[Updated with final Review] Vega FE Benchmarking by PCPER

[Agosto]

Just now, ARikozuM said:

Cinqro 

Definitely not

[Trixanity]

8 minutes ago, Hunter259 said:

Those aren't the same chips. 1060 is not the same GPU as the 1080.

Yes and 'doubling the core count on Polaris' isn't the same chip either. You can't just add cores without making a new chip, so I don't see your point. The architecture and process node is the same and the characteristics should therefore be similar and the chips don't double in power consumption from doubling the core count. So a GCN4 chip (aka Polaris) with 4096 shader cores doesn't use 400W of power. Besides they'd ease the clock speeds a bit if that was the case. Doesn't take much to decrease power consumption substantially while still getting a lot of performance from the wider design.

[The Owl]

Hmm right out of the oven (or maybe not): https://videocardz.com/70663/amd-confirms-radeon-rx-vega-is-launching-at-siggraph

 

 

[Valentyn]

33 minutes ago, The Owl said:

Hmm right out of the oven (or maybe not): https://videocardz.com/70663/amd-confirms-radeon-rx-vega-is-launching-at-siggraph

 

 

Yup AMD marketing has confirmed it.

 

[Taf the Ghost]

Part of me really hopes AMD has found a way, at the driver level, to limit mining on the top-tier cards.  The Polaris GPUs are going to end up being the "Best Selling GPUs that no one ever owned" or something like that.

[cj09beira]

21 hours ago, valdyrgramr said:

Well, the thing is RX will have just gaming drivers meaning it will have better support.  This card, on the other hand, needs really weird driver support because it's not just a gaming card yet a workstation one too.  Which is most likely why this would have performance issues on that end.

right now it seems the gaming mode is like a placeholder, i think both sides of the driver will get more focused on that part as time goes on.

10 hours ago, Trixanity said:

 

the product main focus is 3d and compute, and in that its not broken at all, gaming is just a added bonus that today just works, and i expect it to get a lot better in a month 

5 hours ago, Hunter259 said:

Yeah I see what youre saying but I just understand why they would do it. Purposely gimp the gaming performance for the sole purpose of keeping it under wraps. After all these delays and secrecy and now this bombshell of a review it makes it seem like VEGA was just a waste of resources.

it seems that what they did was focus on the pro side of the driver only, the gaming side was developed to the point of it doesn't crash. and i expect in the future for the gaming side of FE to be just as good as RX Vega, but not now.

5 hours ago, Notional said:

Definitely a bombshell for sure. And I really hope it's not representative of the RX line, because we consumers need proper competition. Just look what is happening on the CPU side now that AMD is competitive at the high end again.

 

Either way, I think AMD really lost the plot with HBM. It not only cost AMD a lot of money, but resulted in the utter failure of Fiji, and has given a lot of problems with Vega. And it was entirely unnecessary, as GDDR5+/6 is still capable.

this is one of those cases where they are taking a hit for the promise of a better future, for amd the faster they can use hbm the better as it will make it cheaper and eventually the better product over gddr, think about it, 2 hbm 2 chips can compete with 11 gddr5x chips, which probably means that when hbm 2 reaches a better economy of scale it will be the better product full stop.

it has better granularity while moving data, it has more bandwidth, it has an advantage when writing to it (might be reading, not sure) as it simply doesn't have to wait before another write (after completion of the first one)

[cj09beira]

5 minutes ago, Taf the Ghost said:

Part of me really hopes AMD has found a way, at the driver level, to limit mining on the top-tier cards.  The Polaris GPUs are going to end up being the "Best Selling GPUs that no one ever owned" or something like that.

they probably can't do anything about it:

they change the driver, they use the older one.

they modify the bios, they mod the bios to make it work again.

there is no way for them to do that, 

amd's architecture is made to have a lot of compute power, and as it seems some times that is a curse for gamers

on a better note, more and more devs are starting to use compute based effects which means perf uplifts for amd (like the fury x in doom) it will take time but its starting to happen.

(partly why amd gpus stay relevant longer)

[cj09beira]

Just now, valdyrgramr said:

Well, gaming mode on the FE was aimed towards two type of people from what I've been reading.  Content creators followed by game devs.

true, but my point is that the dev team is not focused on the gaming side right now

[cj09beira]

2 minutes ago, valdyrgramr said:

I think they're more focused on finishing 64 and 56[pro cards maybe firepros?  IDK these are the two high end pro cards named after their amount of computing units] while trying to get the RX cards out in about 2-3 weeks.

i bet there are many people at RTG spleep deprived right now trying to make all these drivers 

hopefully with the rise of amd in the cpu market they can increase the gpu budget, they need some small things ironed out.

better control of power all around the gpu 

a system to have the voltage of the gpu be individual to each card so that you aren't using more power than needed, (done to increase yields)

improve memory compression to allow smaller gpus like the 580 to need less bus width (memory bandwidth) and get lower power consumption as a result

[Agosto]

2 hours ago, Valentyn said:

Yup AMD marketing has confirmed it.

 

Told you

[Morgan Everett]

9 hours ago, ravenshrike said:

This is actually a VERY good card for game developers as it can do everything they need on the rendering side and then turn around and play the game at 1440p+ all for a thousand dollars. Grab 2, and it gives you the perfect setup for designing a game in DX12/Vulkan and then playing it.

It may be a good card for game developers, but it's not a good gaming card. More to the point, it doesn't herald good news for the remaining Vega products that are aimed at gamers. 

[Morgan Everett]

Just now, valdyrgramr said:

We only know that they will be faster, but we don't know anything other than that.  But, that's AMD's fault for not sending out test samples.

Some SKU's may be faster; other may not be. It's looking likely to be a disappointment to many, clearly.  

[NvidiaIntelAMDLoveTriangle]

I love the fanboys so much. Every day their level of stupidity reaches new heights.

What is it so difficult to understand that a GPU is a GPU. This whole excuse of it's not a gaming card is complete bullshit. Stop this bullshit.

Did all of you forget this:

?

 

[Morgan Everett]

1 minute ago, valdyrgramr said:

All of the RX ones will be faster than this, it's already been confirmed.  I would honestly wait for it to come out rather than raising or lowering your hopes.  This card performs around 1070/1080 level, it's the slowest VEGA card that can game.  This card also is going to have many driver problems considering it's a professional card that can game for content creators and developers.  Bad drivers/driver problems can cause performance issues.  So, I really wouldn't use the slowest VEGA card with driver problems as the basis of how the other cards will perform.  This card has shown really good performance in terms of pro use, but has no drivers that aid it well enough in gaming because as everyone keeps saying...that's not the point of the card.  Even if it is using the same processor/chip there are many other hw and even sw aspects that can boost a card's performance.  So, seriously just wait to see how well the RX ones perform.  I'm not saying that the RX ones will fail or be far superior either.  I'm just not going to make assumptions based off the lowest tier card of the platform with no real gaming drivers and the only thing being the same is the processor itself.

Where has this been confirmed?

 

 

[ARikozuM]

Just now, valdyrgramr said:

I really hope that is sarcasm because the only people toting it a a gaming card is one interviewer for a review company asking how it games, and some random guy benching it for gaming.  AMD themselves has addressed that this isn't a gaming card because it isn't part of the RX line.

Problem is that if this card is meant for content creators (which includes developers), this is rather lackluster if you can't run whatever engine you're using for the most performance to then be scaled down. 

[NvidiaIntelAMDLoveTriangle]

14 minutes ago, valdyrgramr said:

I really hope that is sarcasm because the only people toting it a a gaming card is one interviewer for a review company asking how it games, and some random guy benching it for gaming.  AMD themselves has addressed that this isn't a gaming card because it isn't part of the RX line.  A Lambo is the ideal family van because both are vehicles.

Dude, the difference between a "gaming" card and a workstation card is that the latter has more features or specific features dedicated for that task.

But both have the same fucking GPU. If the workstation card is going to perform good or bad in games, then the gaming one is going to perform just the same because it's the same freaking GPU.

[Morgan Everett]

Just now, valdyrgramr said:

Raja keeps saying this?  I mean he is the head engineer on the matter.

 

Where's he said this, that every RX Vega SKU will outperform the Vega FE? Link to source, please. 

[Morgan Everett]

I've seen page 7; there's nothing on there that confirms that all RX Vega SKU's will be faster than the FE. 

[Morgan Everett]

Nowhere in those tweets does he say, or even imply, that every Vega RX SKU will be faster than the Vega FE, which you claimed was "confirmed". 

 

You've yet to provide any other source that says that, either. 

[PlayStation 2]

Something honestly seems wrong here. 

[Morgan Everett]

5 minutes ago, valdyrgramr said:

He just said not to worry about gaming performance.  I did provide a source.  It had a quote in it about the speed of the FE when Raja was talking about it being fast in terms of professional use, not gaming.  Further mentioned in the source, "Koduri also mentioned in his Reddit post that the new AMD RX Vega will be fully optimized for gaming drivers and it will come out with additional goodies. Koduri refrained from telling what goodies will be available with the AMD RX Vega."  By being optimized for gaming it will perform better and be faster in terms of gaming.  Considering both Jason and Raja saying that the FE is not optimized for gaming, but the RX will be it's only logical from there to realize that the RX cards will perform better/faster in terms of gaming only.  By how much?  We don't know?!  That could be very little to a good bit to a lot.  So, from there we will see what happens when it launches.  I don't have high or low hopes for the card at the moment.

So, as I suspected, no such thing has been confirmed at all; and, contrary to your suggestion, it's not "only logical from there" to draw an inference that all Vega RX cards will perform better. 

 

Like I said: some Vega RX cards may perform better; some may perform worse. Most likely, Vega's going to disappoint many. 

[leadeater]

34 minutes ago, NvidiaIntelAMDLoveTriangle said:

I love the fanboys so much. Every day their level of stupidity reaches new heights.

What is it so difficult to understand that a GPU is a GPU. This whole excuse of it's not a gaming card is complete bullshit. Stop this bullshit.

Did all of you forget this:

 

23 minutes ago, NvidiaIntelAMDLoveTriangle said:

Dude, the difference between a "gaming" card and a workstation card is that the latter has more features or specific features dedicated for that task.

But both have the same fucking GPU. If the workstation card is going to perform good or bad in games, then the gaming one is going to perform just the same because it's the same freaking GPU.

I would be careful in using Nvidia as an example that applies to all and not just themselves, just wait for the Vega RX gaming cards and then you won't have to.

[Delicieuxz]

It sounds to me as though there will be at least two higher-tier Vega Pro GPUs talked about at Siggraph:

 

 

"... and we'll also be talking about our more higher-end capabilities within the professional line with Vega for our professional Radeon Pro offerings, as well." - AMD dude, PCPer video, June 26th

 

~20:10

 

[Taf the Ghost]

RX Vega and Vega FE should share a die (along with any cut-down Vega SKUs), but that's about all we know right now. We don't if the HBM2 is clocked faster, what the current drivers look like, why they've had to build new ones (some Raja has mentioned) or what other aspects will be with the cards.

 

The compute power should put it around the 1080 Ti class, which is where I would expect the top SKU to land, though probably for something like 599. We know how Nvidia builds their main GTX cards, with the Titan X being the "full" version of the 1080, but we don't know if AMD is taking a "base Die + addons" approach comparatively. Given that AMD has Infinity Fabric, they could easily have some extra technology attached to the Gaming dies that aren't on the FE or Instinct models.  We just don't know yet.

 

Oh, and there will likely be a 56 CU lower-tier SKU.

[f22luke]

I'm bummed that we didn't see any comparisons between different test platforms.  I want to see if it works any better on Ryzen with the Infinity Fabric than on an Intel chip.  Does anybody know of anybody that has tested this yet?