[Rumor] AMD Radeon RX Vega indefinitely delayed

[Crumpet]

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

Should have cancelled Fiji?

If i'd had the money at the time I would have bought a Fury or Fury X, and their performance is still great today. Yes they are very hot and thirsty, but eh. If I cared about power draw I wouldn't have 2 gaming pc's, 2 gaming laptops and 5 fishtanks.

 

[jpiniero]

If i'd had the money at the time I would have bought a Fury or Fury X, and their performance is still great today. Yes they are very hot and thirsty, but eh. If I cared about power draw I wouldn't have 2 gaming pc's, 2 gaming laptops and 5 fishtanks.

Yeah that's the point - AMD likely lost money on Fiji. Or didn't make very much.

 

[Crumpet]

Yeah that's the point - AMD likely lost money on Fiji. Or didn't make very much.

Nvidia took them by surprise I feel. But you have to release something to recoup some of the losses. But yes, it didn't go to plan.

 

[Glo.]

AMD actually earned quite a serious amount of cash on Fiji. They were selling GPUs at 300$ and still making money on them, so it was not that bad, as people believe it to be. It was largest 28 nm GPU, with massive interposer, and new, next generation memory type, but at the same time, the process was mature, and cheap.

 

[iBoMbY]

Obviously a troll. He posted the same on Reddit.

 

[Qwertilot]

The main part of the costs are R&D (+ design I guess) and that is already sunk by the time you're getting near release anyway.

 

[SpaceBeer]

1. 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.

2. 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.

3. 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).

1. Polaris 10 is ~15% larger than GP106 (232 vs 200 mm^2), so it should be more expensive to produce. However, if GF is, let's say 20%, cheaper than TSMC, why would AMD go for TSMC? We don't know prices, of course, but it might be GF is better choice, even if TSMC has better process. Plus, we are not sure if Pascal is able to achive higher clocks because of process or better architecture.
And if power efficiency is such a big deal, AMD and AIB partners could easily make RX 480/580 "Eco" with 900/1100 MHz base/boost clocks. Such card would probably be fast and efficient as GTX 1060 3GB. They have already done this with R9 Nano or WX 7100. Or you could buy RX 480 4GB, downclock (undervolt) it and have exactly the same perf/watt for the same price, plus 1GB more VRAM (compared to GTX 1060 3GB)

2. Vega isn't delayed. More than year ago, on AMD's official roadmap Polaris was the only 2016 product and Vega was planned for 2017. (though it was placed close to the beginning of the year, but never confirmed for Q1). And several months ago, they confirmed it will be released in H1. So only if it's not released by 30th June, you can say it is delayed

3. It is confirmed there are lot of changes implemented in new architecture and HBCC is one of the most important. As I know, Vega based GPUs will be able to use both GDDR and HBM. So even if the HBM2 was the reason, nothing would stop them to make let's say 48 CUs GPU with GDDR5 an get some potential 1070 buyers. I doubt HBM implementation is causing any delay, since they have successfully done that 2 years ago. Even if HBM2 price due to low availability is an issue, I'm sure AMD (and any other company) would rather go for lower earnings (due to higher BOM cost) than product delay.

 

[vissarix]

I dont know if its delayed or not....If Vega was good Amd would have talked about it, showing some benchmarks so people might wait for Vega instead of buying the Gtx1080ti or a Titan XP...

Amd showed cherry picked benchmarks on both Fury X launch which according to them was faster then a Titan X but in reality was slower then a Gtx980ti...They did the same with Ryzen 1800X showing best case scenario and sandbagging i7 6900k scores althought Ryzen is a good product because it costs way less then competition...

Chances are that Vega might be that bad that not even cherry picked benchmarks can save it...

 

[tamz_msc]

I dont know if its delayed or not....If Vega was good Amd would have talked about it, showing some benchmarks so people might wait for Vega instead of buying the Gtx1080ti or a Titan XP...

Amd showed cherry picked benchmarks on both Fury X launch which according to them was faster then a Titan X but in reality was slower then a Gtx980ti...They did the same with Ryzen 1800X showing best case scenario and sandbagging i7 6900k scores althought Ryzen is a good product because it costs way less then competition...

Chances are that Vega might be that bad that not even cherry picked benchmarks can save it...

Cherry-picked benchmarks on the 1800X? Everybody could more or less replicate the results that AMD gave in its slide deck.

 

[SpaceBeer]

I dont know if its delayed or not....If Vega was good Amd would have talked about it, showing some benchmarks so people might wait for Vega instead of buying the Gtx1080ti or a Titan XP...

Amd showed cherry picked benchmarks on both Fury X launch which according to them was faster then a Titan X but in reality was slower then a Gtx980ti...They did the same with Ryzen 1800X showing best case scenario and sandbagging i7 6900k scores althought Ryzen is a good product because it costs way less then competition...

Chances are that Vega might be that bad that not even cherry picked benchmarks can save it...

They have been talking about it
https://www.youtube.com/watch?v=bDl6xJJqIAU

And there were some benchmarks/show-offs during New Horizon event and CES. We also know Liquid sky is/will be using Vega. So it works

 

[Harmaaviini]

If their target was a card ~1.1x 1080 I think they would have gone quiet when 1080 was launched and they wouldn't have shown any demos with engineering samples at christmas. Aiming at that performance with a 500+ mm^2 die vs 314 mm^2 would be folly.

Right now even if they are at 1080ti performance level things are not really ideal but also not a disaster. Imagine 250$ 1070 and 375$ 1080. Those would sell. It surely can be done if you look at historical prices with similar chip sizes. And also consider the ASP of those cards over their lifetime so far. I bet they would rather maintain margins and launch Volta as soon as possible though. Coincidentally(?) there are rumors...

 

[Krteq]

...If Vega was good Amd would have talked about it, showing some benchmarks so people might wait for Vega instead of buying the Gtx1080ti or a Titan XP...

Do you remember HD 4800 series vs GTX 200 series few years ago?

 

[coercitiv]

Raja Koduri is his source.

Raja Koduri is not his source unless he actually made the statement. This thread is full of opinions, yet no link to any kind of article / press release / analysis is available to reinforce the OP.

Actually, the last mention of Vega performance that I know of was made by Don Woligroski, AMD Desktop Processor Marketing Manager on April 6th during a Tom's Hardware AMA, and it was the following:

Vega performance compared to the Geforce GTX 1080 Ti and the Titan Xp looks really nice.

Now, whether that's just empty marketing talk or not, it remains to be seen, but at least this is a proper source. That man actually said what I quoted.

 

[vissarix]

Do you remember HD 4800 series vs GTX 200 series few years ago?

Thats 9 years ago...still doesnt make sense for Amd to not show any benchmark if Vega is about to launch in may..

 

[tamz_msc]

Do you remember HD 4800 series vs GTX 200 series few years ago?

As awesome as it sounds if AMD can pull off another RV770, we already know that such a thing is less likely to happen because the Vega die is bigger than GP102, and this time around it's very different from RV770 vs GT200.

 

[OatisCampbell]

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.

I was one of the waiters who bought a 1080Ti, personally I don't care what anyone else buys. So what I want from you is: Buy whatever computers parts you feel like buying, or buy none at all. Whichever makes you happiest and fits your budget.

Without valid source this thread is either a. pointless conjecture b. actionable violation of NDA. My guess is pointless conjecture. Companies haven't released info on unreleased products for many years, we have no more paper launches.

I get why companies don't paper launch first, because it gives their competitor additional time to make adjustments they can, like clock speeds or fan profiles.

In this case, with your competitor's full line on the store shelves with stock, I don't get the logic of holding back if you're got valid performance competition.

Only see two possible outcomes:

1. AMD has cards that compete on performance, paper launch stops competitor's sales to some extent and locks down some market share.
2. People who are too impatient to wait buy other cards anyway.

I suppose there's some risk of "Jinkies our HBM2 supplier's factory burned down! Our plans are thwarted!" but the odds of that have to be low. With NV already commanding huge market share lead and their parts selling well, I might paper launch here if I were Lisa.

 

[Arachnotronic]

Raja Koduri is not his source unless he actually made the statement. This thread is full of opinions, yet no link to any kind of article / press release / analysis is available to reinforce the OP.

Actually, the last mention of Vega performance that I know of was made by Don Woligroski, AMD Desktop Processor Marketing Manager on April 6th during a Tom's Hardware AMA, and it was the following:


Now, whether that's just empty marketing talk or not, it remains to be seen, but at least this is a proper source. That man actually said what I quoted.

In theoretical TFLOPs, Vega 10 does in fact look good compared to the Titan XP/1080 Ti.

 

[Erazor51]

Vega will release in a months time and the wait will be worth it

 

[Krteq]

Thats 9 years ago...still doesnt make sense for Amd to not show any benchmark if Vega is about to launch in may..

Well, in case of RV770 they also didn't show any benchmarks neither a specs before launch.. and it was a quite positive surprise.

 

[Yakk]

Obviously a troll. He posted the same on Reddit.

Ah, there we go, good catch.

 

[Blockheadfan]

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.

Ignoring the rest of the source-less "information" there is no way AMD "anticipated" the bolded part, even causally interested people like myself know that this goes against the precedent of recent Nvidia releases.

Titan - 780Ti, Titan X - 980Ti, Titan Xp - 1080Ti. But suddenly AMD thought the 1080Ti was going to be a mild refresh of GP104? Discredits your entire post in the first sentence in my opinion.

Edit : Also is there an invented-with-no-corroborating-evidence-at-all tag? Rumor to me implies that a website with at least some history of being correct or credibility posted an article. This seems like, until the OP proves otherwise, completely fabricated speculation.

 

[raghu78]

Also is there an invented-with-no-corroborating-evidence-at-all tag? Rumor to me implies that a website with at least some history of being correct or credibility posted an article. This seems like, until the OP proves otherwise, completely fabricated speculation.

This is not a rumour thread. Its completely fabricated stuff. AMD have stated Vega will launch in H1 2017. They have reconfirmed that statement at recent events like Ryzen launch. I call this thread a pathetic FUD attempt.

 

[Bouowmx]

it might be GF is better choice, even if TSMC has better process. Plus, we are not sure if Pascal is able to achive higher clocks because of process or better architecture.

NVIDIA GeForce GTX 1050 [Ti] on Samsung (Global Foundries) maximum 1.9 GHz, compared to 2.1 GHz of every other Pascal product on TSMC.

 

[SpaceBeer]

Same process, different max oc clocks
https://www.techpowerup.com/reviews/ASUS/GTX_750_Ti_OC/28.html
https://www.techpowerup.com/reviews/MSI/GTX_970_Gaming/30.html
https://www.techpowerup.com/reviews/EVGA/GTX_950_SSC/33.html
https://www.techpowerup.com/reviews/ASUS/R9_380X_Strix/26.html

Even if the difference between Pascal cards is only due to different process (and not e.g. power or thermal specifications defined for those cards), would those 10% clock increase help AMD to make much faster cards? Looking at RX 570/580, I would say no. So if GloFo is cheaper, and I beleive it is, at least for AMD, there is no reason to use TSMC. It would be much better for them to improve GCN and be able to achieve higher clocks with GF process. I am sure AMD would be happy to have architecture able to reach 1900 MHz and be power efficient as GP107. In that case Polaris 10 would be much faster than GP106, maybe even close to GTX 1070 (with additional ROPs and higher bandwidth)

 

[Det0x]

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.

Not so fast. (predicted Vega would be 50+% faster then a 1080 over a month ago)

You should follow this thread more closely.. Post #1330 and #1343

If we assume Vega should be running atleast @ 1500mhz core (look further down for "proof"), then we can extrapolate what the scores would be once clocks are finalized

1500/1200 = should net a 25% performance increase with 100% clockspeed scaling

Face detection: 1080TI is ~60% faster then Vega
"simulated 1500mhz Vega" = 194.58
1080TI = 313.26

TV optical flow: Vega is ~4% faster then 1080TI
"simulated" 1500mhz Vega = 72.13
1080TI = 69.22

Ocean surface: Vega is ~20% faster then 1080TI
"simulated" 1500mhz Vega = 4760.57
1080TI = 3944.8

Particle simulation: Vega is ~16% faster then 1080TI
"simulated" 1500mhz Vega = 2123.35
1080TI = 1816.88

T-Rex: 1080TI is ~7% faster then Vega
"simulated" 1500mhz Vega = 18.48
1080TI = 19.773

Video composition: Vega is ~5% faster then 1080TI
"simulated" 1500mhz Vega = 202.2
1080TI = 192.586

Bitcoin mining: Vega is ~7% faster then 1080TI
"simulated" 1500mhz Vega = 1509.07
1080TI = 1408.061

Now the big question is, how much above 1500mhz will the consumer Vega10 be running at ?
Afterall we have gotten that 1.5ghz number from a professional workstation-card leak, and as we know, consumer version tends to be clocked higher..

Not bad considering the 1500mhz Vega10 should be running at a 225w terminal envolope.

https://videocardz.com/65521/amd-vega-10-and-vega-20-slides-revealed


https://videocardz.com/67275/amd-vega-spotted-with-4096-cores-and-8gb-2048-bit-memory

Just yesterday we found the first CompuBench result revealing 64 Compute Units in Vega. Today we share another leak, which confirms 64 CUs, but also memory subsystem configuration.

AMD Radeon RX Vega: 4096 Stream Processors and 8GB memory
SiSoft benchmark detected 64 Compute Units on 687F:C3 device (so not C1 like in the previous leak). This device has 8GB 2048-bit memory configuration, which means two HBM2 stacks, each 4GB and 1024-bit.

This particular variant is clocked at 1200 MHz which translates into 9.8 TFLOPs (in theory) single-float. Quite some distance away from AMD’s promised performance number of 12 TFLOPs so its clear that the performance can increase by atleast +25% once the engineers have refined the clocks.

Here is the GTX 1080's scores for comparrison. And This Vega part is ahead by a reasonable amount, even if its only running by alpha drivers and clocks(1200mhz): http://imgur.com/z68xbLd

So indeed we have atleast some basis for believing AMD will open a can of whoopass

Third option:

1. The printed circuit board (PCB) is designed for a 225w card.
2. Official early showcase had the engineering sample GPU running at lowish clocks, i would guesstimate/hope ~1000mhz.
3. More recent leaks show ES cards running at 1200mhz.
4. Finalized (read non-ES) cards could be running at 1500+mhz once released in one and a half months. (and finally using all the available power (225w))

Above could be just as plausible as your lowballing. And I have no problem believing big daddy Vega could be 50% faster then a 1080 in a "AMD friendly game" as you put it.

This should be quite telling for whats going on..

Do you remember what clockspeed the Ryzen engineering samples was running at. ?