Vega/Navi Rumors (Updated)

[T1beriu]

I did some googling and found a video of Prey at 4K Ultra settings

1. There's no ULTRA preset in Prey. Bad points for Lisa there.

2. It says it's running on Ultra, but the preset it's not even set on the best one - Very High.

3. AA is turned off. It should be SMAA T2X. That video is junk.

At least know roughly know what two Vega's can do!

Uhhh... what? Please tell me "roughly" what two Vegas can do. I seem to have misplaced my bionic eyes. Or "roughly" means it can play Prey?

 

[guskline]

I agree with T1beriu on that Vega Prey demo. Where were the benchmarks? MIA! Sad. I bet Dr. Su was steaming. Raja has some serious explaining to do.

 

[xpea]

so lets summarize this another underwhelming event:
- needs crossfire setup to run Prey at 4k
- availability in August (Siggraph is July 30), I bet in very limited quantity, and September for any real numbers

well it remains me Polaris...

 

[xpea]

I agree with T1beriu on that Vega Prey demo. Where were the benchmarks? MIA! Sad. I bet Dr. Su was steaming. Raja has some serious explaining to do.

What says everything is that AMD had no problem to compare Vega with Titan and P100 in compute, but still keeps silent in gaming ! Sorry guys, but don't expect Vega to beat 1080Ti even in heavy favorable DX12 games...

 

[JDG1980]

It's looking like Vega is really a compute-first part in which gaming took a back seat, and pretty much the only reason they're even releasing a RX Vega gaming card at all is to save face (it would be very bad PR if they cancelled it).

I wouldn't be surprised if the use of only 4 shader engines proved to be a bottleneck (we were promised that there was better load balancing to counteract that, but this is starting to look like BS). I also wouldn't be surprised if AMD decided internally that optimizing drivers for DX11 games isn't worth the immense amount of developer time that it takes, and decided instead to go after the professional market where they can get more money for less work. If so, however, that is going to hurt them on Raven Ridge.

 

[SPBHM]

it seems unlikely that Vega would outperform the 1080 ti in gaming given the specs,
showing a CF kind of makes sense in that case, a Vega CF is likely "the fastest" when comparing to a single Titan Xp, a single Vega is probably faster than the 1080, slower than the 1080 ti.

 

[JDG1980]

it seems unlikely that Vega would outperform the 1080 ti in gaming given the specs

It should. It's a larger die (over 500mm^2, compared to 471mm^2 for GP102) and with more raw power (13 TFlops FP32, compared to about 11.3 for GTX 1080 Ti). And the implementation of tiled rendering should have closed the perf/TFlop gap with Nvidia's offerings. If it's lagging this badly, then either there are hardware bottlenecks (most likely the fact that the front end is still as starved as Fiji, with only 4 shader engines we've seen on the Linux drivers) or else the drivers suck.

Vega was supposed to fix the 4 shader engines limit, but it doesn't look like that actually happened. If the product is really subject to the same kind of bottlenecks that held Fiji back, despite 2+ years to fix them, then Raja should be fired for incompetence and so should the design team.

 

[ThatBuzzkiller]

More than ever AMD needs Microsoft to be on their side for it is their only realistic path towards salvation ...

AMD didn't dominate alone with the K8 for without Microsoft they could not have standardized the new AMD64 extension together to bury their competitors itanium architecture ...

A similar event needs to play out for their graphics products in order to even have hope to directly compete with Nvidia. AMD needs to lobby Microsoft's DirectX technology to their direction even harder ...

AMD used to be good at making high end graphics hardware or CPUs (I guess they improved for the longest while) but they were always bad at standardizing their technology like we see when Intel is practically ALWAYS the one setting standards for x86 extensions (that can't be helped since the two are hardly equals), letting Microsoft have lots of leeway with DirectX when they could've seriously capitalized harder on what we have now with D3D12 multi-threading model and switched to a newer shader bytecode format earlier and then we have HSA which failed to get any sort of traction at all ...

 

[wanderica]

What a disappointment. Didn't they specifically say we'd have more Vega information at Computex or am I taking that comment out of context? I guess an announcement for the actual launch announcement technically qualifies as "more information," but I'm not "glad I waited," Raja.

I think they'd have been better served showing nothing instead of CF Vegas with no FPS counter. When compared to Vega's supposed direct competitor at the top end, the GTX 1080 Ti, there's nothing good we can take away from that demo. I mean, even the Fury X got respectable non crossfire framerates in the above linked benchmarks. So, tell me, AMD, what should I take from your demo? I'm at a bit of a loss. I hope that's not all, and AMD will pull out one of those gotcha! moments, but I'm definitely less hyped now than I was yesterday, for sure.

 

[railven]

I agree with T1beriu on that Vega Prey demo. Where were the benchmarks? MIA! Sad. I bet Dr. Su was steaming. Raja has some serious explaining to do.

Rewatching the Stream to show the wife - that Prey Demo was absolutely terrible. Definitely not a good foot forward. Better they showed nothing and just mentioned a release date/window.

Uhhh... what? Please tell me "roughly" what two Vegas can do. I seem to have misplaced my bionic eyes. Or "roughly" means it can play Prey?

That 2x Vega can run Prey @ 4k! Can't get any rougher than that

 

[Elixer]

Basically, August is when the 'gaming' cards will be coming, and I would be curious to know where the bottleneck is.
It must be HBM2, and/or GloFlo.
I still don't think GloFlo has the capacity to be able to produce everything AMD needs.
As for HBM2, there is another player soaking up the inventory, and it could be the normal supply chain issues that has plagued AMD for a long time.

While this may be good for "back to school" sales, they are going to miss out on a ton of sales over the summer.

 

[Bacon1]

Two 580s get 50fps+ in 4k w/o AA:

https://youtu.be/1YQ3DYl4Ffg

Also vega starts here (and gaming @ 53min):

https://youtu.be/WK9z6NO1qyU?t=3041

She says ultra settings a few times, so I'd assume she means AA as well as maxed settings unlike most people who test with AA off.

Also that scene was very intense with multiple enemies and explosions, while most GPU Reviewers test w/o enemies around as it's more consistent for their results.

This guy is getting 40s with a 1080 Ti when fighting multiple enemies, maybe someone can find the same scene they used in the Vega demo for a more direct comparison?

https://youtu.be/KTnJvEL4RzA?t=1136

 

[Jackie60]

1070 performance two years after Fury x? That has to be the smallest GPU generational increase in the biggest time frame in GPU history. If it's really at 1070/1080 levels AND releasing two months late Raja should lose his job, that is a piss poor effort.

 

[SPBHM]

It should. It's a larger die (over 500mm^2, compared to 471mm^2 for GP102) and with more raw power (13 TFlops FP32, compared to about 11.3 for GTX 1080 Ti). And the implementation of tiled rendering should have closed the perf/TFlop gap with Nvidia's offerings. If it's lagging this badly, then either there are hardware bottlenecks (most likely the fact that the front end is still as starved as Fiji, with only 4 shader engines we've seen on the Linux drivers) or else the drivers suck.

Vega was supposed to fix the 4 shader engines limit, but it doesn't look like that actually happened. If the product is really subject to the same kind of bottlenecks that held Fiji back, despite 2+ years to fix them, then Raja should be fired for incompetence and so should the design team.

yes...
I'm just guessing, looking at Polaris vs Pascal, 480 vs 1060 the gap is huge in terms of gaming performance vs TFLOPS
in Prey even the 1060 3GB beats the 480, so I just don't see enough increase on those specs over the 1080 ti.
now I look at the efficiency gains AMD did from previous gens
https://www.computerbase.de/2016-08...2/#abschnitt_die_gpugenerationen_im_benchmark

and I'm not convinced, until it's out who knows, but adding to the lack of information on gaming performance and demo using CF...

 

[Jackie60]

I watched the Livestream from AMD. Heard a ton about Threadripper. Heard next to nothing about Vega. UGH!!!!!!!!!!!!

Since I just sold one of my RX480s I had running in CF, I'm content to wait until the end of July to see what Vega brings to the table. I really have no choice. The short demo of 2 RX vegas (or was that Frontier Vegas) was underwhelming to say the least.

Perhaps by then the GTX1080TI will drop a bit in price???

If the performance is at that level 1080TIs might start increasing in price.

 

[ryzenmaster]

Not exactly sure what message they wanted to convey with that CF Prey demo, but I can't exactly blame people if they take it as "one vega is not enough". That and having no fps counter made this a really bad choice for a demo. I'd expect even CF Fury to hit 60fps+ here so again what's the point. Now one might say that maybe they wanted to advertise their threadripper platform and how it delivers plenty of pcie lanes to run multiple GPUs on.. but even then the message would be lost since eyes are obviously on rx vega.

Be it as it may, all I can say is that this was yet again brilliant marketing from AMD. They only need to add a bit more confusion and delay to ensure vega demand will be low enough to meet their production capacity.. brilliant.

 

[Elixer]

...
Also that scene was very intense with multiple enemies and explosions, while most GPU Reviewers test w/o enemies around as it's more consistent for their results.

This guy is getting 40s with a 1080 Ti when fighting multiple enemies, maybe someone can find the same scene they used in the Vega demo for a more direct comparison?

It was supposed to be a showcase demo, at least, that is what everyone was expecting, so, doing this particular game & scene is still baffling on why they used crossfired Vega, and why this game?

So, if this was cherry picked to showcase something then we are all not seeing what exactly was the point of this particular setup.
No FPS counter, no GPU utilization numbers, visible tearing, Vega working in crossfire and ???

 

[tamz_msc]

It was supposed to be a showcase demo, at least, that is what everyone was expecting, so, doing this particular game & scene is still baffling on why they used crossfired Vega, and why this game?

So, if this was cherry picked to showcase something then we are all not seeing what exactly was the point of this particular setup.
No FPS counter, no GPU utilization numbers, visible tearing, Vega working in crossfire and ???

Their graphics demos have been underwhelming since the Polaris E3 2016 stuff. I say just do it like NVIDIA - anounce a presentation date, walk up to the stage and launch the card/announce launch dates. No need for demos and teasers.

 

[Bacon1]

It was supposed to be a showcase demo, at least, that is what everyone was expecting, so, doing this particular game & scene is still baffling on why they used crossfired Vega, and why this game?

So, if this was cherry picked to showcase something then we are all not seeing what exactly was the point of this particular setup.
No FPS counter, no GPU utilization numbers, visible tearing, Vega working in crossfire and ???

Yeah I don't understand it much either, just trying to show that the scene was probably very demanding and it appeared to run very well. There is clear tearing since its a 60hz monitor w/o vsync/freesync active... but yeah, no idea what they are trying to show exactly since they didn't have gpu % or even fps counters up...

Their marketing dept is still failing hard .

Its clear their push since Polaris / Ryzen 5/7 has been to go for server / pro market. They got back some marketshare which is good, but now they need the high profit margin sales to make some money.

 

[Veradun]

So... "a couple" was literally "two".

 

[DarkKnightDude]

I bet HBM2 availability is the reason for any delays.

 

[Veradun]

I bet HBM2 availability is the reason for any delays.

My bet is a silicon respin

edit: we'll never know, obv

 

[Ancalagon44]

AdoredTV says he thinks the reason is driver development.

Actually quite believable in AMDs case. Remember, Vega's back end is quite different to GCNs. So, it could take AMD a while to get its gaming performance up to scratch.

 

[Udgnim]

1070 performance two years after Fury x? That has to be the smallest GPU generational increase in the biggest time frame in GPU history. If it's really at 1070/1080 levels AND releasing two months late Raja should lose his job, that is a piss poor effort.

AdoredTV says he thinks the reason is driver development.

Actually quite believable in AMDs case. Remember, Vega's back end is quite different to GCNs. So, it could take AMD a while to get its gaming performance up to scratch.

Vega release issues could end up being due to most of AMD's money & labor being focused on Ryzen / TR / Epyc

I don't remember when exactly (definitely was not this year), but I remember Lisa Su saying something along the lines of AMD being primarily a CPU company

 

[exquisitechar]

Every time Vega is shown off I lose hope that it will be a better card than the Fury X was.

Multiple knowledgeable people have suggested that Vega needs a lot of work to be put into the drivers, though. I won't mind a late July launch if AMD that is the case and AMD delivers.