Tuesday, May 16th 2017

AMD Announces Radeon Vega Frontier Edition - Not for Gamers

Where is Vega? When is it launching? On AMD's Financial Analyst Day 2017, Raja Koduri spoke about the speculation in the past few weeks, and brought us an answer: Radeon Vega Frontier Edition is the first iteration of Vega, aimed at data scientists, immersion engineers and product designers. It will be released in the second half of June for AMD's "pioneers". The wording, that Vega Frontier Edition will be released in the second half of June, makes it so that AMD still technically releases Vega in the 2H 2017... It's just not the consumer, gaming Vega version of the chip. This could unfortunately signify an after-June release time-frame for consumer GPUs based on the Vega micro-architecture.

This news comes as a disappointment to all gamers who have been hoping for Vega for gaming, because it reminds of what happened with dual Fiji. A promising design which ended up unsuitable for gaming and was thus marketed for content creators as Radeon Pro Duo, with little success. But there is still hope: it just looks like we really will have to wait for Computex 2017 to see some measure of details on Vega's gaming prowess.

Vega Frontier Edition is the Vega GPU we've been seeing in leaks in the last few weeks, packing 16 GB of HBM2 memory, which, as we posited, didn't really make much sense on typical gaming workloads. But we have to say that if AMD's Vega truly does deliver only a 1.5x improvement in FP32 performance (the one that's most critical for gaming at the moment), this probably paints AMD's Vega as fighting an uphill battle against NVIDIA's Pascal architecture (probably ending up somewhere between GTX 1070 and GTX 1080). If these are correct, this could mean a dual GPU Vega is indeed in the works, so as to allow AMD to reclaim the performance crown from NVIDIA, albeit with a dual-GPU configuration against NVIDIA's current single-chip performance king, Titan Xp. Also worth nothing is that the AMD Radeon Vega Frontier Edition uses two PCI-Express 8-pin power connectors, which suggests a power draw north of 300 Watts.
For now, it seems AMD actually did its best to go all out on the machine learning craze, looking for the higher profits that are available in the professional market segment than on the consumer side of graphics. Let's just hope they didn't do so at the expense of gaming performance leaps.

After an initial throwback to AMD's times since he became lead of Radeon Technologies Group, where Raja mentioned the growing amount of graphics engineers in AMD, including their commitment to the basics of graphics computing: power, performance, and software. Better basics in hardware, software, and marketing are things that Raja says are responsible for AMD's current market outlook, both from a gamer and content creator perspective, which led to an increase in AMD's graphics marketshare.
RTG's chapter two of Radeon Rising, going beyond the basics, will allow the company to go after premium market dollars, with an architecture that excels on both gaming and CAD applications. Raja Koduri said he agreed with NVIDIA CEO Jensen Huang in that at some point in the future, every single human being will be a gamer.
The final configuration of Vega was finalized some two years ago, and AMD's vision for it was to have a GPU that could plow through 4K resolutions at over 60 frames per second. And Vega has achieved it. Sniper Elite 4 at over 60 FPS on 4K. Afterwards, Raja talked about AMD's High Bandwidth Cache Controller, running Rise of the Tomb Raider, giving the system only 2 GB of system memory, with the HBCC-enabled system delivering more than 3x the minimum frame-rates than the non-HBCC enabled system, something we've seen in the past, though on Deus Ex: mankind Divided. So now we know that wasn't just a single-shot trick.
Raja Koduri then showed AMD's SSG implementation and how it works on a fully ray-traced environment, with the SSG system delivering much smoother transitions in the system. AMD worked with Adobe on integrating SSG capability into Adobe Premiere Pro.
Raja then jumped towards machine intelligence, which Raja believes will be dominated not by the GPU (NVIDIA's green) or CPU (Intel blue) paths, but in true heterogeneous computing.
Raja took to stage results on DeepBench, a machine learning benchmark where NVIDIA dominates at the moment, joking about AMD's absence from the benchmark - since they really didn't have a presence in this area. In a benchmark, AMD pitted Vega against NVIDIA's P100 architecture (interestingly, not against NVIDIA's recently announced V100 architecture, which brings many specific improvements to this kind of workloads), delivering an almost 30% performance lead.
Add your own comment

91 Comments on AMD Announces Radeon Vega Frontier Edition - Not for Gamers

#1
eidairaman1
The Exiled Airman
Too bad this isn't the days of past where workstation boards could be bios modded for gaming.
Posted on Reply
#2
xkm1948
Raja was shying away from Vega's gaming performance as much as he can. And the first card using Vega will be a Radeon PRO workstation card. Not keeping my hopes high then. Realistically anywhere around 1080 performance would be something to celebrate.

Also that card he demoed in the end has 16GB of HBM2? Four stacks of HBM2 or Dual GPUs?

Also Computex will probably be a paper launch. As the availability is June.
Posted on Reply
#3
Dante Uchiha
"Vega has achieved it. Sniper Elite 4 at over 60 FPS on 4K"

vs.

Posted on Reply
#4
Absolution
AMD pitted Vega against NVIDIA's P100 architecture (interestingly, not against NVIDIA's recently announced V100 architecture, which brings many specific improvements to this kind of workloads),
Welp, always one step behind AMD :(
Posted on Reply
#5
efikkan
"Polaris: better hardware"
"Up to 2.8× Performance-per-Watt"
PR usually stretches the truth, but that's far fetched.
Posted on Reply
#6
Absolution
Dante Uchiha said:
"Vega has achieved it. Sniper Elite 4 at over 60 FPS on 4K"

vs.


No idea on settings though

efikkan said:
"Polaris: better hardware"
"Up to 2.8× Performance-per-Watt"
PR usually stretches the truth, but that's far fetched.
You'd think they would let go of Polaris already
Posted on Reply
#7
oxidized
I think it's actually good that raja has toned it down a bit, it'll make look AMD ads less ridiculous. Pretty satisfied by the cpu side actually.
Posted on Reply
#8
iO
xkm1948 said:
...Also that card he demoed in the end has 16GB of HBM2? Four stacks of HBM2 or Dual GPUs?..
Two 8-Hi stacks.
Posted on Reply
#9
Grings
Raevenlord said:
(interestingly, not against NVIDIA's recently announced V100 architecture, which brings many specific improvements to this kind of workloads)
Recently announced, so how would AMD have one already to use in a live demo?
Posted on Reply
#10
xkm1948
Grings said:
Recently announced, so how would AMD have one already to use in a live demo?
p100, not v100
Posted on Reply
#11
Fluffmeister
I'm just happy there was a graph confirming that 16GB is indeed four times the memory of Fury X.

But joking aside, I'm all for frontier editions, but what about consumer editions?

Also, I'm drunk.
Posted on Reply
#12
efikkan
Raevenlord said:
In a benchmark, AMD pitted Vega against NVIDIA's P100 architecture (interestingly, not against NVIDIA's recently announced V100 architecture, which brings many specific improvements to this kind of workloads), delivering an almost 30% performance lead.
In all fairness, Volta is at least not released yet.

But in the Polaris/Polaris refresh PR materials AMD chose Maxwell as comparison, not the more relevant Pascal.

Absolution said:

You'd think they would let go of Polaris already
But stay tuned, Vega is bringing 4× performance per watt. (yeah right)
Posted on Reply
#13
Grings
xkm1948 said:
p100, not v100
i was referring to the 'interestingly not against' part, i know they used a p100
Posted on Reply
#14
xkm1948
Fluffmeister said:
I'm just happy there was a graph confirming that 16GB is indeed four times the memory of Fury X.

But joking aside, I'm all for frontier editions, but what about consumer editions?

Also, I'm drunk.
Is it just me or was Raja trying his best to avoid talking about Vega gaming performance? You would think that high end gaming GPU is great for profiting margins and Raja seems to avoid it as hard as he could.
Posted on Reply
#15
Caring1
Fluffmeister said:
I'm just happy there was a graph confirming that 16GB is indeed four times the memory of Fury X.

But joking aside, I'm all for frontier editions, but what about consuming editions?

Also, I'm drunk.
With the right sauce I suppose they might be easier to digest.
Posted on Reply
#16
Toothless
xkm1948 said:
Raja was shying away from Vega's gaming performance as much as he can. And the first card using Vega will be a Radeon PRO workstation card. Not keeping my hopes high then. Realistically anywhere around 1080 performance would be something to celebrate.

Also that card he demoed in the end has 16GB of HBM2? Four stacks of HBM2 or Dual GPUs?

Also Computex will probably be a paper launch. As the availability is June.
xkm1948 said:
Is it just me or was Raja trying his best to avoid talking about Vega gaming performance? You would think that high end gaming GPU is great for profiting margins and Raja seems to avoid it as hard as he could.
You confuse me.
Posted on Reply
#17
Fluffmeister
I'll give AMD credit here, and as their own slide acknowledges... compute is where the money is, and like Nv it is where they should focus their efforts.

Technology should trickle down, giving it away for cheap right from day one might gain a few fans... but it doesn't pay the bills
Posted on Reply
#18
W1zzard
Fluffmeister said:
compute is where the money is
Not true I think. Biggest margins yes. But there's big margins also in enthusiast gamer and much bigger volume
Posted on Reply
#20
Basard
We're all doomed....
Posted on Reply
#21
Patriot
I mean... nice marketing words... but nvidia is training 100k devs in deep learning, AMD isn't hardly supporting ROCm.
Posted on Reply
#22
MrMilli
But we have to say that if AMD's Vega truly does deliver only a 1.5x improvement in FP32 performance (the one that's most critical for gaming at the moment), this probably paints AMD's Vega as fighting an uphill battle against NVIDIA's Pascal architecture (probably ending up somewhere between GTX 1070 and GTX 1080).

Why is that so? Theoretical FLOPS have no direct relation to actual performance when comparing in between different architectures, as has already been proven by the current situation between AMD and nVidia. If Vega can perform closer to its theoretical performance compared to Fiji then real life performance will be >50% increased. From all the slides/info I've seen, everything point to Vega being more efficient compared to Fiji.
Posted on Reply
#23
Fluffmeister
W1zzard said:
Not true I think. Biggest margins yes. But there's big margins also in enthusiast gamer and much bigger volume
Yeah I guess, that is fair enough. But that in turn begs the question.... when is the enthusiast gamer gonna get their hands on one?

Is this card coming late June really first Vega based card coming to the market?
Posted on Reply
#24
xkm1948
Do the Radeon Pro cards have Crimson gaming driver support? If so one can easily take this during launch and do game benchmarks on it.
Posted on Reply
#25
yotano211
If this thing is going to be release in late June, when is the consumer card going to launch, late July?
Posted on Reply
Add your own comment