Vega Rumors

SighTurtle said:
Sources? Links? Are you the source? Sorry, I don't know who you are.


Die size readjusted from 560mm after GamerNexus did their measurement.
Click to expand...

There must be quite a bit of transistor bloat that doesn't help gaming performance for it to be 512mm2. 35mm2 bigger than GP102, currently 40% slower, and consumes 50+ more watts.

OUCH.
 
_mockingbird said:
RTG is apparently pretty bitter about reviews of Radeon Vega FE that have been published and is considering denying sample units of Radeon RX Vega to sites that published reviews of Radeon Vega FE.

This is not a done deal and may not happen, but is currently under consideration.

Note: RTG functions as a semi autonomous group inside AMD akin to Hong Kong so whatever happens inside RTG doesn't apply to the rest of AMD.
Click to expand...

Just wondering, exactly where does this info come from?
 
Raja responded to PCPer's amended die size measurement (now 512mm²) by saying the actual size is the closest perfect square.

I just spent 10 minutes of my life trying to convince a random guy on the internet that 484 is not, in fact, closer to 512 than 529.

Still doesn't believe me, insists it's 484.

Edit: just saw post above, seems like it was in reference to 500mm figure?
 
razor1 said:


GN review

http://www.gamersnexus.net/hwreviews/2973-amd-vega-frontier-edition-reviewed-too-soon-to-call
Click to expand...


Performance is basically half that of OCed GP102 @ 4K. Can't beat a GTX 1080 FE and draws more power than a Titan Xp during gaming.

amd-vega-fe-power-grw.png


vega-fe-doom-4k.png


vega-fe-grw-4k.png


vega-fe-forhonor-4k.png

Interesting choice of GN to remove / not include GTX 1080 FE @ power consumption chart.
 
Last edited:
-Sweeper_ said:
Performance is basically half that of OCed GP102 @ 4K. Can't beat a GTX 1080 FE and draws more power than a Titan Xp during gaming.

amd-vega-fe-power-grw.png


vega-fe-doom-4k.png


vega-fe-grw-4k.png


vega-fe-forhonor-4k.png

Interesting choice of GN to remove / not include GTX 1080 FE @ power consumption chart.
Click to expand...

I found this curious as well. Would have much rather seen Fury and 1080 cards on all of the charts than a bunch of 1080Ti cards and Titan OC numbers.
 
Ieldra said:
Raja responded to PCPer's amended die size measurement (now 512mm²) by saying the actual size is the closest perfect square.

I just spent 10 minutes of my life trying to convince a random guy on the internet that 484 is not, in fact, closer to 512 than 529.

Still doesn't believe me, insists it's 484.

Edit: just saw post above, seems like it was in reference to 500mm figure?
Click to expand...

Ouch, well at least it was only 10 minutes. But ya, I think Raja meant it in relation to the 500mm figure. Unless he was lying when he said it was smaller than 500mm, and in fact it is bigger than 500mm and therefore it's 529mm. Its not impossible for that to happen, except why lie about die size in the first place then? Unless he meant "closer" rather than "smaller" than 500mm. That's not impossible either. His reply in the tweet could be meant to correct that statement in that way. I guess.
 
SighTurtle said:
Ouch, well at least it was only 10 minutes. But ya, I think Raja meant it in relation to the 500mm figure. Unless he was lying when he said it was smaller than 500mm, and in fact it is bigger than 500mm and therefore it's 529mm. Its not impossible for that to happen, except why lie about die size in the first place then? Unless he meant "closer" rather than "smaller" than 500mm. That's not impossible either. Unless his reply in the tweet indeed was also meant to correct that statement in some way. I guess.
Click to expand...

Because AMD can't answer a single thing clearly.
 
Anyone what to do a mm^2 per performance or transistors per performance?

I'm pretty sure its pretty close to Polaris.
 
Ocellaris said:
Because AMD can't answer a single thing clearly.
Click to expand...

Well while Raja is head of RTG, I would suggest it's simply just a miscommunication on his part.

I found a article dating back to CES where they report Raja states the die size is below 500mm but I can't find anything to support that statement being made.
 
Look I have been criticizing amd here but I do think that MSI marketing guy and amd guy in Twitter repeadetly saying don't compare FE to gaming rx vega over and over must have some credibility. Looks to me they have some features disabled in the FE edition, may be they won't release full fledged gaming driver for another month with all the features baked in.

But again I leave it up to amd to disappoint but I do expect that rx Vega to be faster than FE. Looks to me like FE has half assed gaming support.

Also I think if you are a professional website and AMD has said please wait for rx Vega for gaming then as a courtesy they should but I know people are looking hard for answers and gamer nexus and pc perspective are going all out benchmarking. I highly doubt [H] will do this because they will call a company out but not until they have given them the benefit of the doubt. I expect [H] to fully go all out on rx Vega and criticize it if it's a turd and amd bullshitted about FE vs Rx Vega. But I don't blame all these websites trying to get hits.
 
Last edited:
Even 484mm2 is still more than double Polaris size, with less SPs that a double Polaris would contain. Which is rather curious.
 
Snowdog said:
Even 484mm2 is still more than double Polaris size, with less SPs that a double Polaris would contain. Which is rather curious.
Click to expand...

This.

It's weird as hell.

But basically I figure RTG absolutely jumped the shark with this particular prosumerish product. As much I want to (finally) see an RX Vega put through it's paces I've never sat here hoping for a windfall either.

It's Just this card we're looking at is some aborted offspring of marketing and engineers who got into a bag of MDMA one night or something and couldn't stop when they should have.

Might be time to clean house and hire some people who know how to launch a GPU.
 
harmattan said:
That was my take as well: the guy can't give a single straight answer. We all can do the math, but does he seriously think being coy is a good idea at this point?
Click to expand...

Having dealt with developers in robotics that are located off-shore, partially its a language barrier issue, combined with the fact they're trying to slide around the truth. The truth doesn't come out till they're halfway through the project and realize they can't deliver, however they're not going to tell you that upfront because they like having a job...
 
Pascal had minimal regression in IPC too in order to maximize frequency. Maybe NCU was also redesigned to finally overcome GCN frequency wall.
 
Osjur said:
From 3DCenter: Fury X @ stock vs Vega @ 1052 / 800mhz. CPU is different but it doesn't really matter that much in superposition. That's a quite big regression in IPC compared to Fury X if you ask me. I guess we can rename NCU to SCU (slow compute unit).

VQ3OVEl.jpg

fcYuYBJ.jpg
Click to expand...

Why would they underclock the memory as well? Fury x has more bandwidth than Vega meaning they should actually overclock it so it matches fury x to do a one to one comparison.
 
well they want to compare IPC so bandwidth part probably not that important.
 
razor1 said:
well they want to compare IPC so bandwidth part probably not that important.
Click to expand...
I dont know if HBM will play a significant role but I do know at extreme settings in that benchmark memory bandwidth does play a role especially with HBM it should. But regardless I think AMD just did a bad thing launching this card and they should have enabled game mode after rx vega was launched. I do believe these guys are very incompetent and their driver team is no where near good as nvidias and they should really get their shit together. They made a mistake making a half ass attempt at launch. Launching a product that is not for gamers but has a half baked game mode and that is not certified for professionals. Its like a fucking titan xp not advertised for gamers but does a little better in pro software but then you have quadro cards. IDK I do think that they are having some driver issues that they are working on other wise I see no reason to not have RX vega already even if its a turd. Just take a loss and release it, but looking at FE where its all over the place it does seem probable that they have their shit cut out for them and raja does wish this was based on polaris lol..
 
NKD said:
I dont know if HBM will play a significant role but I do know at extreme settings in that benchmark memory bandwidth does play a role especially with HBM it should. But regardless I think AMD just did a bad thing launching this card and they should have enabled game mode after rx vega was launched. I do believe these guys are very incompetent and their driver team is no where near good as nvidias and they should really get their shit together. They made a mistake making a half ass attempt at launch. Launching a product that is not for gamers but has a half baked game mode and that is not certified for professionals. Its like a fucking titan xp not advertised for gamers but does a little better in pro software but then you have quadro cards. IDK I do think that they are having some driver issues that they are working on other wise I see no reason to not have RX vega already even if its a turd. Just take a loss and release it, but looking at FE where its all over the place it does seem probable that they have their shit cut out for them and raja does wish this was based on polaris lol..
Click to expand...


Not sure about that specific bench.

These are the guys that had DX12 drivers up and going well before nV and it took nV 1 year to catch up on their DX12 drivers (well relatively catch up I mean) to AMD.......

We have heard these excuses way too many times to be hood winked again.

Don't tell me they aren't good now. After all the talk about how much superior AMD's DX12 drivers were. Can't have it both ways......

I expect some performance differences between RX VEGA and VEGA FE but just can't imagine any more than 10% on average. So if they are still trying to get functionality working after forking the base code, god help them that means they are so far behind in drivers one month isn't going to help them at all.

PS if Firestrike shows us anything, that program should be out of the box optimized for anyways lol, regardless right in the base code of the driver.... AMD is well aware of that program and driver team should be very familiar what they have to do to optimize for it should be a piece of cake.

And talking about functionality of the Silicon and having features turned on and off. ANY GOOD PROGRAMMER knows never do optimizations till you have finished all functionality of the application.
 
Last edited:
NKD said:
I dont know if HBM will play a significant role but I do know at extreme settings in that benchmark memory bandwidth does play a role especially with HBM it should. But regardless I think AMD just did a bad thing launching this card and they should have enabled game mode after rx vega was launched. I do believe these guys are very incompetent and their driver team is no where near good as nvidias and they should really get their shit together. They made a mistake making a half ass attempt at launch. Launching a product that is not for gamers but has a half baked game mode and that is not certified for professionals. Its like a fucking titan xp not advertised for gamers but does a little better in pro software but then you have quadro cards. IDK I do think that they are having some driver issues that they are working on other wise I see no reason to not have RX vega already even if its a turd. Just take a loss and release it, but looking at FE where its all over the place it does seem probable that they have their shit cut out for them and raja does wish this was based on polaris lol..
Click to expand...

Might have something to do with the 1H 2017 deadline. Saw the same with Ryzen, release before things were settled and ready. The hope here as with Ryzen is AMD being able to improve things, like BIOS updates, driver updates, etc, etc.
 
SighTurtle said:
Might have something to do with the 1H 2017 deadline. Saw the same with Ryzen, release before things were settled and ready. The hope here as with Ryzen is AMD being able to improve things, like BIOS updates, driver updates, etc, etc.
Click to expand...


Still had 1 year to get the drivers ready, that should be more than enough time.....
 
razor1 said:
Still had 1 year to get the drivers ready, that should be more than enough time.....
Click to expand...

Not to mention they showed Doom running on Vega six months ago, and performance today is relatively the same as it was then. So if they haven't managed any improvements in six months, an extra month won't bring a miracle.
 
Last edited:
SighTurtle said:
Is the HBM shortage/difficulty still the primary guess why Vega is late?
Click to expand...

probably, I just can't see 1 year from tape out the drivers aren't a launch point. Doesn't make sense. We saw 6 month turn arounds with previous AMD and nV products, so in all honesty, its just more excuses from AMD.
 
True Elios, we don't know anything about their 14nm yields of this size of a chip
 
razor1 said:
so who was sleeping? Sounds like some one was high lol.
Click to expand...
I
think raja is probably over involved in marketing of Vega as well. This guy went on stage and said if you tune in until the end we have something special for you about Vega. Then he announced the name rx Vega, I had the biggest WTF moment ever. Since analyst day it seems he has been more quiet looks like Lisa might be shaking up a few things in the GPU department and shutting some people up. I haven't seen Raja tweet too much or hype things since analyst day. If Navi is a turd I think he will be show the door. But I think Navi is actually where they finally dump anything it has to do with GCN.
 
Last edited:
Well they all were involved with the marketing, don't think Raja would have come up with the Volta thing though that sounds like Roy Taylor, something he would say. Raja isn't the type of person to put down the competition. He tries to stay away from talking about the competition if at all possible or direct comparisons when they aren't warranted.
 
You must log in or register to reply here.
Back
Top