Vega Rumors

How's your crow taste Shintai ?

Over what? Some limited edition card that got the checkboxes with a price tag to follow? And the HBM2 is clocked at 1875Mhz. Thank god for Samsung eh? ;)

Seems like you are going to wait a bit longer than expected for your gaming Vega. Good thing you are used to #waitforvega :D
 
Last edited:
Seems like they got their hands on a few 8-Hi stacks. PCI-E Tesla P100 is rated for 9.6 teraflops FP32. Vega Frontier Edition (I won't even bother lol) with its ~13 Tflops of Fp32 and ~25 Tflops (more like 12.5 fp32 apparently) is ~30 faster in deepbench which is fp32 based as per their description on github.

Interestingly enough if you go to the vega website they suddenly stop comparing Vega to the Tesla P100 and switch to the Titan Xp when it comes to specviewperf. Hmm, i'm sure that's unrelated to wanting to compare pro drivers on AMD vs geforce drivers on nvidia, because AMD is ethical.
upload_2017-5-17_1-19-23.png


A more honest comparison would have pitted the Quadro P6000 vs Vega Frontier Edition. Full 3840 cores running at quite higher clock speeds than Tesla P100 which is a cut die running pretty low clocks due to it having to fit into tight power budget (1:2 DP chip don't forget) and no AIO watercooling.

Depending on which slide you choose to look at the P100 scores 122 or 133.

e2d25c5d40bd.jpg

e913a0ad7ce1.jpg

upload_2017-5-17_1-43-29.png

.eJwFwdsNwyAMAMBdGABjXoVsgwgiURMbgaN-VN29d1_1zEtt6hAZawPYz1V57noJz9Kb7sz9amWcS1e-oYiUetyNZAGmEP3LhmBsijYHE8ChxxS9w-R8xJxNhIfexB_Sg7r6_QEHzSLh.4CCJPSOmHgwhOmejKMrhvy_dWb0
 
Last edited:
Seems like they got their hands on a few 8-Hi stacks.
Don't tell that to Schiptai he says it is impossible, and he will point out the error of your ways. Then he will hash tag you with a tweet ala Trump. (he could be Trump).
 
Don't tell that to Schiptai he says it is impossible, and he will point out the error of your ways. Then he will hash tag you with a tweet ala Trump. (he could be Trump).

Shintai said it would not be possible for a gaming card to my understanding. These will have very limited volume and they will not even start shipping till late june. 8Hi stacks are not easy to come by.

Gaming cards won't have 16GB (I sincerely doubt it, but i'm more than willing to be surprised)
 
Over what? Some limited edition card that got the checkboxes with a price tag to follow? And the HBM2 is clocked at 1875Mhz. Thank god for Samsung eh? ;)

Seems like you are going to wait a bit longer than expected for your gaming Vega. Good thing you are used to #waitforvega :D

Whoops, that isn't the right response.

The correct answer is...

"I'm sorry guys, I clearly got this one wrong"

Shintai
 
Or he might just go back and edit his posts. Change the white to black.
Whoops, that isn't the right response.

The correct answer is...

"I'm sorry guys, I clearly got this one wrong"

Shintai


Aw, poor Shintai doesn't know how to gracefully admit being incorrect. It's almost as if being right or wrong isnt her/his ultimate goal...
 
Nevertheless, it shows you that you don't need to buy new hardware to do VR. Most gamers seem to be enjoying 1080.

Crank them down enough and you can use your phone!

I'd agree that a 1080 is 'enough', for now- but VR is going to push the requirements fast.
 
Because going by actual recent catalogue that came out this month there is no 8GB stack HBM2 and AMD only showed a 2-stack GPU at their event with reference to 4-stack being 2018 by certain 'leak' slides that for most are happy to use when it is other data-information.
Meaning a high chance it is faked (especially as it has zero true benchmark figures for a card that is listed unlike Fiji when that leaked on there) and with a figure for the only provided test in the whole of the suite that tbh is very wrong.

So makes it difficult how to treat the rumour of 16GB/1600Mhz that is generated from that benchmark entry.
Cheers

So, after the reveal today, can you guys who nay-say all the time about AMD calm down for once?

Next time don't sound so certain when you don't actually KNOW.
 
Crank them down enough and you can use your phone!

I'd agree that a 1080 is 'enough', for now- but VR is going to push the requirements fast.
Cold you elaborate on how fast is fast? I have been waiting for a long time now, so I am not sure what fast means.
 
It's a financial call they ain't sharing shit other than "Vega is on track for Q2 release".

Oh really? You seem to be so certain.

That was the first Vega product announced, with specs and launch dates. More to come at Computex for consumer Vega no doubt.

16GB HBM2 wasn't possible for AMD, because apparently nobody makes it for AMD to access according to some of you AMD haters. They've proven you all wrong. Instead of admitting you were wrong or being gracious, like "kudos to AMD", you people disgust me with your endless anti-AMD crusade.

Btw, RX Vega Frontier Edition = Founders Edition = Vega FE. Ring a bell?
 
I'm a bit disappointed no talk of the consumer cards, but at least we know some of the leaks were true.
 
Cold you elaborate on how fast is fast? I have been waiting for a long time now, so I am not sure what fast means.

Let me get my crystal ball...

Look, the trend in VR is greater resolution. We've talked about 1440x1440x2, but it'll keep going up; further, game graphics will keep getting pushed, making each of those pixels more costly, meaning more hardware needed to hit 90FPS. This isn't trivial, and we've only barely started to realize the capabilities of modern gaming engines, which can scale well beyond what current hardware can handle.

The only moderating factor, as it always is, will be the console development cycle, but even that is being accelerated!
 
Let me get my crystal ball...

Look, the trend in VR is greater resolution. We've talked about 1440x1440x2, but it'll keep going up; further, game graphics will keep getting pushed, making each of those pixels more costly, meaning more hardware needed to hit 90FPS. This isn't trivial, and we've only barely started to realize the capabilities of modern gaming engines, which can scale well beyond what current hardware can handle.

The only moderating factor, as it always is, will be the console development cycle, but even that is being accelerated!
Right, clear as FUD, I mean mud. Really, give me your guesstimate for a time frame.
 
I'm a bit disappointed no talk of the consumer cards, but at least we know some of the leaks were true.

AMD launched Polaris gaming cards at Computex last year. A year before that they launched Fury X gaming cards at Computex.

I don't know why people find it so difficult to see, with Computex coming up in 2 weeks, that AMD would repeat the same thing again and launch a gaming Vega there, or hell, paper launch it.
 
Seriously, you're trolling now- there's no 'fear', or 'uncertainty', or 'doubt', concerning the vector of VR.
You vehemently argued that it is coming fast. But when I asked you to quantify that, guesstimate even, you turn tail and run. Be brave, give a number so we can have it for the record and laugh at you when you are wrong.
 
Seriously, you're trolling now- there's no 'fear', or 'uncertainty', or 'doubt', concerning the vector of VR.
And, there is FUD. Fear of missing it, falling behind, or wasting money. Uncertainty as to the compatibility, platform longevity etc. Doubts in lots of forms.
 
And, there is FUD. Fear of missing it, falling behind, or wasting money. Uncertainty as to the compatibility, platform longevity etc. Doubts in lots of forms.

Missing what? Falling behind what? Wasting whose money?

You're talking, but not to me, regardless of who you're quoting.
 
Someone earlier asked how could 16 GB of VRAM be usable? I was just pointing out that newer VR headsets are entering the 4K realm. I think that 4K resolution per eye would require a lot of VRAM. I don't think that is derailing the topic.

Do you?

It's a good post. Many higher res OLED panels suitable have been out for a while but not adopted. Much like high Hz 4k panels.
The driving tech (FPGAs etc) is expensive and cost prohibitive in small volumes. E.g. Realtek won't release it's Dp1.4 chips to end users unless you got a million bucks and are e.g. ASUS etc.
E.g. currently to drive a 4k high refresh rate device, you're looking minimum 500USD (even at moderate volumes) just to drive it. Not including the actual display and rest of crap.
Don't ask how I know this :/


Re:Vega/16gb/etc It's okay to be graciously wrong too and admit you fucked up. It happens.
As I said earlier, many companies do not list their full capabilities, stocks and offerings, if you believe their public facing website you will always be mislead.
E.g. some of the shit I price up I have to sign an NDA first before they'll even tell me e.g. they can offer a competitive solution which is NOT listed on website.

What surprised me the most was SSG availability. Wasn't expecting that in any volume till Navi however they had been testing SSG the last 6 months or so prior.
 
E.g. currently to drive a 4k high refresh rate device, you're looking minimum 500USD (even at moderate volumes) just to drive it.

actually a company has a multi purpose board on catalog since the end of 2016 capable of 4k 120Hz. Not sure how much they charge for it:

https://sg.finance.yahoo.com/news/digital-view-announces-120hz-4k-134500652.html?.tsrc=xl
http://www.digitalview.com/media/manuals/svx-4096-2x-manual-rev201-19aug2016.pdf

On hardforum, the first screenshot of a board driving 4k 120hz without frameskipping dates from 29 jan.

20TFLOPS GPUs are a market need, without them, there will not be much demand for such boards.
 
I'm a bit disappointed no talk of the consumer cards, but at least we know some of the leaks were true.

Some people have already been saying that Vega will be an anti climax (adoredtv) . I'll stick by what I said earlier, not expecting that price nor performance will be something that are key to Vega. In that way when I'm wrong I do have something to look forward to :) .
 
existing stuff

20TFLOPS GPUs are a market need, without them, there will not be much demand for such boards.

The first one doesn't have freesync, plus I can't say much other than 4k/120 8 Bit alone is 30Gbps of signal - 1 billion pixels a second. DP cables are 32Gbps incl overhead. Basically you need dp1.4 + DSC or multi cables and that's where it all gets expensive and complicated, especially if you want active sync (pretty much impossible). If you're not a millionaire, 500-600USD plus one off costs is probably a low estimate too for 1.4 as you have added costs on top too incl source and just talking electronic section, no panel, housing etc...Good luck sourcing another solution (we've tried sourcing almost all of them).
Regarding #2 there will be a decision soon on the fate of this project so expect some news in coming days. My approach to the overall design side will revolutionise the entire concept of the monitor industry, so it might just see light of day regardless.

20GFLOPS also is not needed if you're not playing Crysis 10. It also means you have a future ready monitor when 20GFLOPs does come out.
 
Are you trusting AMD's notoriously bad PR images or your own eyes? I see 8+6
My own eyes, and i was asleep. Yeah, 8+6, of course. Still 250W+ TDP card with GP104 and even cut GP104 performance in geometry-heavy load.

I am... less than impressed to say the least. Granted, as was said, AMD always sucked in that, so i guess it should match full GP104 in games... i hope, at least.
 
Where are all the people who claimed that 16GB HBM2 was a foolish fanboy dream now?
Some of us including myself was saying that it cannot be 8-Hi for several reasons; from an electronics perspective the heat generated for the bottom die and logic(plus height) and also it being taken out of all the SK Hynix memory catalogues and part references where other products up to Q4'17 such as GDDR6 are given.
I am still not sure how they are managing 16GB HBM2 tbh nor the 2Gbps, because it seems SK Hynix is having problems even launching 4GB density 1.6Gbps HBM that keeps getting delayed.

SK Hynix were meant to alse be developing 8GB density, but this has been missing from every memory manufacturer as well along with 8-Hi; different engineering challenges.
Cheers
 
My own eyes, and i was asleep. Yeah, 8+6, of course. Still 250W+ TDP card with GP104 and even cut GP104 performance in geometry-heavy load.

I am... less than impressed to say the least. Granted, as was said, AMD always sucked in that, so i guess it should match full GP104 in games... i hope, at least.
I think 8+6 is 300W in this instance with Frontier.
75W motherboard
75W 6pin
150W 8pin

Yeah sorry it may be a bit pedantic I know, because in reality 8+6 is spec as 225W/300W.
Cheers
 
By releasing a dual 8-pin card that performs like GP104? They have proven us all right, in fact!

Also, i am glad Fool's Edition is also applicable to AMD cards too now.

Goal post shifting much brah?

You guys were wrong on 16GB HBM2, didn't even admit it and have to gall to try and shift the discussion into BS territory again.

Performs like GP104? LOL. Prepare to be wrong again, but you naysayers will find something else to twist with your negativity won't you?

RotTR @ 4K, putting up FPS up there with 1080Ti, with ONLY 2GB vram enabled. "But but what are the settings?!" they cried... ignoring the fact AMD's past live game demos at these events have been at the highest available preset, always. Likewise for Sniper Elite 4 @ 4K, which is actually faster than 1080Ti given it's results are on par with 1080Ti running above 1.9Ghz boost clocks.

Honestly can't you naysayers for once say a good thing? It's actually GOOD for you if AMD is competitive, since you can get your beloved corporation's next-gen Volta for cheaper if Vega is up there. Do you really enjoy paying more for your GPUs just because it's NVIDIA?

Really, ALL of you who love NVIDIA GPUs so much should be cheering for Vega to knock it out of the park, it's in your best interest.

Unless ofc, you're not regular consumers/gamers, but major share investors for NV or work for them... :/
 
Btw for those curious about 1x8 pin + 1x6pin or 2x8pin:

Different variants / SKUs.

AMD-Radeon-VEGA-Frontier-Watercooled.jpg


AMD-Radeon-VEGA-watercooled.jpg


The card they demo was the 1600mhz 16GB edition (per Computerbase's coverage).

Linux driver leaks show several device IDs, one of which match this configuration, it is in the middle of the Vega 10 stack with a few other variants above it.
 
Seems like they got their hands on a few 8-Hi stacks. PCI-E Tesla P100 is rated for 9.6 teraflops FP32. Vega Frontier Edition (I won't even bother lol) with its ~13 Tflops of Fp32 and ~25 Tflops (more like 12.5 fp32 apparently) is ~30 faster in deepbench which is fp32 based as per their description on github.

Interestingly enough if you go to the vega website they suddenly stop comparing Vega to the Tesla P100 and switch to the Titan Xp when it comes to specviewperf. Hmm, i'm sure that's unrelated to wanting to compare pro drivers on AMD vs geforce drivers on nvidia, because AMD is ethical.
View attachment 25120

A more honest comparison would have pitted the Quadro P6000 vs Vega Frontier Edition. Full 3840 cores running at quite higher clock speeds than Tesla P100 which is a cut die running pretty low clocks due to it having to fit into tight power budget (1:2 DP chip don't forget) and no AIO watercooling.

Depending on which slide you choose to look at the P100 scores 122 or 133.

e2d25c5d40bd.jpg

e913a0ad7ce1.jpg

View attachment 25126
.eJwFwdsNwyAMAMBdGABjXoVsgwgiURMbgaN-VN29d1_1zEtt6hAZawPYz1V57noJz9Kb7sz9amWcS1e-oYiUetyNZAGmEP3LhmBsijYHE8ChxxS9w-R8xJxNhIfexB_Sg7r6_QEHzSLh.4CCJPSOmHgwhOmejKMrhvy_dWb0
Thanks for the charts. However from the presentations, it seemed like he was saying this card is replacing Radeon Fury X.. Page # 33 on the PDF. http://phx.corporate-ir.net/Externa...WxkSUQ9LTF8VHlwZT0z&t=1&cb=636305760276823867
 
Goal post shifting much brah?
Nope, just stating how it is.

You guys were wrong on 16GB HBM2, didn't even admit it and have to gall to try and shift the discussion into BS territory again.
Quote me. Oh, you cannot, how bad.

Performs like GP104? LOL. Prepare to be wrong again, but you naysayers will find something else to twist with your negativity won't you?
It's a fact, it performs like cut GP104 and GM204 (rofl) in SPECviewperf.
 
Back
Top