Vega Rumors

To be somewhat more fair, the Vega FE should be compared more to the Quandro P5000, professional card:
https://www.newegg.com/Product/Prod...NA&ignorebbr&gclid=CM_Nxtjc5dQCFQ9ahgodFHsJ0g

Basically the P5000 is the professional version of the 1080 with 16GB. No, it does not game as fast as the gaming 1080 either and cost $2500 on sell. Now the gaming performance is like 8%-10% slower then a FE 1080 and more so for an AIB OC 1080. The Vega FE is way cheaper and I would like to see the performance compared to this card in professional applications and games. Biggest difference I see with the P5000 is certified drivers for applications (biggie), Cuda applications and of course lower power requirements. If the FE Vega performs well against this card then it maybe a great buy for a number of professionals.

I will just wait to see what AMD brings to the table with their gaming cards, I probably already put my foot in my mouth earlier. Still AMD is the one that compared the Vega FE to Nvidia's fastest gaming card - go figure plus not having certified professional drivers does not help.

Gaming Vega's will have less ram - some power and heat saved there, HBM2 could be clocked higher. The max temperature for throttling could be set higher, as in over 85c of the FE a.k.a 95c allowing higher clocks. Could have more aggressive powertune options, bigger coolers, more streamlined drivers for speed or games. Plus some of the developers maybe working already to optimize some of the games for Vega. Binning can also used for faster skew versions.
 
Last edited:
I guess another thing that came of the PCPer testing was that a blower can keep the Vega cool if you crank it up. They were staying below 60 degrees when it was maxed out lol.

Although it kind of sounds like it's hitting power limits somewhere, since it still downclocked even at the lower temps.
 
Which was what I was saying all along. It's too early to tell given Raja's quotes which are a week old.

Those quote are a month old, they are from the reditt AMA, last month. and Ryan asked AMD about that and he got the answer that the drivers for FE was forked to work on the RX, so only one month of extra work for the RX drivers all other game optimizations in the past are already in the FE driver.

No such thing as releasing drivers with debug code man, wouldn't make sense on a released product, I can understand something that for specific application developers that are helping AMD create something new, that I have seen from both IHV's getting specific driver releases for testing their apps but to the general public never happens.
 
not having certified professional drivers does not help.

The card is for no one, it was a marketing exercise at a unicorn "professional" customer would wanted to spend $1000 on a card without actual professional driver certification.


Raja is just spouting crap to keep people interested until the last possible minute when they are let down. He also said the FE would deliver smooth 4K gaming right in those same quotes...
 
Last edited:
The card is for no one, it was a marketing exercise at a unicorn "professional" customer would wanted to spend $1000 on a card without actual professional driver certification.



Raja is just spouting crap to keep people interested until the last possible minute when they are let down. He also said the FE would deliver smooth 4K gaming right in those same quotes...

I'm sure it plays minecraft at 4k!
 
There are times I think AMD did better when they just made CPUs, and ATI was its own company. Maybe they should do that again?
 
Don't think they can ever do that, too much IP overlapping between the two departments now.
 
The card is for no one, it was a marketing exercise at a unicorn "professional" customer would wanted to spend $1000 on a card without actual professional driver certification.



Raja is just spouting crap to keep people interested until the last possible minute when they are let down. He also said the FE would deliver smooth 4K gaming right in those same quotes...
The card is in limbo, what professional organization that does serious development work with million dollar budgets, filming industry etc. would skip on $1500 to save some money on drivers that are not guaranteed to work, fully tested etc on their project? Now the home professional, small groups I can see this maybe a great option to save some money but yet get optimized drivers for a number of programs.

I will just give RTG BOD (Benefit Of Doubt) on the gaming cards. I hope we will see some AIB versions right up front with the better cooling etc.
 
Those quote are a month old, they are from the reditt AMA, last month. and Ryan asked AMD about that and he got the answer that the drivers for FE was forked to work on the RX, so only one month of extra work for the RX drivers all other game optimizations in the past are already in the FE driver.

No such thing as releasing drivers with debug code man, wouldn't make sense on a released product, I can understand something that for specific application developers that are helping AMD create something new, that I have seen from both IHV's getting specific driver releases for testing their apps but to the general public never happens.

Raja said:
Consumer RX will be much better optimized for all the top gaming titles and flavors of RX Vega will actually be faster than Frontier version!
RX will be fully optimized gaming drivers, as well as a few other goodies that I can't tell you about just yet....But you will like FE too, if you can't wait:)

Maybe you didn't read that part. So yes, non optimized drivers. Basically what you are working with is likely patched Fiji drivers that don't kick in extra optimizations for the extra abilities like tiling improvements implemented in Vega. Plus there could be profiling engaged, plus other things which slow things down.

But these facts were reiterated last week.

And yes, sometimes you DO put in profiling into release drivers, especially early on, for diagnostic purposes, even for customers. I know I've done it.
 
The card is for no one, it was a marketing exercise at a unicorn "professional" customer would wanted to spend $1000 on a card without actual professional driver certification.



Raja is just spouting crap to keep people interested until the last possible minute when they are let down. He also said the FE would deliver smooth 4K gaming right in those same quotes...

My thoughts exactly.
 
The card is for no one, it was a marketing exercise at a unicorn "professional" customer would wanted to spend $1000 on a card without actual professional driver certification.



Raja is just spouting crap to keep people interested until the last possible minute when they are let down. He also said the FE would deliver smooth 4K gaming right in those same quotes...

Yes, 4k but he didn't specify what detail level! /s
 
The way I see this...

The drivers are so crap that AMD is doing serious damage to themselves by releasing this so early. (I.e. hardware is good but software sucks).

or

This is their GeForce FX. Long wait and it turn out to be power hungry turd. In that case they should just leave GCN and start from scratch.

or

A bit of both. It certainly going to be power hungry and they have to squeeze every drop of power out of it. If that wouldn't be the case there wouldn't be water cooled variants available.

There are quite a bit of warning signs... The way they handled VEGA FE launch is nothing like they did with Ryzen. They clearly aren't comfortable showing real game benchmarks. I'd be extremely surprised if the gaming VEGA would be close to 1080 Ti in performance. Things just don't look very promising at all currently.
 
The way I see this...

The drivers are so crap that AMD is doing serious damage to themselves by releasing this so early. (I.e. hardware is good but software sucks).

or

This is their GeForce FX. Long wait and it turn out to be power hungry turd. In that case they should just leave GCN and start from scratch.

or

A bit of both. It certainly going to be power hungry and they have to squeeze every drop of power out of it. If that wouldn't be the case there wouldn't be water cooled variants available.

There are quite a bit of warning signs... The way they handled VEGA FE launch is nothing like they did with Ryzen. They clearly aren't comfortable showing real game benchmarks. I'd be extremely surprised if the gaming VEGA would be close to 1080 Ti in performance. Things just don't look very promising at all currently.

I thought buying a $1800 product with shitty drivers was a positive! Might blue screen now but imagine, in six months when it's outdated by Volta it'll be golden!!!!!
 
I thought buying a $1800 product with shitty drivers was a positive! Might blue screen now but imagine, in six months when it's outdated by Volta it'll be golden!!!!!

If you want to you can spend any amount of money on things that does not involve that your use is what they created it for. If they did not create it for gaming why bother bitching about it ?

You have a link somewhere for Volta release in 6 months? http://fudzilla.com/news/graphics/43962-new-geforce-will-be-incremental
 
Keep making hula-hops...it won't change the crux...what is next...#WaitForNavi.

Going to enjoy your post from now...compared to the past...thanks for amusing me.

That was weird.

Thanks for contributing to the conversation, in no way does this indicate you have some fundamental need for self worth reinforcement.

I'm just curious as to how far up AMD pushed the voltage to get overall better workable yields. They have a history of doing it and killing any efficiency to be gained for a large percent of silicon.
 
That was weird.

Thanks for contributing to the conversation, in no way does this indicate you have some fundamental need for self worth reinforcement.

I'm just curious as to how far up AMD pushed the voltage to get overall better workable yields. They have a history of doing it and killing any efficiency to be gained for a large percent of silicon.

I absolutely agree. Putting aside the gaming numbers which everyone is fixated on, I'd like to dial one back and see what kind of overall efficiency it could possibly get as a mining card. This ...thing they released is too expensive for that though, need gaming cards.
 
Not weird at all...I can go to your profile and read all of your posts.

You must have a bad memory...take a look for yourself (y)

I have no idea what you're getting at beyond the fact that I'm not trying to feel smugly superior based on the fanboyism of one vendor over another?
 

"Vega has crashed twice during Blender renders so far, using pro drivers. Anyone else experience this?" from gamernexus's twitter feed...


:ROFLMAO:

AMD does have a bad launch track record... seems they resolve things after a month. Maybe if they waited that extra month to start with. :\
 
Maybe you didn't read that part. So yes, non optimized drivers. Basically what you are working with is likely patched Fiji drivers that don't kick in extra optimizations for the extra abilities like tiling improvements implemented in Vega. Plus there could be profiling engaged, plus other things which slow things down.

But these facts were reiterated last week.

And yes, sometimes you DO put in profiling into release drivers, especially early on, for diagnostic purposes, even for customers. I know I've done it.


They are optimized, it was asked and even Rys said so, and Ryan (pcper) asked. So do you realize that? Two direct confirmations from AMD that says FE has game optimizations. The driver was recently forked for the RX Vega.

There is a reason why that people posted that pic with its head in the sand.

Why do you have feel the need to give excuses for AMD's piss poor performance, when AMD themselves are saying otherwise? AMD does that pretty good on its own, like blaming microsoft for Ryzen's CCX problems, they are good at making excuses, don't worry. PR in any company will try to put out the fire.

There will be very little difference between RX Vega and Vega FE.
 
Last edited:
They are optimized, it was asked and even Rys said so, , and Ryan (pcper) asked. So do you realize that? Two direct confirmations from AMD that says FE has game optimizations.

There is a reason why that people posted that pic with its head in the sand.

Why do you have feel the need to give excuses for AMD's piss poor performance, when AMD themselves are saying otherwise?

There will be very little difference between RX Vega and Vega FE.

Well obviously the previous statement by Raja and whatever was implied by PCPer are in direct contradiction. I'll take the quote straight from Raja's mouth that gives much more detail.

And I haven't excused AMD on anything. I told you I'm not passing judgement till RX is released.

In case you forgot, I was right on your side when the RX480 & Fury were crucified for being over power spec, and for lackluster performance/specs.

I am truly neutral territory. But I do want to give a company a fair shake, no matter whom that player is.
 
Well obviously the previous statement by Raja and whatever was implied by PCPer are in direct contradiction. I'll take the quote straight from Raja's mouth that gives much more detail.

OK that means the most recent information we have now, is that Vega FE and RX Vega, will be using similar drivers for gaming, yeah I can see some differences but not much. One month of driver dev, isn't going to be much that is like what two iterations? Yeah but at the time they weren't even working on RX Vega drivers dude, they just forked the drivers. So all gaming drivers are based on Vega FE drivers too. Same base.

So how did Raja at that time, know or even conceivable understand now much faster RX Vega will be in games if they weren't really working on RX Vega drivers? Unless they were already doing game optimizations with Vega FE (which was confirmed by Ryan)
 
OK that means the most recent information we have now, is that Vega FE and RX Vega, will be using similar drivers for gaming, yeah I can see some differences but not much. One month of driver dev, isn't going to be much that is like what two iterations?

Again, my own personal experience doesn't necessarily say this.

For 10+ years this company has been releasing a product that was stable, but slow as hell. For the last year, I have been rewriting from the ground up that 10 years of experience into a new way of doing it. That code was in the background of what was released for the last 11 months. Next month it will be released. It's faster, more stable, and more flexible. But our engineers here have been testing with the release versions with diagnostics to determine the accuracy of the new methods. After all building $1,000,000+ machines which have a critical design flaw can cost you a lot of money. So you take a phased release approach.

And again, I'm not claiming that new DX drivers are going to solve all the performance woes. It might be 1%, 5%, 20%, 30%. It's hard to say without source and some detailed analysis of the internal workings.
 
  • Like
Reactions: N4CR
like this
Again, my own personal experience doesn't necessarily say this.

For 10+ years this company has been releasing a product that was stable, but slow as hell. For the last year, I have been rewriting from the ground up that 10 years of experience into a new way of doing it. That code was in the background of what was released for the last 11 months. Next month it will be released. It's faster, more stable, and more flexible. But our engineers here have been testing with the release versions with diagnostics to determine the accuracy of the new methods. After all building $1,000,000+ machines which have a critical design flaw can cost you a lot of money. So you take a phased release approach.

And again, I'm not claiming that new DX drivers are going to solve all the performance woes. It might be 1%, 5%, 20%, 30%. It's hard to say without source and some detailed analysis of the internal workings.


When was the last time you saw drivers give lets say 15% boost all around in one release?

The only time I can even think of that happening was with nV's GF3's when AMD released their 8500. So that means nV just bundled up many cycles of game optimizations in one release.

Vega's shader array isn't that far away from older GCN products either. And its front end changes are really similar to Polaris.

And btw you can see the improvements over Fiji its ~30% in most cases. Some times more. That is its clock speed increase over Fiji. Front end changes you can see the performance difference in UniEngine. They matched last gen products (nV), that is Polaris, its X 4 tris per clock is working perfectly.
 
Last edited:
When was the last time you saw drivers give lets say 15% boost all around in one release?

The only time I can even think of that happening was with nV's GF3's when AMD released their 8500. So that means nV just bundled up many cycles of game optimizations in one release.

Vega's shader array isn't that far away from older GCN products either. And its front end changes are really similar to Polaris.

Usually you only get that much boost when there was a bug in a previous release with a particular game.
 
So is Vega supposed to do tile-based rendering like Pascal and Maxwell do or not, and is it enabled with the current driver set?
 
So is Vega supposed to do tile-based rendering like Pascal and Maxwell do or not, and is it enabled with the current driver set?


PCper tested that its not active, And from the talks before its more likely then ever that its gotta be done via code.
 
So is Vega supposed to do tile-based rendering like Pascal and Maxwell do or not, and is it enabled with the current driver set?

From what we can tell, no. The average IPC is roughly equal to Fury. And we know Fury didn't have the extra hardware support. So it is suspect that some of this isn't engaged yet.

But no one knows for sure. Hence why there may be a significant performance boost. An improved tiling front end discard will save the back end ROP from a lot of overdraw.
 
Usually you only get that much boost when there was a bug in a previous release with a particular game.

Ummm NO

Sometimes it can be as simple as finding a new way to compute things. Before we had an exhaustive search of 42,000 records in a database x 16. Now we can reduce the query down to 1 due to precaching of data. That's better than an order of magnitude.

Not a bug. Just a better way of doing things.
 
Ummm NO

Sometimes it can be as simple as finding a new way to compute things. Before we had an exhaustive search of 42,000 records in a database x 16. Now we can reduce the query down to 1 due to precaching of data. That's better than an order of magnitude.

Not a bug. Just a better way of doing things.

True, but that's pretty rare for a graphics driver. If it was performing THAT poorly before particularly giving you a >15% hit to performance I'd call it a bug. That may just be semantics, but that's how I'd look at the scenario. Any decent review of the drivers during the initial optimization should have uncovered or corrected those major offenders.
 
Yeah that is a low level change in drivers, so I expect them to do that right off the bat if it was possible via drivers, but the way it was talked about before, it sounded like it had to be done via primitive shaders. So there may be nothing to do in drivers.
 
So they'll price it competitively to make it attractive to people who don't want to spend >$400 on a video card.
$350 for 1070/1080 hybrid performance.

You'll need liquid nitrogen cooling and a nuclear power supply, but hey.
 
So they'll price it competitively to make it attractive to people who don't want to spend >$400 on a video card.
$350 for 1070/1080 hybrid performance.

You'll need liquid nitrogen cooling and a nuclear power supply, but hey.

I would be VERY surprised if they could price it anywhere close to that range. HBM2 isn't cheap, and these cards will be in VERY short supply.
 
So they'll price it competitively to make it attractive to people who don't want to spend >$400 on a video card.
$350 for 1070/1080 hybrid performance.

You'll need liquid nitrogen cooling and a nuclear power supply, but hey.

A couple of things though. I'm betting that NV has a lot more room to drop the price of the 1070 and 1080, and I'm not sure it's even feasible for AMD to sell the first iteration of Vega with HBM2 for $350.
 
Back
Top