s one of AMD’s upcoming Navi line of graphics cards."

    https://www.digitaltrends.com/gaming/google-stadia...
    Reply
  • Stuka87 - Monday, August 14, 2017 - link

    So my question is, can these be under-volted like Polaris can for some pretty decent power savings, and what is the power usage like when you enable AMD's Chill mode. They had stated you get about 90-95% of the performance but at a significantly lower power usage. Reply
  • tamalero - Monday, August 14, 2017 - link

    Does this means that all the future of VEGA 64 will rest in the hands of FINEWINE(tm)'s optimizations and boosts?

    Because right now Vega 64 is nothing but a disappointment.
    Reply
  • Chaser - Monday, August 14, 2017 - link

    This is a letdown. I don't understand why AMD chooses to lag behind Nvidia. The market is ripe for a competitive alternative to Nvidia. AMD hasn't been it. I am very pleased with my GTX 1080 purchase in January. Hopefully, come my next GPU upgrade time, AMD will have something better to consider. Reply
  • Stuka87 - Monday, August 14, 2017 - link

    They don't "choose" to. They had the money to either make an amazing CPU, or an amazing GPU. And the CPU market is larger, so they chose to push R&D budget into Ryzen (Which has payed off big time). Reply
  • TheinsanegamerN - Monday, August 14, 2017 - link

    They chose to split their resources between two GPUs (polaris and vega) rather then focusing on one line of chips. They chose to rebrand and resell the same chips for 5 years.

    AMD isnt rich, but they make quite a few boneheaded decisions.
    Reply
  • Aldaris - Monday, August 14, 2017 - link

    Actually, that looks like it paid off for them in market share. Also, Polaris was always out of stock (irrelevant as to the reasons why. It's still money in AMD's pocket). Reply
  • mapesdhs - Monday, August 14, 2017 - link

    That's a good point; whatever the buyer, a sale is still a sale. However, perhaps from AMD's pov they'd rather sell them to gamers because when Etherium finally crashes there will be a huge dump of used AMD cards on the market that will at least for a time stifle new card sales, whereas gamers tend to keep their cards for some time. Selling GPUs to miners now is certainly money in the bag, but it builds up a potential future sting. Reply
  • mattcrwi - Monday, August 14, 2017 - link

    I would never buy a used GPU that has been run at full throttle 24/7 for months. I'm sure some people won't understand what miners do with their cards or will be enticed by the prices anyway. Reply
  • wolfemane - Tuesday, August 15, 2017 - link

    I own a wide range of 290s and 290xs I picked up at the end of the last mining craze for great prices. Purchased all off miners. They all still work to this day with 0 issues. I've also purchased and sold 10x that quantity across 280 - 290x. Of those only one failed and sapphire replaced it under end of warranty.

    I look forward to the new craze ending. Will get some great cards for dirt cheap, and a vast majority still under warranty.

    Nothing wrong with buying them.
    Reply

Log in

Don't have an account? Sign up now