Vega/Navi Rumors (Updated)

[Crumpet]

Just our own Glo. Plus the various rumors around the web running from 1.2 GHz (probably ES **) up to 1.6 GHz (Probably BS):



** Even if this was an ES chip, there could well be a cut down and/or down clocked Vega at the same performance level

Hmm interesting. Could this perhaps have something to do with the "radiator compatible" note thingy on the Asus Crosshair VI hero?

 

[railven]

AMD is being very quiet. So it seems like they either don't have something great (not shouting from the roof tops) or they are sandbagging like the dickens and plan on surprising people as they did with Ryzen. With the lack of solid evidence out there, we are just stuck reading tea leaves and hoping that AMD is sandbagging (and that they will inject some much needed competition back into the market).

AMD was vocal about Polaris up literally until leaks of Pascal started to leak. Everyone was of opinion that Pascal was going to be far behind Polaris release because AMD show cased Polaris in like early 2016. Then Pascal released and AMD went dark. I recall a presentation post-Pascal appeared to many that AMD was caught off guard.

Now compare it to Ryzen. AMD had info galore. They weren't quiet at all about what they had. Vega was being spoke about and regularly mentioned until 1080 Ti released.

If I had to guess 1080 Ti through a monkey wrench into the mix that will result in Vega launched at a cheaper price point than intended only to get dwarfed by GTX 1080 below it in price and performance and GTX 1080 Ti above it.

I see AMD cranking as much clocks out of it as they can to fit well above GTX 1080 otherwise they'll have a hard sale on their hands. And smaller Vega just got pushed into the <$400 price bracket if it's slower than GTX 1080. With the cost of these dies, I don't see AMD being very happy if any of my opinion holds water.

 

[Crumpet]

I see AMD being "silent" about Vega as a simple part of their advertisement strategy.

Since the initial Vega talk at the beginning of the year we've had a Ryzen 7 release, a Ryzen 5 release, and a whole refresh of the Rx gpu line. Vega previews and spoilers in any of this would take media attention and hype away from those products.

When the media buzz about the Rx refresh dies down, BOOM VEGA.

This keeps the media talking about AMD for the entire 1st half of 2017, keeping the products fresh in peoples minds, in the tech reviewers minds, and in the papers for the average consumer.

 

[Ajay]

Now compare it to Ryzen. AMD had info galore. They weren't quiet at all about what they had. Vega was being spoke about and regularly mentioned until 1080 Ti released.

AMD went dark well before the 1080 Ti launched. Ryzen had more leaks, but info was rather low until after AMD started making public comments (starting with ISSCC) and really ratcheted things up this past winter . In the absence of solid leaks or public info - we are left to speculate. In that regard, either one of us could be correct. Given the success of Ryzen (Intel reacts to Ryzen: Intel Skylake / Kaby Lake Thread) - I'm trying to keep an open mind.

I'm not looking for a high end card - so I won't be crushed if Vega falls short. I just like what competition brings to the market (better products, better price/performance, accelerated timelines).

 

[Mopetar]

AMD is being very quiet. So it seems like they either don't have something great (not shouting from the roof tops) or they are sandbagging like the dickens and plan on surprising people as they did with Ryzen. With the lack of solid evidence out there, we are just stuck reading tea leaves and hoping that AMD is sandbagging (and that they will inject some much needed competition back into the market).

With Ryzen they had at least released a single actual benchmark that looked very favorable. The showed off that they could beat one of Intel's best chips in at least one workload. This naturally lead to some confusion as no one quite expected it, so the inclination was to assume it was an edge case, but it turned out that Ryzen had better SMT, reasonably good IPC, and could hit similar clock speeds to Intel's HEDT products at similar or better TDP ranges.

The only real Vega data we have is some actual gameplay showing 1080-level performance in a game we expect to favor Vega and the explanation that it's an early engineering sample chip using Fury drivers. That's too much uncertainty to make any reasonable predictions. If AMD had released a newer result showing 1080 Ti-level performance, even if it's in an AMD favored game, I'd be more optimistic. As I said previously, the best evidence for Vega being any good is the 1080 Ti being priced at $700, because there's no reason NVidia couldn't have priced it at $800 - $900 if they expect Vega to be another dud.

My optimistic estimate is currently that Vega isn't quite as good as a 1080 Ti in terms of absolute performance. That isn't necessary a problem as if AMD sells it for ~$600, I don't care if it's only 95% of the performance, but I think if they planned on beating the 1080 Ti, they would have dropped some kind of hint by now because there are a lot of people upgrading to Ryzen systems that are getting NVidia graphics cards because the best AMD can offer right now is a 480/580 card and the $$$ you can save on Ryzen over a similar Intel chip means extra money to spend on a GPU that AMD isn't selling. AMD could probably get some of those consumers to forestall some of those sales if they alluded to having something similar. Right now their market is the AMD faithful, and that's not a terribly huge market. Even if they come out with some kind of ridiculous surprise of 15% better performance than a 1080 Ti at the same cost, a lot of people will probably just keep the 1080 Ti and NVidia loyalists will just wait until Volta.

 

[JDG1980]

If I had to guess 1080 Ti through a monkey wrench into the mix that will result in Vega launched at a cheaper price point than intended only to get dwarfed by GTX 1080 below it in price and performance and GTX 1080 Ti above it.

It's hard to imagine AMD getting tripped up by GTX 1080 Ti. They had to know something like this was going to be released, considering that it had been done two times previously (780 Ti and 980 Ti). And since Titan X Pascal was released in August 2016, they would have known back then roughly how much performance to expect. Heck, we all saw it coming on this board long before any such thing was officially announced.

If GTX 1080 + 10% was really the best Vega could do, as the pessimists say, then it's hard to believe how AMD could have ever thought it would be competitive, considering the relatively small size of GP104 and that Nvidia always has a bigger chip slotted in above this. I've said it before and will say it again: given the die size and the architectural improvements, we should expect nothing less than GP102 performance from Vega 10. If this doesn't happen, then it's a dud.

 

[Karnak]

Based on a post from Glo. I don't think Vega will be hitting 1.5 GHz on Air. The fact that Vega will need an AIO to run at 1.5 GHz is kind of bad news - make me think that this clock will be hit with a 'Boost BIOS' like Polaris (and worse perf/watt because of it). I don't think we will see boost clocks above ~ 1.4 GHz. Still, with a major uarch upgrade, I expect Vega to have much better perf/watt than Polaris from significantly better perf/CU and plenty of memory bandwidth (and maybe beyond plenty with HBC - this will be interesting to see).

AMD stated that Vega (with it's NCU) is optimized for higher clock speeds. Even Polaris is (now) able to hit up to 1450 MHz out of the box. And we know the MI25 with 12.5 TFLOPS. To achieve that, Vega needs to hit ~1525MHz.

I don't think we'll see Vega with lower clock speeds than Polaris.

 

[Bacon1]

http://www.anandtech.com/show/11278/amd-radeon-rx-580-rx-570-review/2

Nope, still broken.

Computerbase shows it as fixed

https://www.computerbase.de/2017-04/radeon-rx-580-570-test/4/

 

[railven]

AMD went dark well before the 1080 Ti launched. Ryzen had more leaks, but info was rather low until after AMD started making public comments (starting with ISSCC) and really ratcheted things up this past winter . In the absence of solid leaks or public info - we are left to speculate. In that regard, either one of us could be correct. Given the success of Ryzen (Intel reacts to Ryzen: Intel Skylake / Kaby Lake Thread) - I'm trying to keep an open mind.

I'm basing my opinion that AMD has more moles than we do and probably knew at least the price of the expected GTX 1080 Ti which leads into my next response.

It's hard to imagine AMD getting tripped up by GTX 1080 Ti. They had to know something like this was going to be released, considering that it had been done two times previously (780 Ti and 980 Ti). And since Titan X Pascal was released in August 2016, they would have known back then roughly how much performance to expect. Heck, we all saw it coming on this board long before any such thing was officially announced.

It isn't the product, it was the price. I'm sure AMD already knew what a card like the GTX 1080 Ti could do and probably, like everyone on these forums, expected it to hit between the GTX 1080 and Titan-Pascal in regards to price.

I don't think anyone honestly expected Green for Greed Nvidia would drop the price of the GTX 1080 and slot the 1080 Ti in it's place. I'm sure even AMD didn't expected that. Why I said they probably had intentions of putting Big Vega in the $600 price range, I'd have guess $650, it beats GTX 1080 and wins over enthusiast (because AMD doesn't have that market brand) or causes NV to at least shift prices, but they get to launch at a high price point. With the GTX 1080 Ti @ $700 big Vega has to be a lot closer to it than some what faster than GTX 1080.

With AMD's brand power, they'd be lucky to sell a faster GTX 1080 for $550, but knowing the trends, AMD would be forced to sell a faster card for less, thus putting big Vega in the <$500 price point. And little Vega even lower.

But this is my speculation. I personally hope I'm wrong because AMD can't afford to sell such a die in Vega for less than GP104. It would be suicide.

 

[Ajay]

AMD stated that Vega (with it's NCU) is optimized for higher clock speeds. Even Polaris is (now) able to hit up to 1450 MHz out of the box. And we know the MI25 with 12.5 TFLOPS. To achieve that, Vega needs to hit ~1525MHz.

Well, MI25, based on Vega, is for HPC servers. No doubt AMD can pick the best bins for those GPUs to make sure they hit their targets.

 

[Ajay]

@railven Maybe NV lowered prices because it is true that they are missing their sales targets for their high end enthusiast GPUs.
It's also possible that GloFo wafer prices for 14LPP are lower than TSMC's 16FF+, so a lower price may not be suicide at all.

There's just a lot we don't know, hence, all the speculation. We don't know what all the changes to CGN5 (NCU) will yield. We don't know any details about the actual physical implementation of a CU in silicon or how that will affect perf/watt. Lots of unknowns, including basic things like clocks. I think, as was the case with Ryzen, AMD is waiting to the last possible moment to release info because they want to hit the highest clocks they can reasonably provide. The people who know more, just aren't talking right now.

 

[Karnak]

Well, MI25, based on Vega, is for HPC servers. No doubt AMD can pick the best bins for those GPUs to make sure they hit their targets.

I know, but iirc there was never a (big) difference in clock speeds between server and consumer cards if we're talking about AMD.

 

[Mopetar]

I don't think anyone honestly expected Green for Greed Nvidia would drop the price of the GTX 1080 and slot the 1080 Ti in it's place. I'm sure even AMD didn't expected that. Why I said they probably had intentions of putting Big Vega in the $600 price range, I'd have guess $650, it beats GTX 1080 and wins over enthusiast (because AMD doesn't have that market brand) or causes NV to at least shift prices, but they get to launch at a high price point. With the GTX 1080 Ti @ $700 big Vega has to be a lot closer to it than some what faster than GTX 1080.

AMD can probably adjust prices down to below $500 and still make a profit even though they wouldn't want to do so, but the bigger problem is that if Vega is closer to performance to a 1080 than it is to a 1080 Ti, it doesn't have a lot of purpose. The 1080 Ti is scraping around the edge of 60 FPS at 4K and really it's not a big issue with a free/g-sync monitor so coming in below that means its not a good long-term card for 4K. Better than 1080 performance makes it a great card for high-FPS 1440p gaming for the foreseeable future, but you can get adequate 1440p gaming from a 1070 for a lot less money.

 

[beginner99]

Now compare it to Ryzen. AMD had info galore. They weren't quiet at all about what they had. Vega was being spoke about and regularly mentioned until 1080 Ti released.

Problem with your theory is that AMD did not now about Pascal because it wasn't released yet. So getting surprised I agree and is totally valid. However the 1080Ti isn't a surprise at all. We all knew it was coming and that it will offer similar gaming performance as the original Titan. Also price could be expected to land were it did, slightly more than 980Ti. Meaning there is 0 chance AMD was caught off-guard here. Even I could have guessed with high accuracy which performance/$ to target to beat a 1080Ti.

The only possible thing is that Vega sucks in general. Knowing what NV will do and being able to offer a competing product are 2 entirely different things. But still my point is your theory doesn't seem plausible because it's a different scenario.

 

[beginner99]

It's also possible that GloFo wafer prices for 14LPP are lower than TSMC's 16FF+, so a lower price may not be suicide at all.

Producing at GloFo for sure cheaper for AMD than TSMC due to WSA. Every chip more they sell, the less they have to pay for the WSA. So even if wafers are same price, AMD pays less than NV at TSMC because AMD then has to pay lower penalty for WSA.

There are 2 reasons AMD should price low: The WSA and getting market share. We have already seen that with Ryzen and I expect a similar line-up. A not so cheap top-end model (like 1800x) and then clearly cheaper marginally slower models. So a watercooled top-end model beating 1080Ti by a small margin at same price and then something that is air-cooled, used more power and maybe 5%-8% slower for like $100 less.

 

[alcoholbob]

I'm basing my opinion that AMD has more moles than we do and probably knew at least the price of the expected GTX 1080 Ti which leads into my next response.




It isn't the product, it was the price. I'm sure AMD already knew what a card like the GTX 1080 Ti could do and probably, like everyone on these forums, expected it to hit between the GTX 1080 and Titan-Pascal in regards to price.

I don't think anyone honestly expected Green for Greed Nvidia would drop the price of the GTX 1080 and slot the 1080 Ti in it's place. I'm sure even AMD didn't expected that. Why I said they probably had intentions of putting Big Vega in the $600 price range, I'd have guess $650, it beats GTX 1080 and wins over enthusiast (because AMD doesn't have that market brand) or causes NV to at least shift prices, but they get to launch at a high price point. With the GTX 1080 Ti @ $700 big Vega has to be a lot closer to it than some what faster than GTX 1080.

With AMD's brand power, they'd be lucky to sell a faster GTX 1080 for $550, but knowing the trends, AMD would be forced to sell a faster card for less, thus putting big Vega in the <$500 price point. And little Vega even lower.

But this is my speculation. I personally hope I'm wrong because AMD can't afford to sell such a die in Vega for less than GP104. It would be suicide.

I don't think AMD expected the 1080 Ti to be 2-3% faster than the Titan X (2016), mostly because most people didn't. Most of the rumors ended up being misdirection (aka 6+GB 3200-3320 core cut down GP102 for $899) and Nvidia ended up overdelivering on both performance AND price.

 

[tential]

I see AMD being "silent" about Vega as a simple part of their advertisement strategy.

Since the initial Vega talk at the beginning of the year we've had a Ryzen 7 release, a Ryzen 5 release, and a whole refresh of the Rx gpu line. Vega previews and spoilers in any of this would take media attention and hype away from those products.

When the media buzz about the Rx refresh dies down, BOOM VEGA.

This keeps the media talking about AMD for the entire 1st half of 2017, keeping the products fresh in peoples minds, in the tech reviewers minds, and in the papers for the average consumer.

You've made way too many reasonable posts.

I don't think AMD expected the 1080 Ti to be 2-3% faster than the Titan X (2016), mostly because most people didn't. Most of the rumors ended up being misdirection (aka 6+GB 3200-3320 core cut down GP102 for $899) and Nvidia ended up overdelivering on both performance AND price.

I don't think anyone honestly expected Green for Greed Nvidia would drop the price of the GTX 1080 and slot the 1080 Ti in it's place. I'm sure even AMD didn't expected that. Why I said they probably had intentions of putting Big Vega in the $600 price range, I'd have guess $650, it beats GTX 1080 and wins over enthusiast (because AMD doesn't have that market brand) or causes NV to at least shift prices, but they get to launch at a high price point. With the GTX 1080 Ti @ $700 big Vega has to be a lot closer to it than some what faster than GTX 1080.

With AMD's brand power, they'd be lucky to sell a faster GTX 1080 for $550, but knowing the trends, AMD would be forced to sell a faster card for less, thus putting big Vega in the <$500 price point. And little Vega even lower.

But this is my speculation. I personally hope I'm wrong because AMD can't afford to sell such a die in Vega for less than GP104. It would be suicide.

I literally said the 1080Ti would be $700 hours before it launched. It's not hard to figure out. This has happened SO MANY TIMES. It's honestly mindboggling how any part of the Nvidia release cycle is a surprise to any of you. We post all year here and still it's the same "Woah I can't believe it came in at this price, or Woah, I can't believe there is a Ti card this year! or Woah!!!! I can't believe the Ti card is faster than the Titan card!!!!"

How are you guys still surprised.... this stuff is so obvious at this point.

Also.... lol... do you really think that companies are sitting there watching every rumor off wccftech with us? I don't even....

 

[ConsoleLover]

AMD still has a huge advantage in DX12 and Vulkan, in fact the new overclocked RX 580's to 1445MHz match the 1070FE in Doom.

So I do expect Vega to win in DX12 and Vulkan titles, at least in most of them even against the 1080TI. I expect it will be about 10% faster in Vulkan and DX12 titles, about 10% slower in DX11 titles. That is my expectation.

In specific extremely optimized titles for either Nvidia or AMD I expect it to lose/win more. But on average I expect it to be about 10-15% faster in Vulkan/DX12 and about 10%-15% slower in DX11.

 

[Glo.]

I am actually quite enjoying the doom and gloom over Vega architecture.

 

[unseenmorbidity]

there's no reason NVidia couldn't have priced it at $800 - $900 if they expect Vega to be another dud

Sure, there is, because it limits what amd can sell vega for. It will hurt amd far more than nvidia. This move has been planned for a year. It was a preemptive strike that likely doesn't tell us anything.

This is monopoly building 101. You price your competition out of the market, even if you take a loss doing it. Though, I doubt nvidia is selling at a loss.

 

[Arachnotronic]

Sure, there is, because it limits what amd can sell vega for. It will hurt amd far more than nvidia. This move has been planned for a year. It was a preemptive strike that likely doesn't tell us anything.

This is monopoly building 101. You price your competition out of the market, even if you take a loss doing it. Though, I doubt nvidia is selling at a loss.

There's zero way that 1080 Ti is being sold at anything close to a loss.

 

[zinfamous]

There's zero way that 1080 Ti is being sold at anything close to a loss.

on top of that, imagine how nVidia has so successfully cajoled their customers into making the 1080Ti their 2nd, or even 3rd GPU purchase within a year. It's double, triple dipping on massive, massive margins.

I bet George Lucas is their chief marketing strategist.

 

[iBoMbY]

Though, I doubt nvidia is selling at a loss.

They are easily making $500 profit on each. The production cost is nothing compared to the sales price.

 

[Samwell]

AMD is being very quiet. So it seems like they either don't have something great (not shouting from the roof tops) or they are sandbagging like the dickens and plan on surprising people as they did with Ryzen. With the lack of solid evidence out there, we are just stuck reading tea leaves and hoping that AMD is sandbagging (and that they will inject some much needed competition back into the market).

There is still some time till Computex, where they probably launch. Even a paperlaunch would make no sense before May10, cause Nvidia will probably show off Volta architecture there and would take attention from Vega. After that who knows, maybe paperlaunch and hard launch at computex. But pretty sure no hard launch before, because the rumour mill is way too quiet.

 

[Phynaz]

They probably connected the monitors with an old interface instead of DP. Or were you suggesting that my numbers were wrong?

If which connector on your card you connect a display to matters in this case, then the multi-monitor implementation is broken.

As for your numbers, I have no idea if the are correct or not.