AMD Radeon Vega Frontier Edition 16GB Review

[falchard]

I saw a picture of a new prototype replacing the R9 Nano. Much shorter.

 

[bit_user]

The PCB is long for nothing .. they should make it shorter

You're forgetting where the blower fan goes. It needs to be upwind of the GPU and VRM. I don't see the problem with mounting it on the PCB, BTW.

Even in the case of after-market cards with down-draft fans, you're going to need enough space 2-3 big ones to blow through enough heatsink fins to dissipate all that TDP. Although, you could at least truncate the PCB. in that case.

As for water cooling, then you're right. They could do something like Fury Nano.

 

[mapesdhs]

Anyways, is it possible to include Autodesk Inventor in the suite?...

Note that unless Inventor has been radically rewritten in recent years, the best optimisation you can do for that is to use it on a platform with the highest possible clock (2700K at the budget end, 7700K at the upper end, etc.) Inventor was not optimised for multicore, and the last time I read something about this a few months ago someone said this was still the case. Beyond that, GPU-wise, I guess whatever you can afford.

16 GB is a lot, ...

Not for a workstation card it isn't. Talk to someone doing GIS, medical, etc., they'd happily have 10x that much if they could, hence why the SSG is such a nice idea, great for defense imaging and suchlike. Often wondered what could sensibly replace the Group Station, looks like SSG is the answer.

Ian.

 

[ledhead11]

16GB sounds like a lot, I mainly game, is it 8GB enough?

Most games these days have a use what you give 'em strategy so some numbers won't always matter. Generally here's what I've seen with my TI and 1080's. Game settings can have a lot to do with Vram usage as well but for many by the time you crank it all up the clocks become an issue.

1080p gaming averages 2 to 4+ GB

1440p gaming averages 3 to 6+ GB

4k gaming averages 6 to 9+ GB, MEA and ROTTR love to consume VRAM but most games hang around the 6-8 range.

The 16GB this has is totally aimed at video/3d rendering. In a heavy-load workstation environment it can be totally consumed.

 

[FormatC]

Just for interest:
I get this or next week two different waterblocks for the Vega (FE). One real good AiO and one normal block as feed for my chiller. If I find the time, I will write a short follow-up with better cooling. And - if we have even more time - also for the P6000 and a water cooler mod. It sounds interesting to see the battle of two better cooled cards.

 

[mapesdhs]

Most games these days have a use what you give 'em strategy so some numbers won't always matter. Generally here's what I've seen with my TI and 1080's. Game settings can have a lot to do with Vram usage as well but for many by the time you crank it all up the clocks become an issue.

I gobbled up way more than 3GB just with Crysis by using mega custom settings. Very long draw distance for objects, huge shadow map, etc.

If you want to see VRAM munching, checkout the OCN "best Skyrim pics" thread.

Ian.

PS. ledhead, just sent you a PM!

 

[ddferrari]

Not sure what to think here (in regards to hopes for the fx versions).

I cannot expect to see a huge % boost in games just on a driver update. Maybe 10% tops?

What bothers me the most is the heat/power and throttling concerns.
I dont think those will change with a fx version of the card.

So it would appear that 80-90% of the time Vega 64 will be running closer to 1200mhz, which is a major bummer and why it wont surpass a 1080.

Almost wish there had been a review of the watercooled version as well to see if it was able to maintain higher avg mhz.

The watercooled version is $1500, so even if it did bring Vega FE closer to the roughly $515 1080, it would be an even less flattering comparison.

 

[ddferrari]

The water cooled version of this card is able to maintain it's boost frequency, thus putting it right on par with the gtx 1080, and that is before the gaming version is even out the door with its enabled additional hardware, and better optimized drive. When third party partners get ahold of it and add there own optimizations and cooling solutions we should see a card that is very much as good or not better than the GTX 1080/1080ti and a substantial savings. Not sure what all the fuss is about.

Wow- this is a textbook example of wishful thinking, although there is one (unintentional) grain of truth in your post: AMD has yet again released an unfinished product, with the assurance that it will age like fine wine. Uh-huh.

A few optimizations and better cooling aren't going to transform a card that currently ranks between the 1070 and 1080 in DOOM / Vulkan -AMD's supposed ace in the hole- into a contender. The liquid cooled version of the top RX Vega has been officially announced at $699- so even if it matches the $515, 15-month-old 1080, that's still a failure. And this "substantial savings" you mentioned- savings compared to what?

Vega will not be in the 1080 Ti's league, no matter what tweaks are made. Period. AMD has already stated this. So their advice? Wait, of course! Wait for Navi, and never mind that Volta will be out by then.

 

[ledhead11]

Most games these days have a use what you give 'em strategy so some numbers won't always matter. Generally here's what I've seen with my TI and 1080's. Game settings can have a lot to do with Vram usage as well but for many by the time you crank it all up the clocks become an issue.

I gobbled up way more than 3GB just with Crysis by using mega custom settings. Very long draw distance for objects, huge shadow map, etc.

If you want to see VRAM munching, checkout the OCN "best Skyrim pics" thread.

Ian.

PS. ledhead, just sent you a PM!

Totally right about mod's. I actually forgot about that because I've been doing so much testing with just the base installs on games lately but I do remember my pre-SE Skyrim days. Also tinkered with some WItcher 3 mods. Never did Crysis mods since the game was so demanding to begin with.

A few months back I also jumped through many hoops to get the first Crysis running in 64bit so it would use both the larger system ram and Vram for 1440p/4k. Overall it was cool but not totally stable.

Been playing, testing Crysis 3 again(for like the 100th time) and it still can be demanding for either of my rigs in 1440p or 4k.

 

[footman]

Assuming the miners dont buy up all of the Vega stock on day one I have plans to buy a Vega 64 stock version and swap out the air cooler for an EK waterblock, however I am frustrated to hear that this card suffers from some coil whine, that will be more noticeable when the HSF is removed and replaced with a silent waterblock! Perhaps I need to wait for AIB to build their own Vega 64 PCB's....

 

[Amdlova]

i think i see this movie 2900xt

 

[sehafoc]

I liked this review, and pretty much all of your reviews, but the one thing that has me questioning this review is the choice of DX11 titles...

It seems like you're specifically thumbing the scales by choosing a title that specifically uses gameworks, specifically under MAX settings.

Why not include a Mantle title and same welp nVidia doesn't work with it so AMD does better (I realize that since Mantle was at least open and the basis for Vulcan that doesn't really make sense, but I hope you're understanding the point I'm making)?

Or better yet, if you're only selecting limited games how about choosing titles that aren't so contentious.

 

[kyotokid]

...well the only major advantage I see is more VRAM for GPU rendering. Nvidia has pretty much capped their prosumer cards at a maximum of 12 GB (Titan Xp) so they don't overshadow their expensive top end Quadro line in this department.

The one big disappointment for myself is I work with CUDA based render engines (primarily Iray) and would love 16 GB as that would handle probably 99% of the scenes I create (the 1070 maybe 75%m 1080 Ti about 85 - 90% and about to dump an extra 500$ - 600$ on a Titian XP to for just one more GB of VRAM). Pity, as to get the same amount of VRAM I would have to shell out an additional 1,500$ for a Quadro P5000.

Actually a better performance comparison for the Radeon Frontier would be Nvidia's ghastly priced (6,300$ - 7,000$ depending on vendor) Quadro GP100 as it also has 16 GB of HBM2 and similar FP16/32 floating point performance.

True the more cores/threads the better the render times, but if the file exceeds the available VRAM the process either crashes or, in the case of Iray, dumps to the CPU, and it all becomes moot.

 

[doverm]

@kyotokid

https://www.pcper.com/reviews/Graphics-Cards/Radeon-Vega-Frontier-Edition-16GB-Air-Cooled-Review/Professional-Testing-SPEC

Here you go, some comparisons with P2000/4000/5000. You ll to figure out yourself the most likely gap between both cards though.

 

[falchard]

You can use the AMD Pro Renderer now. But definitely something you would need to test for.

 

[ddferrari]

I liked this review, and pretty much all of your reviews, but the one thing that has me questioning this review is the choice of DX11 titles...

It seems like you're specifically thumbing the scales by choosing a title that specifically uses gameworks, specifically under MAX settings.

Why not include a Mantle title and same welp nVidia doesn't work with it so AMD does better (I realize that since Mantle was at least open and the basis for Vulcan that doesn't really make sense, but I hope you're understanding the point I'm making)?

Or better yet, if you're only selecting limited games how about choosing titles that aren't so contentious.

Thumbing the scales? They only tested three games in total, one of which was DX12 and then DOOM (Vulkan)- AMD's supposed bread and butter. If Vega FE had a shot at being a contender, surely DOOM or a DX12 title was its chance. The review is more than balanced, whether you like the results or not.

There are only 12 games out that use Mantle, so that would be a very misleading representation of the card's overall performance when 99.9% of all titles do not use Mantle or DX12. I'm sure they pick games based on popularity, not API, because that would definitely be biased toward AMD.

 

[bit_user]

...well the only major advantage I see is more VRAM for GPU rendering.

I wonder if Vega's cache-oriented approach to GPU memory will avoid the sort of hard limits you mention.

IMO, definitely worth looking into, before buying your next GPU for rendering work.

 

[siarheiwwe3]

I am on the market for the powerful card. This one looks good