RX Vega Predictions

[Pemalite]

caffeinade said:

With the release of AMD's RX Vega nearing, one must ask themselves, how will it perform, and when will it finally come out.

Expect one of the Vega GPU's to perform around the Geforce 1070 - 1080.

caffeinade said:

RX Vega is AMD's next generation of GPUs set to release in  Q2 2018.

At-least one Vega variant is launching this year.

caffeinade said:

RX Vega is aiming to compete with Nvidia's high end range of GPUs.

And the hype train begins... Hopefully it's not a let down. People overhyped Phenom, Phenom 2, Bulldozer, Ryzen, Fury... And more. And they were all dissapointments in retrospect.

caffeinade said:

RX Vega has many improvements over AMD's past generation of chips, in respect to the architecture.
The PS4 Pro currently uses Some Vega tech in the custom GPU it posses, where as the Scorpio apparently does not.
RX Vega supports support rapid packed math (two FP16 operations in the same time as a FP32 operation) like the PS4 Pro and Switch, and the last few generations of Nvidia cards.

Can you outline what those improvements are and what technology the Playstation 4 uses that Vega has?

Also Scorpio does have some features/characteristics found in Vega.
And I quote:
"According to Goossen, some performance optimisations from the upcoming AMD Vega architecture factor into the Scorpio Engine's design, but other features that made it into PS4 Pro - for example, double-rate FP16 processing - do not."

http://www.eurogamer.net/articles/digitalfoundry-2017-the-scorpio-engine-in-depth

caffeinade said:

Personally I think that Vega will release in May of 2017, or early June 2017.

That tends to follow past precedents. May to July is when most of AMD's GPU's tend to drop as that fits in with the OEM cycles. It's expected.

   

caffeinade said:

I hope it (realistically):
    Is about 15 — 20% faster than the Titan Xp or GTX 1080Ti, FP32 TFLOPs wise (about 13.5 TFLOPs).
    Has about 750 — 1000+ GB/s of memory bandwidth (GTX 1080 Ti has ~ 484 GB/s).
    Launches at or below 700 — 800 US.
    Delivers on the promises of the advanced architecture.
    Has a consumer SKU with 8GB or more of RAM.

I am looking forward to reading your responses.

I was hoping you were going to be realistic. I would not expect 20% better performance than Titan unless we are looking at some edge-case scenario's like resolutions higher than 4k and lots of Asynchronous Compute.

HBM 1 had a bandwidth of 512GB/s and 4GB on Fury.
HBM 2 will double both for the same amount of chips/complexity and one must assume, cost.
So 1024GB/s and 8GB with a possible 2048GB and 16GB configurations.

shikamaru317 said:

Some specs for 1 of the 3 Vega SKU's leaked recently:

<SNIP>

Only problem is, we don't know where that leaked SKU falls on the heirarchy. If it's the most powerful Vega chipset, it's kind of dissapointing, 12.5 AMD tflops is not likely to beat the 11 tflop 1080ti, let alone the 12 tflop Titan XP. If it's the middle or lowest Vega chipset though, it sounds pretty good, sounds like it can at least beat the standard 1080, which would be great depending on pricing. I'm hearing that the slowest of the 3 is going to be faster than the 1070, which suggests this might be the middle SKU.

Those specs have been known for over a year already. It's not really new.

However, according to the 3D Mark leak, Vega had a core clock of 1.2Ghz.

4096 Shaders * 2 instructions * 1.2Ghz Core clock. = 9.8 Teraflops. And that fits nicely in the roughly doubling of a Radeon RX 470/570.

We also need to keep in mind that Vega's NCU architecture adopts many of the advancements found in Maxwell and Pascal, so the amount of work it should be able to do should be vastly superior to that of Polaris and older Graphics Core Next designs.

With that, don't compare AMD's flops against nVidia's. It will never end well, it's inaccurate and doesn't tell the entire story, Especially when the chippy's aren't even on the market yet.

Chazore said:
The more leaks we get, the more it seems likely that AMD are releasing a late catch up to the 1070 and 1080, which is quote disappointing, especially if the 1080 catch up is going to cost what Nvidia already priced theirs down to since the release of the 1080ti.

AMD is in a bit of a bind at the moment. They have a finite amount of resources available.
Plus the Radeon division has just had a massive shakeup about a year or so ago after just renaming GPU's for half a decade and spinkling a few highend cards on top... So we won't see the benefits of that shakeup for another year or two.

vivster said:
If it's not at least 3-5% above the 1080 it's DOA. Expecting a price tag similar to the 1080, maybe above.

There is no such thing as a bad GPU though. Only a bad price. If it it's cheaper than nVidia it could be a big winner.