During CES 2017, chipmaker AMD officially revealed the first details about its Vega graphics processing unit (GPU) architecture. However, it wasn’t until April that we found out the first Vega cards would make their way to market in the second quarter of 2017, most likely in June, but certainly before July.
Unfortunately, questions still linger. We don’t have, for instance, an exact date on which the Vega GPUs will be released, nor do we know how much they will cost or what the spec range will consist of. After all, only one card based on the architecture has been announced and that’s the Radeon Vega Frontier Edition, designed to appeal to developers and creative professionals alike.
AMD held a livestream event on February 28, during the 2017 Game Developers Conference (GDC 2017), its second annual “Capsaicin” event. While the firm did share some more tidbits regarding its Vega GPUs, no concrete details were given regarding price, release date or power profiles – and AMD Ryzen received even less attention.
That said, we did learn that Bethesda plans to work with AMD to ensure that its games take every advantage of Vega and Ryzen technologies – from how its games exploit their benefits to adopting them into its server technology.
Without further ado, here’s everything else we know about Vega just following AMD’s Capsaicin 2017 event.
Cut to the chase
- What is it? The follow-up to AMD’s Polaris GPU architecture
- When’s it out? July at the latest
- What will it cost? Hopefully cheaper than Nvidia’s lot
AMD Vega release date
Although we expected to find out the release date of AMD’s Vega GPUs at its Capsaicin event earlier this year, it’s still lost on us when exactly we’ll see the cards on store shelves. All we know for fact right now is that the Frontier Edition, a non-consumer release, is coming out next month in liquid- and traditional air-cooled flavors.
We can also confidently say that the more affordable, consumer-facing AMD Vega products will be available to purchase this quarter. Otherwise, it’s unclear what day – or even what week – we’re bound to see AMD’s (hopefully) Nvidia GeForce GTX 1080 Ti contender start making its rounds. Then again, we’re even more out of the loop when it comes to AMD Vega pricing.
AMD Vega price
That’s right, with presumably a month to go before Vega arrives on our doorstep, we still know next to nothing about how much the Vega GPUs will cost.
What we do know is that the current Radeon RX 500 lineup starts as low as $169 (about £136, AU$219), and though it’s mostly defunct at this point, the Radeon R9 Fury X was still going for a cool $389 (about £313, AU$506) up until its demise.
So then, the question remains: what can this information prepare us for?
If AMD wants its Vega GPUs to be competitive on the high-end with Nvidia’s Pascal-series GTX 1080 and 1080 Ti, then maybe we will see AMD cards begin to approach the $699 or £699 (about AU$930) mark.
At the same time, there’s no doubt in our minds the company intends to keep up its low- to mid-range offerings come Vega so that less hardy spenders aren’t left in the dust.
AMD Vega specs
Following its 2017 Capsaicin livestream event, AMD has managed to maintain predictably mum on exact specifications for Vega GPUs.
We still don’t have confirmed clock speeds or memory details of these cards yet. What we do have, in classic journalistic form, are leaks. Leaks that stem from Wccftech in particular, fetched from a Linux graphics driver update submitted by AMD itself.
From that, we can now assume that Vega 10, the company’s flagship GPU based on the next-generation architecture, will boast 64 compute units, each housing 64 GCN stream processors, thereby bringing the total number of stream processors to 4,096.
Compare that to the 3,584 stream processors featured in the GTX 1080 Ti, and the Vega 10 could truly champion the AMD brand. Those stream processors are divided into four shader engines, too, each featuring two asynchronous compute units, one render backend and four texture blocks.
Digging even deeper, the render back-end is made up of 16 render output units (ROPs), with 64 ROPs in total. The RX 480 and 580 boast only half of that, though it should also be noted that the GTX 1080 Ti packs 88 ROP units in its sleek, silvery shell.
The texture blocks consist of 16 texture mapping units each. Of course, 16 texture mapping units times four texture blocks times four shader engines equals 256 texture mapping units in total. In comparison the RX 480/580 bear a total of 144 texture mapping units.
AMD claims that Vega’s high-bandwidth (HBM) cache controller has been rated to improve maximum frame rates by 50% and minimum frame rates by 100% over the previous generation with GDDR5. In an interview at GDC, AMD further explained that this tool increases effective memory size by mapping some into system RAM and smartly storing data on local memory.
We also know that the highest-end Vega GPU will contain the same amount of geometry engines for rendering polygons as the previous generation. However, these chips will be able to handle more than twice as many polygons per clock cycle, at 11, as the R9 Fury X’s four per clock.
This era of Vega GPUs will also ditch GDDR5 memory altogether for a new format known as HBM2, or high-bandwidth memory. AMD believes this decision will reduce the footprint of Vega graphics by 50%.
To that end, Vega chips will be more equipped to handle compute tasks than ever – specifically 8-bit, 16-bit, 32-bit or 64-bit operations in each clock cycle – the benefits of which will extend beyond video games.
Finally, AMD’s new Pixel engine will debut in Vega, now moved to a client of the GPU’s L2 memory cache. This will enable it handle graphics workloads which perform frequent read-after-write operations with less overhead on the rest of the GPU.
(We’ve also seen that at least one of the Vega graphics cards uses an 8-pin and 6-pin power connector, as opposed to the R9 Fury X’s dual 8-pins, according to PCWorld, but not from AMD.)
Stay tuned for more details regarding everything AMD Vega, as we’ll be updating this page with the latest as it happens.