AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

[Silent_Buddha]

The quote seems to be



The next couple of months is from now to end of July. We knew FE is mid-late June and were hoping RX Vega in July sometime. Nothing about that quote changes that.

Ah the internet.....

Queue inflaming WCCFTech "article" then pharma posting that here as well.

OK, I figured it'd be something like that. The Reddit poster apparently heard what he wanted to hear versus what she actually said. Either that or they aren't a native English speaker and interpreted "over the next couple of months" as being over 2 months (as in after 2 months) instead of across 2 months.

Regards,
SB

 

[Malo]

Yeah I think people just interpreted it as 2 months after FE was released. This is all showing just how much hype (and hope) is behind Vega. They really need to execute on this one

 

[CSI PC]

That said and judging from the officially available results at github and (guesstimating from) the screenshot from AMDs FAD (courtesy of heise.de), I fail to see where there is a dramatic difference.

Yes, the 288 ms for the M40 seem to be wrong, since they should read 226 ms (it's 288 including column "n" [Forward (msec)], but this is included for M40 only). And yes, the DeepBench results seem to be added "Total Time" (column T in Baidu's github results) only and yes, that part of the benchmark seems to use only FP32 according to the github description.

But WRT the comparison Vega vs. Tesla P100 (PCIe), I do not see anything dramatically wrong here, if you do not insist on geometric means.

If ignoring the discrepancies.
Aren't Heise inferring the P100 (9.3 TFLOPs FP32) difference is only around 9.5% rather than the 33%, which means if AMD had used the Mezzanine P100 (10.6TFLOPs FP32) that GPU would at least tie with the Frontier
And this is ignoring that the test really should use the Quadro P6000 (12.1 TFLOPs FP32) that fits better in terms of the FP32 test and critically with the SpecViewperf AMD also did but with Titan X and not the Quadro GPU designed for such purposes with its professional drivers.
Still it is par for the course generally (just look at the arguments between Nvidia/Intel/Google) and AMD are not the only ones to have done that, most of them have been guilty to some extent.
Cheers

 

[pharma]

Would like to hear the wording that she uses. Did she say that it'll be launching "in the next 2 months" or "2 months after" FE?

Would have been nice for that poster to have at least attempted to quote what she actually said as apparently other people heard something different. There aren't too many combinations of words that could be misinterpreted in such a way, IMO, unless you aren't a native English speaker.

Regards,
SB

http://vocaroo.com/i/s0Qz6Mlo7wHn

 

[Deleted member 13524]

At JP Morgan conference:

https://streamable.com/9p5fo

"The first shipping Vega will be the Frontier Edition, which will ship with 16GB of memory and it will ship towards the later half of June. You will see the enthusiast gaming platform, the machine learning platform, the professional graphics platform very soon thereafter. And soon we will be launching Vega across all the market segments over the next couple of months".

For all we know, RX Vega (assuming it's a Frontier Edition with 4+4GB stacks and higher core clocks) could be coming a week after the Frontier Edition. Or it could be coming in mid July.

In practice, the actual availability window for RX Vega (and Mi25 and Radeon Pro) is between last week of June and 2nd week of July.
And this is assuming AMD is only showing us the big Vega 10.

 

[Silent_Buddha]

So, from the sounds of it, enthusiast gaming Vega products "very soon" after Vega FE and the rest of the consumer market segments over the next 2 months. So, I'd guess enthusiast Vega in June and mainstream and maybe budget (assuming Vega goes that low, which it may not) in July.

Now just hopefully they are decently competitive.

Regards,
SB

 

[CarstenS]

If ignoring the discrepancies.
Aren't Heise inferring the P100 (9.3 TFLOPs FP32) difference is only around 9.5% rather than the 33%, which means if AMD had used the Mezzanine P100 (10.6TFLOPs FP32) that GPU would at least tie with the Frontier
And this is ignoring that the test really should use the Quadro P6000 (12.1 TFLOPs FP32) that fits better in terms of the FP32 test and critically with the SpecViewperf AMD also did but with Titan X and not the Quadro GPU designed for such purposes with its professional drivers.

Well, AMD seems to compare PCIe cards and apart from that the heise article does not mention the SPEC benchmark, where the results undoubtedly show the influence of software (drivers) more than of the underlying hardware.

 

[Deleted member 13524]

So, from the sounds of it, enthusiast gaming Vega products "very soon" after Vega FE and the rest of the consumer market segments over the next 2 months.

At the risk of being critically pedantic, the "over the next couple of months" is a time period that started yesterday, so it's May 22nd + 2 months and not late June + 2 months.

 

[lanek]

Well, AMD seems to compare PCIe cards and apart from that the heise article does not mention the SPEC benchmark, where the results undoubtedly show the influence of software (drivers) more than of the underlying hardware.

This was my point, theres a lot of difference between results depending the software codes ( that i suppose contain specific optimizations ) and specific drivers ( who contain surely specific optimizations ), etc.. for be honest if theres benchmarks number who are really hard to compare, this case is surely on the top 3. Meaning machine who are configured can change a lot ( let alone OS, Hardware, links etc )... It is not like looking at a 3Dmark numbers. I really doubt that AMD will take too much risk on professional markets by openly lying on numbers. ..( and at the same time, im not quite sure they really want to show all during this presentation ).

 

[pharma]

I really doubt that AMD will take too much risk on professional markets by openly lying on numbers. .

Curious what professionals in this space thought about AMD's benchmarks after looking at the April 5, 2017 benchmarks posted by PCPerspective in their Pascal Quadro Roundup.

Edit: Conclusion - Wait for third party reviews.

 

[CarstenS]

This was my point, theres a lot of difference between results depending the software codes ( that i suppose contain specific optimizations ) and specific drivers ( who contain surely specific optimizations ), etc.. for be honest if theres benchmarks number who are really hard to compare, this case is surely on the top 3. Meaning machine who are configured can change a lot ( let alone OS, Hardware, links etc )... It is not like looking at a 3Dmark numbers. I really doubt that AMD will take too much risk on professional markets by openly lying on numbers. ..( and at the same time, im not quite sure they really want to show all during this presentation ).

Who said anything about lying?

 

[3dilettante]

I really doubt that AMD will take too much risk on professional markets by openly lying on numbers. ..( and at the same time, im not quite sure they really want to show all during this presentation ).

It's AMD's financial analyst day, so the target of the information isn't necessarily the professionals that would be buying the hardware. As far as things AMD has presented before for investors and analysts, selectivity in data points and possibly an optimistic lean on the data is comparatively tame.


On a tangent, for some reason I my brain keeps adding a Y after the F in the acronym for the event.

 

[lanek]

Who said anything about lying?

This phrase was not directed to peoples here.. but i have allready see some forum "afficonados" coming with that on other places..

 

[CSI PC]

Well, AMD seems to compare PCIe cards and apart from that the heise article does not mention the SPEC benchmark, where the results undoubtedly show the influence of software (drivers) more than of the underlying hardware.

AMD compared SPECViewperf results on the AMD Vega page but used Titan Xp instead of a Quadro with its drivers/library support: http://pro.radeon.com/en-us/vega-frontier-edition/
Using non Quadro GPUs for such applications can be a headache and why it scores quite low relative to Frontier - I would be curious if you find any professional visualisation publications actually testing a Titan X/Titan Xp that way, easier to find Quadro Pascal GPUs though.
But I guess I am a bit cynical because in both situations they used the 'lesser choice' GPU sold by Nvidia for specific task presented - lets be honest it is not really ideal comparison to take a more general top GPU that has massive FP64 and ignore the other top GPU that focuses on 30% more FP32 performance when one is only doing an FP32 DL benchmark as AMD did, or not use the Quadro product line as Nvidia design for professional visualisation but use the one with consumer drivers.
If they wanted to use one GPU from Nvidia then it should had been the Quadro GP6000, or if really wanting to be selective at least the Quadro GP100 (which would if accounting for the whole DeepBench results be about equal due to being a 10.3 TFLOPs FP32 model instead of 9.3 - caveat if going by Heise and still ignoring the discrepancy with other tests).

But like I said, most will not do a test with Titan X/Xp as it does not use the professional drivers/libraries designed for such work.

Here is one that did test the earlier Pascal Titan X, and as you can see it loses out most of the times by a fair margin to a Maxwell M6000 and the couple Titan X wins are not selected by AMD anyway, but it is a Pascal Titan against Maxwell Quadro.
http://www.tomshardware.de/geforce-quadro-workstation-grafikkarte-gpu,testberichte-241759-2.html

As an example:

And we can see how much the M6000 loses to the P6000 and Quadro GP100 on the 1st SolidWorks test chart.
Just to add neither myself nore Heise are saying AMD lied, just AMD were very selective what GPUs they used and how to calculate results, like I said though they are not the only ones to do that recently and we see often the arguments between Google-Nvidia-Intel.

Cheers

 

[Silent_Buddha]

At the risk of being critically pedantic, the "over the next couple of months" is a time period that started yesterday, so it's May 22nd + 2 months and not late June + 2 months.

Yes, if my post wasn't completely clear on that. Mainstream and potentially budget cards likely coming in July (as I said), or in other words within 2 months of FE.

Regards,
SB

 

[pharma]

Here’s how I’d parse that statement. As per AMD’s previous comments, the only Vega shipping in June will be the Frontier edition. Lisa Su is saying that Vega will launch across all platforms against the next couple of months, which could be read several different ways. If she means “The next couple of months after June,” then we’re looking at consumer launches in August and September (assuming “couple” is read loosely). If we read it more tightly, then we could see launches earlier, in July – August. But there’s very much a question as to where AMD will focus its efforts, and how it will price Vega.
...
If AMD starts pushing Vega into consumer markets in July and can keep the card on store shelves at reasonable prices, it’ll be a sign that whatever production issues may have occurred, they’ve been ironed out. But if it continues to focus on the highest end of the market, or to emphasize data centers and machine learning, it could mean the GPU will only see limited release to consumers. We won’t know which way the situation will break until AMD releases more concrete information or launches the GPU.

https://www.extremetech.com/computi...ch-dates-server-rollouts-talks-zen-follow-ups

 

[Silent_Buddha]

I'm not sure Aug. or Sept. would fit her quote of "very soon thereafter" for "enthusiast gaming platform". Even if it is tied to the shipping of FE products at the end of June that would imply July. I'm not sure how Sept. would even be applicable here as that would be 3 months after end of June which is outside of the "next couple of months" wording.

I'm sure AMD are very focused on trying to get product out and into OEM hands before the school shopping season begins (sometime in Aug.). So I'd imagine August would be the latest they have planned. Of course, things can happen and the best laid plans can go awry. So we'll have to see what happens over the summer. However, this was from their financial analyst day where they have to be as accurate as humanly possible. While they can be vague, to outright misrepresent when product is expected to be introduced wouldn't go over well.

Regards,
SB

 

[xpea]

I'm not sure Aug. or Sept. would fit her quote of "very soon thereafter" for "enthusiast gaming platform". Even if it is tied to the shipping of FE products at the end of June that would imply July. I'm not sure how Sept. would even be applicable here as that would be 3 months after end of June which is outside of the "next couple of months" wording.

English is not my native tongue, but it seems that for AMD fans, "couple of months" equal "2". Is it true ? For sure "couple" implies more than one, so we can discard that. But why 2 and not 3 and even 4 ? if it's 2 months, why don't say "Vega will be available in 2 months" ? My understanding of English says that couple can be 2, 3 and even 4, no ? Someone can confirm ?

 

[3dcgi]

A couple means 2 though when speaking informally some people intend it to mean some vague number more than one. Most people are sloppy with language when speaking informally.

 

[seahawk]

I would not bet on that. "the couple" means 2 persons as a pair. "A couple of ... " is usually used as a synonym for "a few", so yes I think it can be anything from 2-4 months.