• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

Radeon RX Vega thread

So it's true - 300W TDP for slower one (1500MHz) and previous leak had water cooled one at 375W
Yeah, 300W for Instinct MI25 running at 1500MHz in these specs seems a bit on a higher side. I actually expected the 1,5GHz one to be able to fit into 250W because if it will really compete with GTX1080 then this on is only 180W - they are closing on needing twice as much power for the same performance with GCN...

And at 1500MHz it has whopping 0,3TF more than full GP102 Titan officially has.
0,150 more actually as Titan Xp sustained boost FP32 is 12150 GFlops.

This makes even Fury X look like successful design - at least Fury X could outperform reference 980ti in few games.
Looks like Vega is going to struggle even in Doom or Sniper Elite to match 1080ti just going by pure TF number.
I'm sure that there will be some good wins for Vega - although Doom being not very demanding from the start and limited on 200 fps is fast becoming obsolete as a benchmark.
 
One editor from Tom's Hardware seems to know the TPD for consumer Vegas, including the recommendation for board partners not to design air cooled solutions.
Format_C / Tom's Hardware Editor said:
Ich kenne mittlerweile auch die TDP (wurde ja auch hier schon erwähnt), die den Boardpartnern für deren Kühllösungen mitgeteilt wurde und die so definitiv stimmen soll. Einschließlich der Empfehlung, möglichst keine luftgekühlten Karten zu bauen. Trotzdem wird es diese geben, das ist amtlich.
3DCenter Forum - AMD/ATI - VEGA (Vega10, Vega11, Vega20) - 2017 - Seite 317

Rough translation:
I know now the TDP (was already mentioned here), which was told to the boardpartners for their cooling solutions. Including the recommendation to not build air-cooled cards. Nevertheless, there will be air-cooled solutions.
 
Yeah, 300W for Instinct MI25 running at 1500MHz in these specs seems a bit on a higher side. I actually expected the 1,5GHz one to be able to fit into 250W because if it will really compete with GTX1080 then this on is only 180W - they are closing on needing twice as much power for the same performance with GCN...

I need to find the posts from people who were saying Vega was going to be magical and not consume power like Polaris did. GCN is a giant power hog compared to Pascal. I wouldn't be surprised if Vega's power use relative to performance is exactly the same as Polaris despite being on TSMC instead of GloFo.
 
^ Ouch.

I just sold my RX470 for a nice profit [riding the wave of bitcoin miners getting every card on the market in my territory], and while I am content to wait to see how will Vega end up being, I think that 1070 will remain one of the strong picks for me in a month or two.

Do we know when will Vega arrive to sales? Early Aug?
 
Expecting retailers to do Vega + i7 + custom CPU/GPU liquid cooling kit bundles

And maybe even throw in an air conditioner
 
I need to find the posts from people who were saying Vega was going to be magical and not consume power like Polaris did. GCN is a giant power hog compared to Pascal. I wouldn't be surprised if Vega's power use relative to performance is exactly the same as Polaris despite being on TSMC instead of GloFo.
There is nothing magical about being (way) more efficient than Polaris.
Vega will definitely be more efficient than Polaris, at least for the same absolute performance per watt.
 
Speaking of Vega there was a Tweet somewhere about a slight delay again, what was it about?
Something along the lines of short term delay, long term gains.
 
Speaking of Vega there was a Tweet somewhere about a slight delay again, what was it about?
Something along the lines of short term delay, long term gains.
It was from Charlie:
AMD made the right call to delay Vega, trust me here. They took a short term hit for long term gain and credibility. It took guts to do.
https://twitter.com/CDemerjian/status/872770113228992512

Honestly I don't think AMD made a difficult decision.
You either launch disastrous or you launch later, both is bad, but the first one definitely worse.

It seems more like Vega will underwhelm and is coming later but at least in some capacity.
 
It seems more like Vega will underwhelm and is coming later but at least in some capacity.

With all the delays I can't help but feel sorry for the red team, it's the complete opposite of their CPU segment.
I don't know how many delays we've had so far but it's probably around 8 or so.
 
It was from Charlie:

https://twitter.com/CDemerjian/status/872770113228992512

Honestly I don't think AMD made a difficult decision.
You either launch disastrous or you launch later, both is bad, but the first one definitely worse.

It seems more like Vega will underwhelm and is coming later but at least in some capacity.

At this rate they're seriously going to be launching against Nvidia's 11XX/20XX line, which will be the third option, and most disastrous of all.
 
At this rate they're seriously going to be launching against Nvidia's 11XX/20XX line, which will be the third option, and most disastrous of all.
This is looking more likely with every delay. It looks like Nvidia is coming out with new hardware come this Fall which doesn't leave much room for Vega. It is also likely that the next gen Nvidia will outperform it and beat it in performance/watt.
 
Definitely seems to be shaping up to be a disappointment... I really hope it turns out well for them. Competition was much needed in the CPU space, as much as the GPU space. Another reason I hope this succeeds in performance is that these cards could very well be the foundation of what goes into next gen consoles!
 
This is looking more likely with every delay. It looks like Nvidia is coming out with new hardware come this Fall which doesn't leave much room for Vega. It is also likely that the next gen Nvidia will outperform it and beat it in performance/watt.

It is a disaster, AMD should have been on Navi already to match/exceed nVidia.
 
AMD unveiled the full specs of the MI25:
AMD-Radeon-Instinct-5-1000x562.png


FP64 is just supported with a 1/16 rate:
Radeon-Instinct-MI25-2.png

https://videocardz.com/70440/amd-announces-radeon-instinct-mi25-specifications

Aww man, the Instinct MI8 is just a rebranded Nano. Was hoping for 8GB HMB1 at least.
 
I need to find the posts from people who were saying Vega was going to be magical and not consume power like Polaris did. GCN is a giant power hog compared to Pascal. I wouldn't be surprised if Vega's power use relative to performance is exactly the same as Polaris despite being on TSMC instead of GloFo.
Aren't the 14 TF Volta cards around 250 Watts, so what's the big deal.... Water cooled Vega's will be faster and therefore have a higher TDP ceiling....

a gtx 1080 is looking like a better and better pick up this fall....
The Vega gaming cards are said to be more efficient and faster at gaming applications than the Frontier Edition and the MI cards, so how are you propping a GTX1080 when it's been in the market for so long, yet you don't even know the price of gaming Vega.. You 're not aware of the benefits of it's architectural improvements over prior GPU arch's and you have not seen real world results....Personally, I'm really excited to see what vega does with it's SSD implementation for cache and how it affects games like DOOM, games with mega textures et al...and that's just one of the benefits too...
 
Aren't the 14 TF Volta cards around 250 Watts, so what's the big deal.... Water cooled Vega's will be faster and therefore have a higher TDP ceiling....
If we go with the 50% more performance in the same power envelope which GV100 has over GP100 then you'll have ~13,5TF for a ~180W GTX1080 replacement and ~18TF for a ~250W Titan Xp replacement (and sameish for a future 1080Ti replacement). Coupled with the fact that AMD will probably need 12TF to compete with NV's 9TF in the form of GTX 1080 means that top 300W Vega will be competing with a cut down GV104 part running at ~150W. What's the big deal? GCN basically can't compete with what NV have anymore. That's the big deal.
 
You've waited this long, what's one more month, when you'll most probably get GTX 1080 performance for a 1070 pricepoint or below...
Not that I disagree, but I doubt you'll be able to get a Vega card in a month. Not to mention I doubt you'll get the nice AIB cards for a while yet.
 
Aren't the 14 TF Volta cards around 250 Watts, so what's the big deal.... Water cooled Vega's will be faster and therefore have a higher TDP ceiling....

The Vega gaming cards are said to be more efficient and faster at gaming applications than the Frontier Edition and the MI cards, so how are you propping a GTX1080 when it's been in the market for so long, yet you don't even know the price of gaming Vega.. You 're not aware of the benefits of it's architectural improvements over prior GPU arch's and you have not seen real world results....Personally, I'm really excited to see what vega does with it's SSD implementation for cache and how it affects games like DOOM, games with mega textures et al...and that's just one of the benefits too...

yeah pricing will pay a big part and i don't mean to be overly negative, just was hoping for something a bit more encouraging as i'm planning to build a new pc later this year and am really hoping for vega to a be a viable option there
 
That sounds too good to be true, but ok my friend I shall wait.

The pricing isn't too good to be true, AMD knows when they need to price down, and this is one of those instances. On the other hand, you'll probably be guzzling power and heat.
 
This is looking more likely with every delay. It looks like Nvidia is coming out with new hardware come this Fall which doesn't leave much room for Vega. It is also likely that the next gen Nvidia will outperform it and beat it in performance/watt.

Nvidia's present gen already obliterates anything AMD has in performance/watt, no one is expecting Volta to regress from Pascal there. So yes, the next-gen Nvidia will dramatically outperform it and absolutely annihilate it in performance/watt.

See, the thing about performance/watt isn't really that super important in the mainstream market. Sure, the RX 480 needs like double the power to have the same performance envelope as the 1060 but ultimately with electricity as cheap as it is, the argument is largely academic.

The real challenge with performance/watt is when you're selling GPUs that run up right against the limits of the PCI-e spec for power draw like the 1080 Ti, which will happily eat up every single bit of the 375W that PCI-e can provide over the slot + 2x 8-pin connectors. So when you have a GPU which has such a sheer performance/watt advantage like Pascal hitting the limit, what chance does GCN that has half the performance/watt have of competing with that?

There's a reason why AMD is forced to use the expensive, hard to manufacture, scarcely available HBM2 and also recommend all AIB partners use watercooling. It's the only way they can even get within the same performance galaxy as Pascal. Vega is going to be available in minuscule quantities and cost them and the AIB's a ton to manufacture. Meanwhile Nvidia is using ordinary commodity-level GDDR5X and AIB's are populating 1080 Ti's with regular old aircooling and everything is fine and power draw is fine and temperatures are fine and they are coasting to sales of millions of units of 1080 and 1080 Ti in an entirely uncontested market space without spending a ton of money on the cooling system or the VRAM.

I would not be shocked if the top Vega can barely hang with the 1080 while sucking down all 375W alloted by the PCI-e spec and requiring a Fury X-style closed loop cooler to do it. 1080 Ti is untouchable at this point by Vega because GCN probably needs 600W to have the same performance envelope as Pascal at 375W there's just no way to make a video card that consumes 600W and sell it to consumers and AMD are not crazy enough to try.
 
I'm assuming no one is insane enough to sell a video card which requires more than 2x 8-pin PCI-e power connectors.

75W PCI-e slot + (150W PCI-e 8-pin x 2) = 375W

I mean maybe AMD are insane and they'll sell a Vega video card which requires 4x 8-pin PCI-e power connectors to run. Then they could sell a video card that consumes 600W. You would probably need LN2 to cool it because the heat dissipation would be impossible even for custom loop water and a minimum 1000W power supply but it could be done. It would be the 3Dfx Voodoo 5 6000 all over again.
 
Those TDP figures are insane. Nvidia GTX 1080 has a TDP of 180 watts which is similar to the RX 580 which is an AMD mid range card. 300 watts for the entry level Vega? How do they think they can compete? Also, Nvidia is probably refreshing their GPUs early next year. With all the delays, they will a very short window to sell their GPUs competitively.. Nvidia will smoke them with Volta if not with Pascal from the look of things. It seems to be at least that Vega is going to be DOA, but I still might be wrong here.
 
Those TDP figures are insane. Nvidia GTX 1080 has a TDP of 180 watts which is similar to the RX 580 which is an AMD mid range card. 300 watts for the entry level Vega? How do they think they can compete?

1080 is a 314 mm2 card.
Big Vega is expected to be about 500 mm2. (1080Ti/Titan ballpark)

Actual gaming performance might be a problem here, power consumption is not.
 
More reports on Vega being very power hungry: MSI: ‘Damn, [RX Vega] needs a lot of power'

I'm assuming no one is insane enough to sell a video card which requires more than 2x 8-pin PCI-e power connectors.

75W PCI-e slot + (150W PCI-e 8-pin x 2) = 375W

I mean maybe AMD are insane and they'll sell a Vega video card which requires 4x 8-pin PCI-e power connectors to run. Then they could sell a video card that consumes 600W. You would probably need LN2 to cool it because the heat dissipation would be impossible even for custom loop water and a minimum 1000W power supply but it could be done. It would be the 3Dfx Voodoo 5 6000 all over again.

Unless something has changed, PCIE spec doesn't allow two 8-pin power connectors on one device, only 8+6. 8+8 can be considered out of spec already but then why stop here? You can put 8+8+8+8 and have yourself a nice card with 675W peak consumption.
 
Most if not all pascal can run at 2ghz core clocks.

As such something like Xp is 15.3 gflops and 1080ti is 14.3gflops and 1080 is 10.2gflops.

I think that's why Nvidia base flops is not indicative
 
Most if not all pascal can run at 2ghz core clocks.

As such something like Xp is 15.3 gflops and 1080ti is 14.3gflops and 1080 is 10.2gflops.

I think that's why Nvidia base flops is not indicative

Nvidia FLOPS also have a higher average game performance than the equivalent AMD FLOPS, mainly because GCN is better at compute but games aren't really compute-heavy at all.
 
Fiji was on the same process as Hawaii, yet way more efficient.

All depends on what they plan to compete with. The higher you clock a chip relatively to its projected clocks, the less power efficient it is.

Fiji was power efficient because it was clocked relatively low, not only the chip but the HBM1 memory as well. Fiji was competing with GM200 based 980Ti which was also clocked relatively low (hence the great OC headroom).

Vega 10 seems to be clocked pretty high - I don't believe that they'll be able to compete with 1080Ti this time as for that they'd need a much wider chip than just an upclocked+optimized Fiji. If they'll try to compete with 1080 and they need to clock the chip so high as to recommend board makers to use WC only on their custom boards this can mean only that they weren't expecting GP104 to be as good as it ended up being.

I think that they'll try to be noticeably faster than 1080 on average at the same price (wasn't the case with Fiji vs 980Ti) and this is why they need to clock the chip out of it's comfortable power range. This however can be easily countered by another cut down GP102 based card by NV so I don't know why they even bother, tbh. So far it looks like the lower positioned Vega 10 card (~1070'ish level) might be a much better option for consumers - as it was with both Nano and RX 470.
 
I would not be shocked if the top Vega can barely hang with the 1080 while sucking down all 375W alloted by the PCI-e spec

so you're basically expecting vega to be less power efficient as polaris.
won't say that might not be possible...it's amd after all. but doubt that's very likely.

i could start arguing, that the same people who allways insist that you can't compare flop figures between architectures are presumeing that performance/flop ratio between vega and pascal will be the same as with gcn while vega is the biggest architectual change for amd cards in over half a decade. but to be honest, im kinda glad that expectations are kinda low this time around. so there's potential for a positive surprise.
 
so you're basically expecting vega to be less power efficient as polaris.
won't say that might not be possible...it's amd after all. but doubt that's very likely.

i could start arguing, that the same people who allways insist that you can't compare flop figures between architectures are presumeing that performance/flop ratio between vega and pascal will be the same as with gcn while vega is the biggest architectual change for amd cards in over half a decade. but to be honest, im kinda glad that expectations are kinda low this time around. so there's potential for a positive surprise.

That would mean vega is barely even 30% faster than a furyx. That would be at least r600 levels of failure
 
Is Vega meant to be for the pro market or is it going to be universal?

I am talking in terms of Apple and recently released iMac and iMac Pro and GDDR5 vs. HBM2? Is HBM2 likely to be the next memory standard they use or will Polaris continue with GDDR6?
 
Is Vega meant to be for the pro market or is it going to be universal?

I am talking in terms of Apple and recently released iMac and iMac Pro and GDDR5 vs. HBM2? Is HBM2 likely to be the next memory standard they use or will Polaris continue with GDDR6?

Vega is supposed to be for the Pro and high end gamer market. A pro version of Vega, the Frontier Edition, will launch next tuesday. The RX Vega consumer cards to compete with the 1070 and 1080 will supposedly be launched at SIGGRAPH at the end of July.

The Polaris based RX 500 series will remain AMD's low to mid range cards.
 
PCWorld has a preview up on Vega FE

Close to Titan Xp performance in gaming, and higher performance in synthetics

While AMD didn’t want to reveal any gaming performance, it agreed to give us a taste of how Radeon Vega Frontier Edition performs in gaming. So we switched out the 8K Dell panel for a pair of Acer 34-inch, wide-aspect 3440×1440 panels, and AMD let us play games on both the Titan Xp and the Radeon Vega Frontier Edition.

To show it wasn’t just an API advantage, AMD let us play Doom using Vulkan, Prey using DirectX 11, and Sniper Elite 4 using DirectX 12. All of the games were set to their highest game settings, and we played at the native resolution of the panels. Although the identical panels were FreeSync-based, FreeSync was switched off on the AMD GPU.

[…]

From what we’ve seen, that concern may be misplaced. It appears to be plenty fast and, at least for the settings and the games we played, indistinguishable from the competition. Our original estimates after seeing Radeon Vega Frontier Edition with Sniper Elite 4 at Computex still hold: The cards appears to be faster than Nvidia’s GTX 1080 and close to that of a GTX 1080 Ti card.
 
Top Bottom