AMD's Vega for gamers is finally here. The Radeon RX Vega 64 and RX Vega 56 mark AMD's return to the high end of the graphics-card world. After a long, long stretch wherein 2015's Radeon R9 Fury X and R9 Fury were asked to hold the fort against Nvidia's Pascal onslaught, the company is relieving them with what promises to be competition for Nvidia's long-dominant GeForce GTX 1080 and GeForce GTX 1070.
The Vega 10 GPU that's riding in on those cards is a massive and massively complex piece of silicon. It packs 12.5 billion transistors into a 486 mm² die fabricated on GlobalFoundries' 14nm LPP FinFET process. (For comparison, the Nvidia GP102 chip aboard the GTX 1080 Ti and friends is a similarly massive 12 billion transistors on a 471 mm² die.) The card's compute resources are organized into 64 "Next-gen Compute Units." Each of these hosts 64 stream processors for a total of 4096. The full Vega 10 chip has 256 texture units and 64 ROPs, too.
While the basic organization of Vega 10 may sound similar to Fiji before it, the similarities largely end at those broad outlines. I would love to explore Vega's many changes and capabilities in more depth, but when you have two days and change to review two brand-new graphics cards in the wake of testing and writing for a CPU review, stuff has to be left on the cutting-room floor, and a deep dive on Vega's new talents is one of them. We've known broadly what the new bits of Vega would be since January, however, so my architecture introduction is as good a place as any to start if you need to catch up. I'll be trying to add more information to this article as time goes on, but AMD should have a white paper available soon with full architectural details if you want to know much, much more.
The Radeon RX Vega 64 and RX Vega 56
The implementations of the Vega 10 GPU that will be available to consumers were revealed at SIGGRAPH a couple weeks ago, but to recap, AMD will be selling the fully-enabled Vega 10 GPU aboard the Radeon RX Vega 64 in both air- and liquid-cooled varieties. The Radeon Vega 56, on the other hand, loses eight NCUs to the world's tiniest chainsaw, and it'll only be available as an air-cooled card in its reference form. Curiously, AMD left all 64 ROPs intact on both the RX Vega 64 and RX Vega 56, meaning the cuts to the 56 may hurt less than they otherwise might.
|GTX 970||1050||1178||56||104||1664||224+32||224 GB/s||3.5+0.5GB||145W|
|GTX 980||1126||1216||64||128||2048||256||224 GB/s||4 GB||165W|
|GTX 980 Ti||1002||1075||96||176||2816||384||336 GB/s||6 GB||250W|
|Titan X (Maxwell)||1002||1075||96||192||3072||384||336 GB/s||12 GB||250W|
|GTX 1070||1506||1683||64||120||1920||256||259 GB/s||8GB||150W|
|GTX 1080||1607||1733||64||160||2560||256||320 GB/s||8GB||180W|
|GTX 1080 Ti||1480||1582||88||224||3584||352||484 GB/s||11GB||250W|
|Titan Xp||1480?||1582||96||240||3840||384||547 GB/s||12GB||250W|
|R9 Fury X||---||1050||64||256||4096||1024||512 GB/s||4GB||275W|
|Radeon RX Vega 64
|Radeon RX Vega 64
|Radeon RX Vega 56||1156||1471||64||224||3584||2048||410 GB/s||8GB||210W|
Both cards also ship with 8GB of HBM2 memory on board. That memory communicates with the Vega 10 chip across a 2048-bit bus. AMD doesn't disclose as much, but on the RX Vega 64 cards, that HBM2 runs at an effective rate of 1890 MT/s, and on the RX Vega 56, it runs at about 1600 MT/s.
|Asus R9 290X||67||185/92||4.2||5.9||346|
|Radeon R9 295 X2||130||358/179||8.1||11.3||640|
|Radeon R9 Fury X||67||269/134||4.2||8.6||512|
|GeForce GTX 780 Ti||37||223/223||4.6||5.3||336|
|Gigabyte GTX 980||85||170/170||5.3||5.4||224|
|GeForce GTX 980 Ti||95||189/189||6.5||6.1||336|
|GeForce Titan X||103||206/206||6.5||6.6||336|
|GeForce GTX 1070||108||202/202||5.0||7.0||259|
|GeForce GTX 1080||111||277/277||6.9||8.9||320|
|GeForce GTX 1080 Ti||139||354/354||9.5||11.3||484|
|GeForce Titan X (Pascal)||147||343/343||9.2||11.0||480|
|Radeon RX Vega 64 (air)||99||396/198||6.2||12.7||484|
|Radeon RX Vega 64 (liquid)||107||429/215||6.7||13.7||484|
|Radeon RX Vega 56||94||330/165||5.9||10.5||410|
We didn't have time (yet) ahead of publication to run our fancy Beyond3D test suite, so here are some potential peak rates for the RX Vega family. Compared to the GTX 1080, the RX Vega 64s trail slightly in pixel fill and peak rasterization rates, but they bring a prodigious array of shader and texturing power to the table, along with much higher memory bandwidth. The RX Vega 56 similarly trails and trounces the GTX 1070 in these theoretical peak rates. Our tests will tease out whether those theoretical victories translate into real performance shortly.
AMD's reference design for both the air-cooled RX Vega 64 and RX Vega 56 uses the same black shroud with an axial fan exhausting air directly out of the rear of the card. While this design may bear some similarities to the Radeon RX 480 before it, they're ony skin-deep. The shroud on this card is mostly metal and features a full metal backplate, as well. The Radeon logo on the side of the card lights up with red LEDs when the card is on.
One neat little touch from the R9 Fury X returns on board the RX Vegas: the "GPU Tach." This line of red LEDs (or blue, if you flip a DIP switch) will show you your GPU's occupancy, presuming you can see it from your desk or other site of sitting. The LEDs can also be turned off.
The Radeon RX Vega 56 will list for $399, or $20 more than the GTX 1070's original $379 suggested price. The RX Vega 64 air-cooled card will sticker for $499, or the same as the GTX 1080's most recent suggested price. Both cards are also available as limited editions and as parts of "Radeon Packs," which we described in more detail in our RX Vega reveal. We've been talking about RX Vega for ages, so now it's time to shut up and share performance numbers.