AMD's Radeon VII GPU is the world's first 7nm unit for consumers


(Ace Dragon) #1


While it’s not certain yet if the performance can match Nvidia’s new RTX 2080 chip, it will give you double the memory for a lower price. In addition, it gives a bit of a performance boost over the Vega 64 (so it should give good performance for Blender’s Eevee engine as soon as the company learns to make most robust drivers).

At the least, it is good to see AMD is still able to push GPU tech. in some areas.


(Stefan Werner) #2

1 TB/s memory bandwidth is neat, I want to see how that translates to performance. In many GPU applications, memory access is the bottleneck.


(BigBlend) #3

Everyone says it’s expensive but for 3D works it’s cheap. For gaming I can’t see why anyone would buy RTX 2080 or this card just to play a few games.


Is Radeon pro duo 32GB worth it for cycles?
(razin) #4


(Ace Dragon) #5

If these benchmarks hold or rise once the drivers mature, then it could indeed be the best GPU you can buy for content creation (in the sub 1000 dollar range).

That is assuming AMD can back this up with good drivers, bad software can be the breaking point for even the best hardware.

Still, that is not preventing gamers on sites like wccftech from becoming broken records, they are again trotting out their forecasts of AMD going bankrupt or at least going into major financial pain (just like last year and the year before). The fact that AdoredTV spun the hype machine into overdrive didn’t help either (but that is not AMD’s doing).


(razin) #6

does nvidia talked about the rumored 11xx series ? because that series will compete with the radeon 7, the rtx line i see it mostly for real time ray tracing & for ray tracing acceleration and for ai not for content creators that use adobe/davinci & other gpu accelerated stuff that doesn’t need raytracing


(Tomáš Luža) #7

16 GB VRAM? :open_mouth:


(Ace Dragon) #8

Jensen Huang of Nvidia responds and trash-talks the new Radeon.

Either he is acting just like his fans in the gaming community, or the Nvidia fans in the community got their tribal behavior from him. It gives a good reason to abandon Nvidia if not for their solid drivers.

The comments also have a very interesting bit, Lisa Su of AMD is actually his niece.


#9

What a weak, hubristic character. That’s just primitive & an exemplary behavior of being afraid…

Injured wild animals are especially fearful because they know that they have less chance of …

&

IMHO, a normal person would be honored to finally meet somewhat decent adversary (especially, if opposing is really a close relative). I think he’s taking it personally, showing his muddy waters - hurt ego & shallow pride. Maybe it’s coming his time to leave… to live and let live :expressionless:

Hope for more happy times ahead :relieved:


(BeerBaron) #10

I’d agree that it’s a bit unbecoming for a CEO to trash-talk like that, but what he says is 100% true. This is a 7nm chip (much better process) with 16GB of expensive HBM (more memory bandwidth), 30% higher FLOPS, requiring almost 50% more power and still it barely beats the 2080, at the same suggested retail price, months after its launch. Imagine what NVIDIA could do with 7nm, almost 300W TDP and 1GB/s HBM.

This product just doesn’t make any sense for gamers who can’t make use of that much memory. It makes more sense for content creators, but that’s not such a big market and then the lack of CUDA support makes it less attractive.


(Ace Dragon) #11

It depends on whether the supposed technologies that were supposed to supercharge the original Vega cards actually get their debut when the driver updates start coming out. It also depends on whether or not AMD’s “fine wine” approach actually works this time (which didn’t do much for the original Vegas).

At the least, AMD is confirming their promise to gamers that they will now be releasing new cards on a more regular basis (ie. not forcing people to wait until they get the Navi units released).


(mathieu) #12

Well, TSMC said 7nm would be either 25% more perf. for the same TDP or 50% less energy for the same perf. Just like for the Vega64, they balanced toward performance. The Vega 64 with 5% less frequency can work with 0.9mV and be as efficient as the 1080Ti (MRays/Watt in Cycles, at least, without my optimizations…).
So I guess the RadeonVII with the right undervolting can become an efficiency king if TSMC tells true about it’s process. And the 16GB are great for rendering.
And if the gamers and miners don’t like it, the price will be better for us :slight_smile:
A lot of ifs, but let’s see…


(Ace Dragon) #13

Guru3d confirms that AMD has a series of benchmarks in their footnotes.

The majority of the benches fall between 20 percent and 35 percent, but there’s a surprising 68 percent FPS jump for Fallout76. I’m not sure what’s in Bethesda’s code that allows that, but it could give a hint that these numbers could get better.


(Grzesiek) #14

Very eager to get my hands on one (at least one) of them. Just like Vega 56/64, the TDP peeks are never reached in Rendering. So for Blender this will be awesome.

Agreed that for gaming, well they had to do something. NAVI simply wasn’t ready, so this is a “stop gap” measure.

One site mentioend that there will be “only 5000” units on sale… and that each untill will loose AMD money, which is understandable. But for rendering… the extra 30%, is a nice boost.

Still eager to see though what RTX and its RTX features will bring for Blender.

Lastly, I wonder how Nvidia CEO will respond when they release a the “rumoured” GTX 1180… (no RTX, means I Hope singnificantly lower price…)

Still back to Vega 2, VII, Radeon VII… ahh these names. Wonder if they will switch to the R5 R7 R9 naming scheme from the past just with roman numerals? Radeon V Radeon VII Radeon IX… good old days i guess hehehe.

Still, saving my cash for 1 (or 2) of these, and hope that someone releases water blocks for them :slight_smile:


(Grzesiek) #15

Reviews are out:



(Grzesiek) #16

Direct Blender comparison, sight but noticeable boost. From 110s to 95s … LuxRender seen a significant boost. Almost 60% improvements… hmm

https://www.pcper.com/reviews/Graphics-Cards/AMD-Radeon-VII-Review-Supercharged-Vega/Content-Creation-Power-and-Temps-Fina


(bernardothecat) #17

Hmmm. I’m curious if that’s a driver issue? If not: Jeroen Bakker starts work at the Blender Institute this week, welcome! He will first focus on Cycles OpenCL optimizations, funded by AMD.


(Grzesiek) #18

Indeed, good point… they only look at final number, so the kernel compile is part of that count.

But their results compare all three latest VEGA variants, so they are all equally impacted by the kernel compile.

So if Jeroen manages to find speed improvements, both will be equally impacted. Lets say that he manages to reduce compile by 15 seconds.

That would be 80s vs 95s… … still about 19% improvement… nothing close to the 50% improvement in LuxRender. There simply has to be more work done on OpneCL Cycles


(nur rakhman) #19

as for gaming purpose i have one suggestion for amd i think they need to increase the number of ROPs of their top tier card.Ex: from 64 to 96 or even more to get better performance


(Grzesiek) #20

Yup, but that will be for NAVI (I hope)

GPU wise, Vega2 is an interesting beast. about 25% boost in performance with reduced CU count (60 from max 64). But all the issues many reviewers have experienced… surpised by that.

And in the end price/performance on slighly lower then RTX 2080 (win some/loose some)

At this point I’m considering just getting 2 Vega56 and undervolt them a tad and get better overall rendering performance with similar power envelope at a reduced price…