Radeon RX Vega Discussion Thread

I'm betting that Vega will be GTX 1080 performance for $400. Willing to wager $100 on the bet.

Not happening. Vega will not be $400 AMD rather not even release it if that was the case. Yea you might Vega have series start at $399.99. No way in hell top end Vega is $400. Wanna wager that $100. I am down lol.
 
Vega should be Fury series and 390 series replacements, how they stack up to nV's cards is a different matter since nV has 4 cards (2 in performance, which is the bracket the 390x series, and 2 in Enthusiast,which are in and Ti and Titan P,) problem with this is, Titan is not something AMD can go toe to toe against unless they have the performance in respectable power consumption. So 1080ti will be something they have to get to, to actually have a decent product on the market, something that can sell. If they don't get to that performance tier, everything else is going to matter BIG time, power consumption, can't be crazy on these cards. They can't go up to 275 watts and at gtx 1080 performance, anyone purchasing that card that is a 100 watt power difference in respect to their competition, that is a 50% increase in power. That isn't small, that is going from a 750 watt power supply to over 1000 watts.

OEM's will shun the card, look Dell the third largest OEM started selling Fury's at system costs 200 bucks or something around there above systems with gtx 1080's. That tells ya the power supply and cooling they need to put into the rest of the components, they are being beefed up considerably.

So if at gtx 1080 performance levels, they need to be around 200 watts otherwise the card just won't sell anywhere. If at 1080ti performance, yeah they can go up to 275 watts no problem, but from what we have seen so far to reach the 1080ti, unlikely that will happen.

There was a Chinese marketing rumor. Saying Vega was 225w, idk we will see if it uses more than 200w and up to 250w it would have to be in dead center of 1080 and 1080ti to be acceptable.
 
There was a Chinese marketing rumor. Saying Vega was 225w, idk we will see if it uses more than 200w and up to 250w it would have to be in dead center of 1080 and 1080ti to be acceptable.

There is also an AMD Instinct slide that says less than 300 and another that says 250 TDP. which is to be believed, both are higher than any of nV's products.

And we know AMD's TDP ratings are BS.
 
Not happening. Vega will not be $400 AMD rather not even release it if that was the case. Yea you might Vega have series start at $399.99. No way in hell top end Vega is $400. Wanna wager that $100. I am down lol.
Yes, I mean the base model starts at $400. Of course there will be higher end or AIB partner cards for more. And I was serious about the bet. Let's do it.
 
Yes, I mean the base model starts at $400. Of course there will be higher end or AIB partner cards for more. And I was serious about the bet. Let's do it.

so according to you VEGA will jump straight for mediocre RX 580 performance to GTX 1080? so, it will double the performance of current offer and leave nothing to fill the gap?. they flooded the market in the 100 - 250$ with very similar RX 470/570 and RX 480/580 between 4GB/8GB and you think they will have nothing between that tier of performance and GTX 1080?. they need to fill GTX 1070 levels of performance and that's what entry VEGA should be aiming for 400$ if and only if HBM2 allow for it and that would be kinda sad IMHO.

Gratz NKD for the free 100$ you won ;)
 
There is also an AMD Instinct slide that says less than 300 and another that says 250 TDP. which is to be believed, both are higher than any of nV's products.

And we know AMD's TDP ratings are BS.

Is it too much to ask until the product is out? Instinct is passively cooled so probably running hotter than usual. Why they have it rated at <300w. It probably justifies the wattage when it comes to deep learning stuff. Vega on the other hand will have probably vapor chamber cooler for reference but I expect it to be mostly after market solutions like the rx 580 and also a water cooled version for special edition model (rumor). Likely lower tdp then instinct based card.

I mean heck Nano was pretty damn good when it came to power/performance. Amd can certainly make that happen with Vega.
 
Araxie That is a good point. AMD would be wise to have a 1070 competitor.

Knowing AMD, they'll have 5 SKUs between the RX 580 and a 1080 competitor.

I mean, look at the mess that was the variations of 470s and 480s. 4 different official SKUs all within the same 30% performance bracket. ..
 
Knowing AMD, they'll have 5 SKUs between the RX 580 and a 1080 competitor.

I mean, look at the mess that was the variations of 470s and 480s. 4 different official SKUs all within the same 30% performance bracket. ..

Well they have way too many spots and options to hit. They have no card above 2304 shaders. They could have 2900, 3500 and then full 4000 (I am just rounding it here). That could be 349.99, 449.99 and 549.99. To fill in above rx 580. This will likely be the case here.
 
Is it too much to ask until the product is out? Instinct is passively cooled so probably running hotter than usual. Why they have it rated at <300w. It probably justifies the wattage when it comes to deep learning stuff. Vega on the other hand will have probably vapor chamber cooler for reference but I expect it to be mostly after market solutions like the rx 580 and also a water cooled version for special edition model (rumor). Likely lower tdp then instinct based card.

I mean heck Nano was pretty damn good when it came to power/performance. Amd can certainly make that happen with Vega.

Not too much to ask for, but to even expect that is kind of in the realm of imagination. Realistically, we know AMD's numbers for Vega are, power, Tflops, etc. We also have seen what they have shown off for performance. The only way Vega will come out differently that what AMD has shown, is

A) AMD is showing specific things in the best light
B) AMD is sandbagging

(A) Is what AMD has a habit of doing. And any company in their position would do.

Nano was good but it was still throttling like mad with its reference cooler. It rarely reached its "supposed" tflop numbers while using in full tilt. Nano used highly binned chips for lower voltage too. Look if the rumor of a water cooled version comes out to be true, expect it to be using a butt load of power over 275 watts. There is NO need or an AIO unless power consumption is uncontrollable without it. The cost of an AIO for a card that has HBM, HBM2, isn't justified for a card going up against a card that doesn't need those types of VRAM.
 
Last edited:
Knowing AMD, they'll have 5 SKUs between the RX 580 and a 1080 competitor.

I mean, look at the mess that was the variations of 470s and 480s. 4 different official SKUs all within the same 30% performance bracket. ..

I'm guessing they have a small vega for RX 590(1070) and 590x(1080) and big vega will be Vega Fury(1080 Ti) and Vega Fury X(Titan Xp). Only thing that makes sense really, and i goes with how their line up has gone for the last few years, the polaris launch being the exception because they had no high end cards. Whether or not they'll actually meet those performance goals is another question.
 
Not too much to ask for, but to even expect that is kind of in the realm of imagination. Realistically, we know AMD's numbers for Vega are, power, Tflops, etc. We also have seen what they have shown off for performance.

Nano was good but it was still throttling like mad with its reference cooler. It rarely reached its "supposed" tflop numbers while using in full tilt. Look if the rumor of a water cooled version comes out to be true, expect it to be using a butt load of power over 275 watts. There is NO need or an AIO unless power consumption is uncontrollable without it. The cost of an AIO for a card that has HBM, HBM2, isn't justified for a card going up against a card that doesn't need those types of VRAM.

AIO is rumored to be only soecial edition. Nothing more. Full Vega won't just be AIO there will be air cooled.
 
AIO is rumored to be only soecial edition. Nothing more. Full Vega won't just be AIO there will be air cooled.

Think about it, what is the difference between an AIO and phase change vapor chamber? What is the thermal dissipation ability vs each other in the limited amount of space of 2 slots? Is is going to be so much that its going to make a huge difference, when we are seeing Polaris with water cooling doesn't do much for it because its architecture and node are hitting limits?
 
I'm guessing they have a small vega for RX 590(1070) and 590x(1080) and big vega will be Vega Fury(1080 Ti) and Vega Fury X(Titan Xp). Only thing that makes sense really, and i goes with how their line up has gone for the last few years, the polaris launch being the exception because they had no high end cards. Whether or not they'll actually meet those performance goals is another question.

As much as I like AMD, I doubt we'll see a 1080 Ti classed product.
 
Yeah, I'm not sure I see Vega beating 1080 Ti. I mean, I guess it's not impossible but hoping for 1080-level performance already seemed like a stretch. But, you never know...
 
Not too much to ask for, but to even expect that is kind of in the realm of imagination. Realistically, we know AMD's numbers for Vega are, power, Tflops, etc. We also have seen what they have shown off for performance. The only way Vega will come out differently that what AMD has shown, is

A) AMD is showing specific things in the best light
B) AMD is sandbagging

(A) Is what AMD has a habit of doing. And any company in their position would do.

Nano was good but it was still throttling like mad with its reference cooler. It rarely reached its "supposed" tflop numbers while using in full tilt. Nano used highly binned chips for lower voltage too. Look if the rumor of a water cooled version comes out to be true, expect it to be using a butt load of power over 275 watts. There is NO need or an AIO unless power consumption is uncontrollable without it. The cost of an AIO for a card that has HBM, HBM2, isn't justified for a card going up against a card that doesn't need those types of VRAM.
(C) We do not know the full potential as in fully OC, the HardOCP full ability.

The 225w number only represents stock number which is probably faster than the 1080. After that we have no clue, if it is a racer, even with guzzling gallons of fuel, it could be the top performer. For example OC speeds of 1800 or 2000. An AIO or waterblock versions made for true Ethusiast (who don't give a shit about watts but true performance) would be very cool :cool:. Maybe you're just scared it is going to pounce Nvidia rich ass :p. I rather let HardOCP put it under the test bench and let us know what they find for the real potential of the card. Stock is for pussies. Now the flip side it has nada or very little, another FuryX which actually some did get over 1100+ but more the exception than the rule kinda card.

As for Nano throttling, lol, I must have a "Golden Mermaid Nipple Sample" version. Now I am hoping for a Nano replacement since Nividia looks like they have nothing high end coming for the SFF user or multi system type user.
 
  • Like
Reactions: NKD
like this
Now I am hoping for a Nano replacement since Nividia looks like they have nothing high end coming for the SFF user or multi system type user.

The Nano really offered nothing in the SFF department. Sure, it allowed it to fit in a tiny niche of the ITX segment that wasn't covered by the 2 49%+ segments with no GPU vs GPU. But then there was other flaws. And multi system?

I doubt AMD will try the Nano approach again, it was a plain disaster both sales and functional. And there is a reason why nobody else does it. Its a contradiction product.
 
Nano has to throttle because its rating is too similar to Fury/Fury X.
It is impossible for it to sustain such rating without something giving; it has a TDP of 175W vs 275W for Fury/X and yet officially has very similar clocks and identical FP32 TFLOPs.
Just not possible with that TDP rating at sustained levels.

Here are Tom's results.

aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9RL1ovNTI0NTU1L29yaWdpbmFsLzAxLUNsb2NrLVJhdGUucG5n



TPU:

clocks-bf4-2560.gif


This will vary game-to-game though but far from consistent even to reach 950Mhz.
And going forward this will become even tricker for AMD with the more dynamic Boost-power algorithm they use (just look at the difficulty they had with official spec for Polaris models in this regard).
Fiji was pushed just has hard to its limits as Polaris was by AMD for a launch product.

Cheers
 
(C) We do not know the full potential as in fully OC, the HardOCP full ability.

The 225w number only represents stock number which is probably faster than the 1080. After that we have no clue, if it is a racer, even with guzzling gallons of fuel, it could be the top performer. For example OC speeds of 1800 or 2000. An AIO or waterblock versions made for true Ethusiast (who don't give a shit about watts but true performance) would be very cool :cool:. Maybe you're just scared it is going to pounce Nvidia rich ass :p. I rather let HardOCP put it under the test bench and let us know what they find for the real potential of the card. Stock is for pussies. Now the flip side it has nada or very little, another FuryX which actually some did get over 1100+ but more the exception than the rule kinda card.

As for Nano throttling, lol, I must have a "Golden Mermaid Nipple Sample" version. Now I am hoping for a Nano replacement since Nividia looks like they have nothing high end coming for the SFF user or multi system type user.


Maybe you do have a special nano cause every review I have read, they never were able to sustain its max clocks!

Really have to wait for H to test AMD's over clock ability?

Come on, we got how many good over clocking chips from AMD in the past 6 generations lol?

Don't we know AMD is pushing their chips TOO much, TOO high to just to get to competitive parity with nV GPU?

I can list out pretty much ever GCN chip has been like that, and its been getting worse in recent generations.
 
Its a prosumer type with 32gb of GDDR5. Not sure where you are going with this. Its not even a gaming card.
 
Yeah, so why you bothering with thread crapping this subject with a non gaming workstation polaris dual GPU that will sell for $1000 and nobody buying a gaming card will care about?
 
Yeah, so why you bothering with thread crapping this subject with a non gaming workstation polaris dual GPU that will sell for $1000 and nobody buying a gaming card will care about?

If Vega is about to launch at 12Tflops. You think a dual Polaris card with 11.45tflops is the right thing? Its quite interesting for gamers too. ;)

Feel free to argument otherwise, if possible.
 
We have only seen a 2-stack version of Vega and their slides going back several months mention Vega 10 1st with 2-stacks and 4-stacks mentioned as part Vega 10 x2 sometime 2nd half of 2017.

So can they even hit 16GB (as a single card) with only have 4GB stacks currently available?
Vega 20 mentions 4-stacks.

Might explain why they felt the need to go with Polaris in the professional/HPC segment for now, but this unfortunately limits AMD a bit.

Cheers
 
Last edited:
Everyone is talking about 1080 level performance. What about better than 1080 level performance? If the VEGA came out with just 1080 performance then AMD isn't leading in anything. AMD has to do better than to match the performance of a year old card (1080). One year later and the best AMD can do is match the performance level from a year ago? They would need to sell those cards for dirt cheap and high-end users would be better off with the "N" cards. When will AMD actually lead at the high-end?
 
Everyone is talking about 1080 level performance. What about better than 1080 level performance? If the VEGA came out with just 1080 performance then AMD isn't leading in anything. AMD has to do better than to match the performance of a year old card (1080). One year later and the best AMD can do is match the performance level from a year ago? They would need to sell those cards for dirt cheap and high-end users would be better off with the "N" cards. When will AMD actually lead at the high-end?

Never.
 
Everyone is talking about 1080 level performance. What about better than 1080 level performance? If the VEGA came out with just 1080 performance then AMD isn't leading in anything. AMD has to do better than to match the performance of a year old card (1080). One year later and the best AMD can do is match the performance level from a year ago? They would need to sell those cards for dirt cheap and high-end users would be better off with the "N" cards. When will AMD actually lead at the high-end?

last time AMD did lead anything was with the vanilla HD7970 six years ago since then, the best they can do is play the catch up game.
 
Everyone is talking about 1080 level performance. What about better than 1080 level performance? If the VEGA came out with just 1080 performance then AMD isn't leading in anything. AMD has to do better than to match the performance of a year old card (1080). One year later and the best AMD can do is match the performance level from a year ago? They would need to sell those cards for dirt cheap and high-end users would be better off with the "N" cards. When will AMD actually lead at the high-end?


To keep the card in PCI-e specs and take the lead they need to hit ~50% increase in perf/watt.... That is ALOT. To match the 1080ti they need to hit 40%, To match the gtx 1080 they need 60% increase in perf/watt.

See the mountain they have to climb?
 
To be fair.
AMD can lead with some technology development, but unfortunately before it can ever be fully utilised to its potential such as HBM that they used before its time or say Mantle (but that could be deemed a catalyst).
Nvidia seem to time their technology commitments to a more precise window for maximum leveraged use.

Cheers
 
I'm hoping that Vega is GTX1080 level in DX11 and GTX1080Ti level in DX12. That has to be the minimum IMO. Any less than that, is not great news.
 
I've seen some people suggesting that AMD won't launch at Computex beacuse they don't show up on the exhibitor list on Computex's website. AMD has indicated on their site that they will show up at Computex. Plus, AMD did a Macau Press event for Polaris that occurred just before Computex last year. So, not impossible for that to also happen again.
 
IIRC they didn't show up on Computex' website last year either for quite some time.
 
I don't think nVidia showed up either, only their manufacturer partners did. But the scene was definitely dominated by nVidia GPUs from every one (1080 and 1070 had just been released).
 
I'm hoping that Vega is GTX1080 level in DX11 and GTX1080Ti level in DX12. That has to be the minimum IMO. Any less than that, is not great news.
I'm fairly certain Vega can reach GTX 1080 level, though that is Nvidia's best from last year. If priced aggressively, say $400, I think AMD would have a winner.

However, I could still see a higher end model in the $600 range that can compete with 1080 Ti. This seems somewhat unlikely, but you never know. As you say, maybe AMD can eek out an edge in DX12/Vulkan.
 
GTX 1080 is what I would consider to be the bearest minimum Vega (talking flagship here) would need to be even remotely called a success, since anything less than that, you are looking at a 20% performance increase over Fury X as a result of both Architectural change and node shrink, while nVidia in this regard has achieved near 70% improvement (980ti to 1080ti), an very poor showing in comparison to that.

I'd wager we'll see something like a 50% increase over its old Fury X line, which puts it inbetween 1080 and 1080ti. 1080ti is ideal, but a bit of a stretch IMO.
 
Yes. If Vega is less than GTX 1080 then it would be a bust for enthusiasts (though even 1070 level performance might sell at the right price). I have to think they can reach that level if they want to stay in the game.

1080 Ti, seems unlikely but we can hope. I speculate that with Nvidia launching two top-end products within weeks of each other, they must have thought AMD was going to be competitive at that level. To lock that segment in now, versus in a month (or whenever Vega launches) with AMD having a viable alternative.
 
I'm fairly certain Vega can reach GTX 1080 level, though that is Nvidia's best from last year. If priced aggressively, say $400, I think AMD would have a winner.

However, I could still see a higher end model in the $600 range that can compete with 1080 Ti. This seems somewhat unlikely, but you never know. As you say, maybe AMD can eek out an edge in DX12/Vulkan.

I wonder if they can even price it at $400 with HBM2? Those things aren't cheap and at some point AMD has to start increasing their margins.
 
Back
Top