Tuesday, June 5th 2018

AMD Demonstrates 7nm Radeon Vega Instinct HPC Accelerator

AMD demonstrated the world's first GPU built on the 7 nanometer silicon fabrication process, a Radeon Vega Instinct HPC/AI accelerator, with a 7 nm GPU based on the "Vega" architecture, at its heart. This chip is an MCM of a 7 nm GPU die, and 32 GB HBM2 memory stacks over four stacks (4096-bit memory bus width). It's also the first product to feature a removable InfinityFabric interface (competition to NVIDIA's NVLink interface). There will also be variants based on the common PCI-Express 3.0 x16. The card supports hardware virtualization and new deep-learning ops.
Add your own comment

97 Comments on AMD Demonstrates 7nm Radeon Vega Instinct HPC Accelerator

#1
Jism
Other websites reported, that the 7nm vega is twice as efficient compared to the 14nm. And that the performance over the previous one is around 35% more. So we're looking at 1080Ti levels here.

Edit: god, lisa is so sexy with that GPU.
Posted on Reply
#2
cucker tarlson
Mostly due to x2 vram bandwidth I suppose. It'd be a goood card for gamurz, except it'll never be for them.
Posted on Reply
#3
Jism
cucker tarlson said:
Mostly due to x2 vram bandwidth I suppose. It'd be a goood card for gamurz, except it'll never be for them.
Not really. I think memory speed aside from just bandwidth is another important factor for the performance of that card.

You could have 900Gb/s with memory operating at 100Mhz. It's still slow.
Posted on Reply
#4
cucker tarlson
What are you talking about ? Bandwidth is what matters. Look at Fury X vs 980Ti at 4K. If anything HBM is better than gddr, not only does it have higher bandwidth but also lower latency.

Aside from that, I'm MAD at AMD for totally showing the middle finger to gamers with Vega. They'll be held in my contempt for this for quite some time. 290 was great, but after that it's been a shitshow in the enthusiast segment.
Posted on Reply
#5
Jism
cucker tarlson said:
What are you talking about ? Bandwidth is what matters. Look at Fury X vs 980Ti at 4K. If anything HBM is better than gddr, not only does it have higher bandwidth but also lower latency.

Aside from that, I'm MAD at AMD for totally showing the middle finger to gamers with Vega. They'll be held in my contempt for this for quite some time. 290 was great, but after that it's been a shitshow in the enthusiast segment.
Yeah and what creates this bandwidth? The speed of the ram. It's a very important role for graphics cards. As i said you can have 900Gb/s a second running at 100Mhz or 400Gb/s running at 800Mhz. The difference is that the 800Mhz with a much slower total bandwidth can / will be faster then the 900Gb/s. It just depends on what you are using this for. Compare a highway with 2 lanes with a limit of 180mph or a highway with 6 lanes and a limit of 80mph. Which one do you think is faster for the quick operations?

It was a design trade-off. They design primarily for professional market and create a second division from the same chip for just gaming. Nvidia on forehand cuts the FP64 performance by 1/64 to prevent people from buying gamers cards that are intended for the professional market. AMD does something in a simular way. But they both where professional chips at some point before going into gaming.
Posted on Reply
#7
RejZoR
I just wish RTG releases a competing card again and kicks NVIDIA's ass repeatedly for few years so we get back some good old competition again. This is fucking sad man, I miss the days when we awaited every GPU launch with excitement. It's all "meh" now, even GTX 1080 series, sure they were fast, but there is nothing really exciting about it. RX Vega had a lot, but they also cocked up bunch of stuff that could make it great or exciting. Ugh.

I feel so desperate that NVIDIA or AMD giving us SMAA as a setting to replace FXAA and MLAA option in graphic settings would be considered as "super exciting thing" at this point...
Posted on Reply
#8
W1zzard
Jism said:
Other websites reported, that the 7nm vega is twice as efficient compared to the 14nm. And that the performance over the previous one is around 35% more. So we're looking at 1080Ti levels here.

Edit: god, lisa is so sexy with that GPU.
Those claims are not combined, it's one or the other, I confirmed with amd
Posted on Reply
#9
R0H1T
W1zzard said:
Those claims are not combined, it's one or the other, I confirmed with amd
This thing's on 7nm TSMC, right?
Posted on Reply
#10
Prima.Vera
Jism said:

Edit: god, lisa is so sexy with that GPU.
????
Posted on Reply
#11
ppn
35% more core clock 2.0GHz be as power hungry as vega64 at 1.5 OR @1.5 GHz power would be 1/2. Now the 35% comes from bandwidth 484 to 1,28. So in fact +166% bandwidth improvement and same clock so it averages 35% not using the core benefits that 7nm offers and what they really mean by that.

W1zzard said:
Those claims are not combined, it's one or the other, I confirmed with amd
Posted on Reply
#12
techy1
Prima.Vera said:
????
give the poor man a break - he is just thirsty for any kind of new gpus and can not controll his emotions at this point... heck if Jensen would wipe out gtx 11x0 I would be in love
Posted on Reply
#13
R0H1T
ppn said:
35% more core clock 2.0GHz be as power hungry as vega64 at 1.5 OR @1.5 GHz power would be 1/2. Now the 35% comes from bandwidth 484 to 1,28. So in fact +166% bandwidth improvement and same clock so it averages 35% not using the core benefits that 7nm offers and what they really mean by that.
A Vega64 can be undervolted pretty easily, AMD ~ like often ~ ships them with too much & so Vega64 runs hotter than it needs to. See R9 Nano or the upcoming Vega Nano for instance, Nvidia on the other hand ships them almost perfectly.
Posted on Reply
#14
Jism
W1zzard said:
Those claims are not combined, it's one or the other, I confirmed with amd
https://nl.hardware.info/nieuws/57569/computex-amd-demonstreert-7nm-vega-gpu
Volgens AMD werkt de nieuwe chip 2x zo efficiënt als de eerste generatie Vega GPU en kunnen de prestaties tot 35% hoger zijn dan bij de bestaande Vega-kaarten.
Their source was AMD at computex. The translation is is that the new chip is supposed to be twice as efficient and performance can be 35% higher compared to current generation of Vega cards. But this is the Accelerator and not the vega card. However, the Vegas WAS an instinct before. So theoretically, 35% equals 1080ti levels or maybe a bit higher / lower. Still competing tho.

R0H1T said:
A Vega64 can be undervolted pretty easily, AMD ~ like often ~ ships them with too much & so Vega64 runs hotter than it needs to. See R9 Nano or the upcoming Vega Nano for instance, Nvidia on the other hand ships them almost perfectly.
Correct. There's an overhead of a few percent on any given AMD card. Why this is i'm not sure. Perhaps they want to maintain stability in all possible situations. They test these cards by putting them in working condition in the oven. It's just to ensure that the cards will sustain every possible scenario of such bad ventilated cases and so on.
Posted on Reply
#15
Vayra86
Here we go again.

If this card is 1080ti performance, its already too late. Not interested, and as expected.
Posted on Reply
#16
R0H1T
Vayra86 said:
Here we go again.

If this card is 1080ti performance, its already too late. Not interested, and as expected.
Do you really expect 32GB HBM2 to feature on a consumer dGPU atm?
Posted on Reply
#17
Vayra86
R0H1T said:
Do you really expect 32GB HBM2 to feature on a consumer dGPU atm?
No of course not, I expected a much quicker move towards GDDR5X / 6 by now and improvements to GCN. However, what we get is ANOTHER product nobody really cares for, which is just a shrink of something that wasn't good to begin with.

Double density and arriving at a 35% perf increase ... :oops:
Posted on Reply
#18
TheGuruStud
Vayra86 said:
No of course not, I expected a much quicker move towards GDDR5X / 6 by now and improvements to GCN. However, what we get is ANOTHER product nobody really cares for, which is just a shrink of something that wasn't good to begin with.

Double density and arriving at a 35% perf increase ... :oops:
It's not for you and never was... In fact, unless you're doing work, it's not for any of us.
Posted on Reply
#19
Jism
Vayra86 said:
No of course not, I expected a much quicker move towards GDDR5X / 6 by now and improvements to GCN. However, what we get is ANOTHER product nobody really cares for, which is just a shrink of something that wasn't good to begin with.

Double density and arriving at a 35% perf increase ... :oops:
It's on a smaller node, was said, written, announced from the start. Like going from 480 to 580 on a smaller node thus packing higher clocks.

Who cares. As long as the price is competetive we all good.
Posted on Reply
#20
IceShroom
cucker tarlson said:


Aside from that, I'm MAD at AMD for totally showing the middle finger to gamers with Vega. They'll be held in my contempt for this for quite some time. 290 was great, but after that it's been a shitshow in the enthusiast segment.
It is the gamers who first showed middle finger to AMD, not AMD.
Posted on Reply
#21
Vayra86
TheGuruStud said:
It's not for you and never was... In fact, unless you're doing work, it's not for any of us.
It doesn't excel in other areas either relative to the competition's offering.
Posted on Reply
#22
Flyordie
Well, its good news for Vega64 owners then. Means they plan on keeping Vega around for at least another 1.5 years or so. So developers will optimize further for it.
Posted on Reply
#23
TheGuruStud
Vayra86 said:
It doesn't excel in other areas either relative to the competition's offering.
I can buy several of them for the same price, though, correct?
Posted on Reply
#24
medi01
Is it possible to estimate chip size from the slides?
Posted on Reply
#25
ppn
medi01 said:
Is it possible to estimate chip size from the slides?
My calculation based on the assumption that the green pcb is the same size on both 14nm and 7nm VEGA is 232-242mm2 or ~1/2 VEGA64~ RX580
Posted on Reply
Add your own comment