[Rumor] AMD Radeon RX Vega indefinitely delayed

[Mockingbird]

Initially, AMD has intended for the Radeon RX Vega to compete with the Geforce GTX 1080 Ti which AMD has anticipated to be a mild refresh of the Geforce GTX 1080 with 10% increase in performance.

Radeon RX Vega has a close loop design similar to the one seen on the Radeon R9 Fury X. One of the problem is that it is already running out of the box at its thermal limit with no overclocking headroom.

The other more serious problem is that Radeon RX Vega costs significantly more to make than the Geforce GTX 1080 and even the Geforce GTX 1080 Ti.

When NVIDIA announced the Geforce GTX 1080 Ti at $699 and dropped the price of the Geforce GTX 1080 to $499, it threw the monkey wrench at AMD.

The Radeon RX Vega doesn't perform well enough for a $699 price tag, but costs too much to produce such that it could be sold at $499.

AMD is stuck in the impossible position of having to somehow increase the performance to justify the price tag or cut cost enough so that it can be sold at a lower price.

As a result, Radeon RX Vega has been indefinitely delayed.



We already have a thread discussing Vega rumors and speculation. Please use it rather than starting new threads if you're just posting rumors or speculation.

AT Moderator ElFenix

 

[coercitiv]

Name a source please.

 

[Crono]

Name a source please.

This. Otherwise it's conjecture or prognostication, which I'm okay with as long as it's stated as such and posted in the appropriate thread (not a fan of clickbait-y titles, and it's a little late for April Fools).

 

[Glo.]

I am laughing my ass off right now.

Are you tech enthusiast, that really want to buy new hardware, or just fanboy/cheerleader of brands, who want to know which company has "won"?

If you are tech enthusiast, you would not posts this thread.

Whole doom and gloom over Vega is ridiculous. Thankfully, nobody so far has guessed correctly the performance of the GPU. So this job of being silent about it was good thing from AMD.

 

[sandorski]

I find this too difficult to believe. There's no way they designed a single Vega that required that high a Price just to be Marketable. I'm guessing there are at least 3 Vega Models at different Price Points, not just 1 that requires a near $700 price point to break even.

 

[Mockingbird]

I find this too difficult to believe. There's no way they designed a single Vega that required that high a Price just to be Marketable. I'm guessing there are at least 3 Vega Models at different Price Points, not just 1 that requires a near $700 price point to break even.

There're more than one Vega.

The "little" Vega is still on track to be released, but the "big" Vega that people have been waiting for has been delayed.

 

[Glo.]

There're more than one Vega.

The "little" Vega is still on track to be released, but the "big" Vega that people have been waiting for has been delayed.

Ahem, did your source have told you that there are two GPUs based on Big Vega? One is air cooled, and other is water cooled? Did your source told you also, that none of those GPUs is cut down version of big Vega?

P.S. All I have is rumors coming from retail line, and the rumors say that there are 4(!) GPUs based on big Vega GPU coming up. Two for consumer market, and two for professional market, and machine learning. MI25 is obvious one, second is replacement for Radeon PRO SSG, and the two are consumer ones.

 

[Mockingbird]

Ahem, did your source have told you that there are two GPUs based on Big Vega? One is air cooled, and other is water cooled? Did your source told you also, that none of those GPUs is cut down version of big Vega?

P.S. All I have is rumors coming from retail line, and the rumors say that there are 4(!) GPUs based on big Vega GPU coming up. Two for consumer market, and two for professional market, and machine learning. MI25 is obvious one, second is replacement for Radeon PRO SSG, and the two are consumer ones.

No, I was not.

 

[Newbian]

There're more than one Vega.

The "little" Vega is still on track to be released, but the "big" Vega that people have been waiting for has been delayed.

Provide links please or it won't end well for you.

 

[Shmee]

Rumor tags have been added to the thread, as there are no sources sited. OP, in the future, please use source tags, or state that this is your opinion.

 

[exquisitechar]

Do you think AMD is filled with morons? Who else would think the 1080ti would be a refresh of the 1080 with 10% higher performance?

What to believe: this, or that rumor from semiaccurate forums that Vega is far beyond expectations and that AMD thinks it'll gain them 50% market share? Hmm...

 

[mohit9206]

I don't know why people who are waiting for Vega don't just buy Pascal. Its quite obvious Vega won't match the power efficiency of Pascal and the secrecy around Vega probably means amd knows it has a dud on its hands. With All these red flags, its safe to assume Vega will disappoint just like these Polaris rebrands.

 

[sandorski]

I don't know why people who are waiting for Vega don't just buy Pascal. Its quite obvious Vega won't match the power efficiency of Pascal and the secrecy around Vega probably means amd knows it has a dud on its hands. With All these red flags, its safe to assume Vega will disappoint just like these Polaris rebrands.

Vega is not Polaris. Making assumptions based upon Polaris is not "safe".

 

[Glo.]

I don't know why people who are waiting for Vega don't just buy Pascal. Its quite obvious Vega won't match the power efficiency of Pascal and the secrecy around Vega probably means amd knows it has a dud on its hands. With All these red flags, its safe to assume Vega will disappoint just like these Polaris rebrands.

LOL.

Ok.

 

[richaron]

I don't know why people who are waiting for Vega don't just buy Pascal. Its quite obvious Vega won't match the power efficiency of Pascal ...

As stated above, source please or make it obvious this is another in a long list of conjecture painting AMD in a bad light.

 

[exquisitechar]

I don't know why people who are waiting for Vega don't just buy Pascal. Its quite obvious Vega won't match the power efficiency of Pascal and the secrecy around Vega probably means amd knows it has a dud on its hands. With All these red flags, its safe to assume Vega will disappoint just like these Polaris rebrands.

Disagreed. 2016 to early 2017? Buy Pascal. Right now? Vega is around a month away. For a big purchase like a high end GPU it is not much of a wait. Unless if you have so much money you don't care and will just swap Pascal with Vega if it's faster.

The Polaris refresh tells us nothing about Vega, there are no red flags. On the contrary, actually. I've seen a few people in the industry hinting that it's a good product. In any case, I doubt it will be a disaster like the OP says. GTX 1080 + 20% is my prediction of it's performance at the absolute minimum, based on tidbits of information like MI25's TFLOP rating.

 

[Glo.]

Disagreed. 2016 to early 2017? Buy Pascal. Right now? Vega is around a month away. For a big purchase like a high end GPU it is not much of a wait. Unless if you have so much money you don't care and will just swap Pascal with Vega if it's faster.

The Polaris refresh tells us nothing about Vega, there are no red flags. On the contrary, actually. I've seen a few people in the industry hinting that it's a good product. In any case, I doubt it will be a disaster like the OP says. GTX 1080 + 20% is my prediction of it's performance at the absolute minimum, based on tidbits of information like MI25's TFLOP rating.

MI25, and highest end Vega, are supposedly the same, golden Sample chip, based on retail rumors. They can clock up to 1.5 GHz. MI25 has under 300W TDP, and my rumors say that RX Vega with Liquid Cooling has 275W TDP, so this is pretty in line.

687F:C1 GPU from CompuBench leak has 1.2 GHz core clock. Make of it guys whatever you want to. However, that 687F:C1 GPU has been already demoed by AMD in Doom 4K, Vulkan, in the same settings that Techpowerup tests their GPUs. Check out LinusTechTips video on the matter, they accessed the options.

 

[mohit9206]

Name a source please.

As stated above, source please or make it obvious this is another in a long list of conjecture painting AMD in a bad light.

Raja Koduri is his source.

Vega is not Polaris. Making assumptions based upon Polaris is not "safe".

But Vega is still GCN. So its a given that power efficiency won't be close to Pascal.

Disagreed. 2016 to early 2017? Buy Pascal. Right now? Vega is around a month away. For a big purchase like a high end GPU it is not much of a wait. Unless if you have so much money you don't care and will just swap Pascal with Vega if it's faster.

The Polaris refresh tells us nothing about Vega, there are no red flags. On the contrary, actually. I've seen a few people in the industry hinting that it's a good product. In any case, I doubt it will be a disaster like the OP says. GTX 1080 + 20% is my prediction of it's performance at the absolute minimum, based on tidbits of information like MI25's TFLOP rating.

There needs to be 3 Vega cards. One to match each of 1070,1080 and 1080Ti.
For the card you are describing, the 1080Ti competitor, how much power would it consume? 350W? What kind of sane person would want such a inefficient card.
Amd made the mistake of choosing GloFo and now they're paying for it.
They should break the contract, take the loss and move to a more power efficient and better process.

 

[Newbian]

I don't know why people who are waiting for Vega don't just buy Pascal. Its quite obvious Vega won't match the power efficiency of Pascal and the secrecy around Vega probably means amd knows it has a dud on its hands. With All these red flags, its safe to assume Vega will disappoint just like these Polaris rebrands.

Do you think power efficiency matters as much as performance per dollar?

If they can release their big vega that's last I checked been advertised at 10% or so more powerful then the gtx 1080 for around that price or a bit less this is what people care more for.

Especially for the cheaper cost of freesync and it means amd gamers can finally play at 1440p on high and above game settings on more modern games.

 

[mohit9206]

Do you think power efficiency matters as much as performance per dollar?

If they can release their big vega that's last I checked been advertised at 10% or so more powerful then the gtx 1080 for around that price or a bit less this is what people care more for.

Especially for the cheaper cost of freesync and it means amd gamers can finally play at 1440p on high and above game settings on more modern games.

10% more performance for 50% more power consumption isn't what most people want.

 

[Newbian]

10% more performance for 50% more power consumption isn't what most people want.

Says who?

Yes for some having better power consumption is what they would want for certain specialized builds but big vega would be targeted more for gamers and performance or price matters.

Even more so if they are able to keep the temperatures down as we have seen with the 500 series for some people.

 

[sandorski]

Raja Koduri is his source.

But Vega is still GCN. So its a given that power efficiency won't be close to Pascal.



There needs to be 3 Vega cards. One to match each of 1070,1080 and 1080Ti.
For the card you are describing, the 1080Ti competitor, how much power would it consume? 350W? What kind of sane person would want such a inefficient card.
Amd made the mistake of choosing GloFo and now they're paying for it.
They should break the contract, take the loss and move to a more power efficient and better process.

GCN is meaningless in predicting the differences.

 

[KompuKare]

Amd made the mistake of choosing GloFo and now they're paying for it.
They should break the contract, take the loss and move to a more power efficient and better process.

The problem with GF's process is less about it being bad as such, rather AMD decided or forced themselves to run it way outside of the sweetspot with Polaris and even more so with the 500 series refresh. Once they knew the characteristics of the process (years before Polaris shipped) they should have made the decision to run wide and slow rather than fast. Yes, their perf/area would have suffered even more vs GP107 but surely between GF's process being worse than TSMC's and the WSA they should be able to afford to do so. Plus by running inside the sweet spot for the process they (& their OEM partners) could have saved on VRMs and cooling. Remember the cut-down workstation Radeon Pro WX5100 uses Polaris 10 and gets by with 75W. Sticking to 0.9V or 0.8V and max 1000MHz or so in a wider design (say with 3000+ shaders) should have gotten them competitive with GTX 1060 at a similar power usage even before the efficiency gains the Vega architecture is expected to bring.

But as the charts produced by The Stilt for the Ryzen launch show, GF's process is actually very good when running at or below its sweetspot and it's only when you try to overvolt and overclock it past a certain point that it gets very bad. Unfortunately, RX 580 seems like some theoretical Ryzen 7 1850X with a base speed of 4.0GHz with voltages and power consumption figures to match.

As for Vega being delayed, I had assumed the big workload of launching Ryzen was the reason and that (hopefully) the driver team is using this delay to good effect since Vega seems to the biggest departure from GCN so far.

Of course, it is possible that someone at AMD was not rational and really wants to HBM2 after they spend so many resources getting it to this stage, and that HBM2 is still just too expensive for mainstream and at this stage they have no backup GDDR5X plan. Surely, even years ago they would have realised that HBM2 for small Vega would very very risky though? Still decisions are not always rational (for instance the brilliant DP performance of Tahiti and Hawaii just made their products appears power hungry and they still didn't really break into the HPC market).

EDIT: was there an earlier rumour that Vega would be TSMC anyhow?

 

[Crumpet]

I honestly don't understand all the people I see who say "I was waiting for Vega but it took too long and I bought a GTX---- instead".. Well, what do you want? Do you want us to feel sympathy for you? Do you expect us to do the same? what do you want from us?

Vega has already made Nvidia sit up, and at the moment that's all the positive information I need to sit and wait. Another 8 weeks or so in the whole scheme of things won't make a huge difference if they are still on for their Q2 target, and then we'll know.

Hells, with Vega supposedly coming this quarter, and Volta being hurried out supposedly next quarter.. I think you'd have to be mad (or utterly uninformed) to purchase Pascal right now, unless you have the cash to just switch things up with the new releases.

 

[jpiniero]

I am skeptical but AMD probably should have cancelled Fiji so who knows. They are surely taking their sweet time on this though.

EDIT: was there an earlier rumour that Vega would be TSMC anyhow?

There was. It doesn't seem like it is true though.