AMD Vega Discussion Thread

[RejZoR]

such a familiar tune...

Yeah, well, none of old GPU's had so many new radical freatures implemented as Vega has.

 

[EarthDog]

Yeah, well, we will all be waiting with bated breath for these improvements using the radical new architecture/features.

 

[notb]

Yeah, well, none of old GPU's had so many new radical freatures implemented as Vega has.

And yet it usually took AMD so long to optimize the GPUs that they should have known so well from the beginning.

Honestly, the situation with Vega looks like if they'd found this GPU on an archeological exploration and are in the middle of discovering how it works.

BTW: Do you know that it took us centuries to learn how Mayan calendar worked? And it turned out to be very impressive (considering their R&D budget...), but, sadly, not very useful - we already had atomic clocks and stuff...

 

[cucker tarlson]

I was almost sure that the new and important features were disabled in Vega RX drivers. So, much more potential in next few driver updates than the usual 5-10%. Interesting times ahead then...

Possible that they raise power draw even higher without much performance gain in return as far as things go for the current games ?

 

[TheoneandonlyMrK]

And yet it usually took AMD so long to optimize the GPUs that they should have known so well from the beginning.

Honestly, the situation with Vega looks like if they'd found this GPU on an archeological exploration and are in the middle of discovering how it works.

BTW: Do you know that it took us centuries to learn how Mayan calendar worked? And it turned out to be very impressive (considering their R&D budget...), but, sadly, not very useful - we already had atomic clocks and stuff...

Ah well ,we will get there eh ,at least your not waiting on a driver then eh ,leaves more time to mooch around your threads of interest.

ill pass you some unreliable but interesting none the less Facts this pic was taken with a mild Oc on core of 4% and 960 memory clocks i think , mild as i said note the power used by each component of the gpu.
i ran it many times it comes out the same despite reboots and software updates.

And from it I infer that half Vegas issue is its memory's power draw in use ,and the memory (from samsung apparently) so usually good, is an ongoing ,evolving early product in its self with others soon to market similar tech, vega could well improve on a technical level quite easy , or not, might be the reason Navis on next gen memory.

still with all that's not enabled, and reasonable cooling i think Vega will be alright, well see , clearly.

its only unreliable because of vegas limitless issues with monitoring software , could be all out of whack for all i know but hwinfo64 is not normally wrong and what could be compared to gpuz indicated the same values.

 

[HD64G]

Possible that they raise power draw even higher without much performance gain in return as far as things go for the current games ?

Or they may reduce power draw as some features do for nVidia GPUs. Time will prove what's what.

 

[kurosagi01]

So Vega56 has hit the UK e-retailers..*launch price* starting at £379.99.

 

[turbogear]

In Germany Mindfactory had it for 409€ and it was sold out in first few minutes after Launch.
Caseking has it listed with delivery starting 4th September and price 509€.

 

[the54thvoid]

Seems even more limited than FuryX was.

 

[AlienIsGOD]

Vega prices at the canadian egg

Cant afford a 56 at those prices

 

[RejZoR]

At this point, why anyone even bothers advertising suggested prices. Why not just say, we got some new graphic cards and we have absolutely no clue how much they'll cost. So, go and buy them. The situation is insane.

 

[Eric3988]

I managed to purchase a 56 on Amazon for the actual MSRP. First time early adopter though, I don't know how normal the price gouging is. Anyways, I think the 56 is going to be the sweet spot for some time. How long? That ball is in Nvidia's court. Maybe Volta will be another fantastic performance increase like last time, but nothing is a given. The 56 will be a good high end card for 2 years or so I figure, so not a bad buy at all, if it can be had as MSRP.

 

[EarthDog]

You got one for $399? Nice!!!

 

[kurosagi01]

Personally if they can keep cost between £380-470 mark in the UK then it would be competing with the 1070.

 

[Th3pwn3r]

I managed to purchase a 56 on Amazon for the actual MSRP. First time early adopter though, I don't know how normal the price gouging is. Anyways, I think the 56 is going to be the sweet spot for some time. How long? That ball is in Nvidia's court. Maybe Volta will be another fantastic performance increase like last time, but nothing is a given. The 56 will be a good high end card for 2 years or so I figure, so not a bad buy at all, if it can be had as MSRP.

I wouldn't exactly call it a high end card personally but high end for 2 years...maybe if you AMD continues on the way they have in the GPU market.

 

[Eric3988]

I wouldn't exactly call it a high end card personally but high end for 2 years...maybe if you AMD continues on the way they have in the GPU market.

Is the 1080 TI going to be high end in 3 years. All GPUs have a life cycle, some are longer than others. I'm hoping with maturing drivers the Vegas will maintain better performance than their Pascal peers. In other news, what has me concerned is that I still dont have a shipping date in spite of ordering within minutes of the card going live. Here's to hoping it won't be a month until it's in my hands. If I have to wait that long and see an EVGA 1080 close to $500, I just might spring for that instead of the Vega. Vega is attractive because of its MSRP at it's performance, but take that away and team green pulls away in the lead...

 

[Fluffmeister]

In 3 years time Volta will have already blown the doors off everything available right now.

Out of interest, how long after launch do people think AMD drivers are acceptably up to speed?

 

[SKD007]

In 3 years time Volta will have already blown the doors off everything available right now.

Out of interest, how long after launch do people think AMD drivers are acceptably up to speed?

in my experience, it takes 3 to 5months for them to iron out issues and to start seeing improvements.

 

[the54thvoid]

The driver improvement is an irresistible myth people depend on. A few review sites have carried out the research and they show clearly, Nvidia cards do not get worse and AMD improvements are game dependent and average 3-7%.
And that's on the games out at the time. Most driver improvement is very early as technical issues are overcome. The Finewine idea is a myth and falls apart in individual games. And that wine tastes awful if it's served in an Nvidia bar.
Buy what you want for what you need it for in the moment. Hope it holds up well enough in the future but games advance and all cards fall back. Both FuryX and 980ti are doing okay but compared to 1080ti, they're way back.

 

[Vya Domus]

The driver improvement is an irresistible myth people depend on. A few review sites have carried out the research and they show clearly, Nvidia cards do not get worse and AMD improvements are game dependent and average 3-7%.
And that's on the games out at the time. Most driver improvement is very early as technical issues are overcome. The Finewine idea is a myth and falls apart in individual games. And that wine tastes awful if it's served in an Nvidia bar.
Buy what you want for what you need it for in the moment. Hope it holds up well enough in the future but games advance and all cards fall back. Both FuryX and 980ti are doing okay but compared to 1080ti, they're way back.

The idea behind finewine is misunderstood , because most GCN features span across all cards if optimizations for something like Polaris are made they will be applicable to a 7000 series card as well.

On Nvidia this is not the case , features and capabilities are not maintained entirely throughout the generations. So even though older cards do not get worse once a new generation comes along optimizations for the previous architectures are no longer the main focus.

 

[the54thvoid]

The idea behind finewine is misunderstood , because most GCN features span across all cards if optimizations for something like Polaris are made they will be applicable to a 7000 series card as well.

On Nvidia this is not the case , features and capabilities are not maintained entirely throughout the generations. So even though older cards do not get worse once a new generation comes along optimizations for the previous architectures are no longer the main focus.

But Vega isnt GCN.......

And Maxwell and Pascal are quite compatible.

Slam dunk.

 

[Vya Domus]

But Vega isnt GCN.......

And Maxwell and Pascal are quite compatible.

Slam dunk.

Nope , Vega still supports the same features previous GCN architectures had it just has extra stuff and handles some things differently. Read carefully what I wrote.

Pascal for example has improved context switching aka is more asynchronous compute capable than Maxwell which isn't at all to more precise.

They are compatible but that works only 1-way unless you want performance degradation.

No slam dunk.

The idea is Nvidia's architectures have been segmented throughout the years in terms of features and capabilities.

 

[Th3pwn3r]

So even though older cards do not get worse once a new generation comes along optimizations for the previous architectures are no longer the main focus.

Which makes sense.

 

[BiggieShady]

But Vega isnt GCN.......

She's GCN alright ... through and through.
Here is Fiji (die shot wasn't colored so I highlighted some areas for you) and Vega (fancy colored die shot):

As you can see Vega is evolution of Fiji which is evolution of Tonga.
Also, Pascal and Maxwell are basically Fermi ... yup, Fermi with ever increasing shader module hierarchy (each level of hierarchy has its own cache) and finer shader grouping for scheduler that has ever increasing multiple issuing.

Gotta listen to Kramer when building gpu architecture.

 

[Vya Domus]

Maxwell still uses static scheduling , that element alone makes it quite different from Fermi...

It seems that only now , with Volta , they finally caught up with Fermi in that aspect and exceeded it as a matter of fact.