Vega Rumors

Bahanime said:
First, they said Zen won't be above 3Ghz. There's no way it's competitive against 5960X or 6900K for half the price...

Then the same bunch said there's no way Vega will clock high. They said it's a bigger chip so 1200mhz is it! It wouldn't be possible at 1500mhz, no chance in hell faster than that below 300W.. blah blah.

They claim there's no chance for Vega to reach Pascal perf/w, no way for it to match the 1080Ti, it's only around 1070 or 1080 level!

Now they say there's no way it's 16GB HBM2, because this and that, as if they know it all.

I wonder why these same people are such downers on everything AMD related. Really if you hate AMD so much, piss off out of these AMD related threads. Your negativity has been proven wrong many times, enough is enough?
Click to expand...


After awhile you get used to seeing the same names repeating it's not possible, you can tune them out or on occasion call them out on it if you like, but they live in their own world and your not going to kick them out of it. There is a reason some of them are now banned from the AMD CPU forum, cause all they did was derail threads. So my answer is stay focused on the subject and not them. Rumors and leaks are fun anyway, never know what is actually true or not and now we have 16GB and it could be HBM2 or perhaps a GDDR variant we didn't know about or AMD lied and they planned a 4 stack HBM2 model all along. Will know the actual truth in about month.
 
NKD said:
Is samsung in some sort of exclusive agreement with HBM2? I mean they have had this for a bit, what is stopping AMD from knocking on samsung's door to get more chips. If Hynix keeps fucking up and can't get their shit together to ramp up supply. I am sure there is no such thing as binding agreement that keeps screwing you over and delaying your product that hurts your position in the market. Unless ofcourse Samsung has some sort of exclusive agreement with Nvidia.
Click to expand...

Closest I can find https://www.forbes.com/sites/patric...14nm-manufacturing-with-samsung/#431760574e99
 
I don't see the HBM technologies as being standardized; in theory, anyone (who is not playing currently, like Micron maybe?) could make HBM and claim that it's 'HBM2' or 'HBM3' or something else. These HBM standards aren't being run through JEDEC, are they? [semi-rhetorical]

Currently, the only reason that there's the appearance of standardization looks like because there's only two players, Hynix and Samsung, and only two customers, AMD and Nvidia. That keeps things tidy for the moment.
 
N4CR said:
Well why did we just see a leak of a 64CU card with 16gb? Presuming it's legit (seems to be) then it's single GPU.
Either AMD has 8gb stacks or are using (like the Fury x many of you like to say it's a clone of) four stacks.... or have figured out a way to tack GDDR on there (unlikely).

Which one is it? I think 4 stacks is looking more likely at this point and it's been done before. Has AMD Specifically said Vega can't use 4 stacks?

edit: if it was vega20 memory pooling/etc 64CUs makes no sense
Click to expand...

Or its a dual card or simply fake/misread. The performance numbers for example doesn't match a 1600Mhz part that it claims to be. But we should be able to know within a month or so what actually comes out. And if it contains 16GB or not, 1600Mhz or not.
 
Last edited:
NKD said:
Is samsung in some sort of exclusive agreement with HBM2? I mean they have had this for a bit, what is stopping AMD from knocking on samsung's door to get more chips. If Hynix keeps fucking up and can't get their shit together to ramp up supply. I am sure there is no such thing as binding agreement that keeps screwing you over and delaying your product that hurts your position in the market. Unless ofcourse Samsung has some sort of exclusive agreement with Nvidia.
Click to expand...

There is no such thing as exclusive agreements. It was something a fanboy somewhere made up and everyone else seem to run with it because they dreamt of some kind of upper hand.

Whoever pays gets it, that's how it works for all the memory makers. If AMD is willing to pay they get it.

Assuming there is a limited amount, then the problem for AMD is that the other companies buying are using it for 10000$ FPGAs and likewise costly parts like GP100 and GV100. That may work out fine for a Instinct card, but certainly not a gaming card.
 
Bahanime said:
Why do you say it with such certainty, as if it's a fact? Are you privy to insider info that the rest of us do not have access to?
Click to expand...

We've known since Polaris that Shintai is Lisa Su's gypsy brother from another mother.
 
Shintai said:
Or its a dual card or simply fake/misread. The performance numbers for example doesn't match a 1600Mhz part that it claims to be. But we should be able to know within a month or so what actually comes out. And if it contains 16GB or not, 1600Mhz or not.
Click to expand...

Maybe fake too. Early/partner drivers are not always full speed.
The big pile of wtfs grows higher.
 
Bahanime said:
Why do you say it with such certainty, as if it's a fact? Are you privy to insider info that the rest of us do not have access to?
Click to expand...

Must be why Nvidia have used HBM2 for a year before AMD, not to mention certain FPGA makers, right?

Business is business, whoever pays the most get it. Plain simple.
 
cybereality said:
So a bunch of sites are reporting an unsubstantiated reddit post and not even quoting the source? Some fine journalism right there.

View attachment 25030

88f0f2bf2507cf431bf6b63e9f15a4e5b4205bd1ce57f995016c5e75265093a5.jpg
Click to expand...


Fire strike 10k, is around gtx 1080 levels, its not that great.
 
Bahanime said:
First, they said Zen won't be above 3Ghz. There's no way it's competitive against 5960X or 6900K for half the price...

Then the same bunch said there's no way Vega will clock high. They said it's a bigger chip so 1200mhz is it! It wouldn't be possible at 1500mhz, no chance in hell faster than that below 300W.. blah blah.

They claim there's no chance for Vega to reach Pascal perf/w, no way for it to match the 1080Ti, it's only around 1070 or 1080 level!

Now they say there's no way it's 16GB HBM2, because this and that, as if they know it all.

I wonder why these same people are such downers on everything AMD related. Really if you hate AMD so much, piss off out of these AMD related threads. Your negativity has been proven wrong many times, enough is enough?
Click to expand...


Who said 1200 was the highest, I remember everyone thinking 1500 is going to the max based on the instinct cards and the target flops.......

Don't try to make things up now about other people.

16gb HBM2 is out of question for Vega unless they are planning to put out a card that costs 1k or more. Its basic economics. 8GB 2 stacks will cost 100 bucks at least that is without the interposer and costs associated with that, add in another 50 bucks or so to be on the safe side, now you want 16 gb?

Think about it.
 
NKD said:
Is samsung in some sort of exclusive agreement with HBM2? I mean they have had this for a bit, what is stopping AMD from knocking on samsung's door to get more chips. If Hynix keeps fucking up and can't get their shit together to ramp up supply. I am sure there is no such thing as binding agreement that keeps screwing you over and delaying your product that hurts your position in the market. Unless ofcourse Samsung has some sort of exclusive agreement with Nvidia.
Click to expand...


Samsung has no exclusivity with nV. AMD has exclusivity with Hynix for HBM and that is why they can't get HBM 2 for Samsung. Look AMD cornered their future to Hynix and HBM, noone else did that but themselves, prior to Raja coming back, so he has no choice but to play ball with that agreement. They need to procure a certain amount of HBM from Hynix before they can get it from anyone else.
 
razor1 said:
Who said 1200 was the highest, I remember everyone thinking 1500 is going to the max based on the instinct cards and the target flops.......

Don't try to make things up now about other people.

16gb HBM2 is out of question for Vega unless they are planning to put out a card that costs 1k or more. Its basic economics. 8GB 2 stacks will cost 100 bucks at least that is without the interposer and costs associated with that, add in another 50 bucks or so to be on the safe side, now you want 16 gb?

Think about it.
Click to expand...

I browsed through several of the Vega threads, there were a lot of huff and puff about how it's 1070 or 1080 level, 1200 - 1300mhz (cos it's a big chip! and GCN doesn't clock well!).. pretty obvious there's a lot of naysayers that like to crap on anything AMD related.

Even your statement, "16GB is out of the question.. unless.. " Right, unless you know for a fact, don't say such things. Do you know for a fact that Vega must be priced so low that it cannot be profitable with 16GB HBM2? Economics aside.

Leaks and rumors be salty, just admit it could be right or wrong but this level of certainty by some without insider info, is appalling.

ps. If 1500mhz was that obvious, we wouldn't have to have so many posters gloating about the 1200mhz leaks earlier. Is it obvious? It should be. 25 Tflops FP16 with Vega should be at least 1525mhz from basic maths. And the next obvious thing, Pro cards clock less than consumer cards. Factor that too.
 
Bahanime said:
Sourced from SK Hynix? Because you left out the important fact, AMD + Hynix collaborated on HBM2 to market.
Click to expand...

Hynix haven't even been able to make any HBM2 products yet. So obviously no. I even wonder if they are going to try make any future HBM or just call it a day here after 2 major disasters.
 
Shintai said:
Hynix haven't even been able to make any HBM2 products yet. So obviously no. I even wonder if they are going to try make any future HBM or just call it a day here after 2 major disasters.
Click to expand...

See, you left out such an important detail, that AMD helped Hynix develop & bring to market HBM2.

I could easily suggest, such partnerships have benefits for the collaborators. Such as, first dibs to the best quality HBM2 modules, removing it from their catalog for other 3rd parties. But, ofc, it's only a suggestion. I wouldn't say it's certain because there's no evidence or fact to back it up.

You can wonder, sure. Just don't claim without evidence please.
 
Bahanime said:
I browsed through several of the Vega threads, there were a lot of huff and puff about how it's 1070 or 1080 level, 1200 - 1300mhz (cos it's a big chip! and GCN doesn't clock well!).. pretty obvious there's a lot of naysayers that like to crap on anything AMD related.

Even your statement, "16GB is out of the question.. unless.. " Right, unless you know for a fact, don't say such things. Do you know for a fact that Vega must be priced so low that it cannot be profitable with 16GB HBM2? Economics aside.

Leaks and rumors be salty, just admit it could be right or wrong but this level of certainty by some without insider info, is appalling.

ps. If 1500mhz was that obvious, we wouldn't have to have so many posters gloating about the 1200mhz leaks earlier. Is it obvious? It should be. 25 Tflops FP16 with Vega should be at least 1525mhz from basic maths. And the next obvious thing, Pro cards clock less than consumer cards. Factor that too.
Click to expand...

If the 1200 mhz scores are true, @1500 its still at 1080 performance levels, there is a 30% difference between the 1070 to the 1080 in performance, and 1200mhz to 1500 mhz is 25% difference lol.

You are mixing and matching what others have been saying to create your own narrative.

Dude how can you take economics aside when looking at premium products, 16gb of HBM 2 with one interposer and a large die GPU just that package alone will be in the 400 buck range, then AMD has to add in their profits lets say 30% then the retailer's profits lets say another 30%? Where are you at? And no one wants to make 30% on a premium product, they would want 50%. You are close to 700 bucks at 30% and that is without the pcb and other components.

16gb for a consumer card with HBM2 is out of reach unless they price it at 1k period, another words it has to be up to 1080ti or Titan Xp performance otherwise its a dud. Why would anyone invest such huge amounts to be a dud. And that open cl bench it doesn't look like it will come close to those performance ranges.
 
Last edited:
Bahanime said:
See, you left out such an important detail, that AMD helped Hynix develop & bring to market HBM2.

I could easily suggest, such partnerships have benefits for the collaborators. Such as, first dibs to the best quality HBM2 modules, removing it from their catalog for other 3rd parties. But, ofc, it's only a suggestion. I wouldn't say it's certain because there's no evidence or fact to back it up.

You can wonder, sure. Just don't claim without evidence please.
Click to expand...


Its only advantageous if they can get competitive products at similar price ranges to market, AMD hasn't been able to do that, where ever the problem is coming from AMD or Hynix, it doesn't matter, its a problem on AMD's side since its their products. We are well aware that AMD has exclusivity with Hynix for HBM, and without that being properly executed, they can't go anywhere else, even Raja stated that the cost of creating the infrastructure and channels for HBM must be recouped.
 
Last edited:
Dont expect any info today, its just analyst day. So it will be something like a lot of promises, a lot of marketing and a lot of hype.

The real deal seems to be may 31st for Vega.
 
N4CR said:
We've known since Polaris that Shintai is Lisa Su's gypsy brother from another mother.
Click to expand...
Careful because he is very quick to cry that he is being abused to the forum admins. I've been warned twice already.
 
Why so much obsession with 16GB?
there is around ZERO evidence that 8GB is too little for today games, or even new games for the lifetime of Vega at its intended target resolution range of 1440p-2160p.
The 2 resolutions more relevant for vega are 2560x1440 120hz and 4k 60hz, where Freesync price matters most. It needs better performance than 1070 in both resolutions, which even Shintai believes it will achieve. the biggest risk for Vega right now is not showing up for the fight.
 
Just seems pretty clear there will be a 16GB variant of Vega now, despite so many saying that would not happen. Not really a obsession and more is always better then less but most of us are having a good laugh at some people at the moment. That is why some people should stay away from the word impossible.
 
geok1ng said:
Why so much obsession with 16GB?
there is around ZERO evidence that 8GB is too little for today games, or even new games for the lifetime of Vega at its intended target resolution range of 1440p-2160p.
The 2 resolutions more relevant for vega are 2560x1440 120hz and 4k 60hz, where Freesync price matters most. It needs better performance than 1070 in both resolutions, which even Shintai believes it will achieve. the biggest risk for Vega right now is not showing up for the fight.
Click to expand...

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
 
geok1ng said:
Why so much obsession with 16GB?
there is around ZERO evidence that 8GB is too little for today games, or even new games for the lifetime of Vega at its intended target resolution range of 1440p-2160p.
The 2 resolutions more relevant for vega are 2560x1440 120hz and 4k 60hz, where Freesync price matters most. It needs better performance than 1070 in both resolutions, which even Shintai believes it will achieve. the biggest risk for Vega right now is not showing up for the fight.
Click to expand...

Because of the way HBM is configured and what we know so far of the Vega memory bus, having 16gb would imply that it would reach double the bandwidth of an 8gb setup.
 
Magic Hate Ball said:
Because of the way HBM is configured and what we know so far of the Vega memory bus, having 16gb would imply that it would reach double the bandwidth of an 8gb setup.
Click to expand...

It would just require 8hi HBM2 stacks. But no memory maker have shown any yet. And it doesn't double the bandwidth.

Vega 10 = 2 stacks
Vega 20 = 4 stacks (2019ish).
 
Shintai said:
It would just require 8hi HBM2 stacks. But no memory maker have shown any yet. And it doesn't double the bandwidth.

Vega 10 = 2 stacks
Vega 20 = 4 stacks (2019ish).
Click to expand...

So if there are only 4hi stacks, and they have 16gb of RAM... do the math, there would have to be a second memory bus or a wider bus.

Derp.
 
geok1ng said:
Why so much obsession with 16GB?
there is around ZERO evidence that 8GB is too little for today games, or even new games for the lifetime of Vega at its intended target resolution range of 1440p-2160p.
The 2 resolutions more relevant for vega are 2560x1440 120hz and 4k 60hz, where Freesync price matters most. It needs better performance than 1070 in both resolutions, which even Shintai believes it will achieve. the biggest risk for Vega right now is not showing up for the fight.
Click to expand...

New VR headsets in development are pushing for higher resolutions and maintain high frame rates. Here is the specs on the $300 Microsoft VR headset. Think of these early entries into VR as the 640 x 480 monitors of yesteryear.

Headset specs

2x LCD displays, 1440 x 1440
2.89” diagonal display size (x2)
Front hinged display
95 degrees horizontal field of view
Display refresh rate up to 90 Hz (native)
Built-in audio out and microphone support through 3.5mm jack
Single cable with HDMI 2.0 (display) and USB 3.0 (data) for connectivity
Inside-out tracking
4.0 meter cable
Sensors: Accelerometer, Gyroscope, Magnetometer, Proximity
Dimensions (L x W x H) 195.8 x 94.8 x 106.59 mm
Weight: 350 grams

Before you say, "VR who cares!" remember that we are getting new Nvidia libraries to push hardware, HDR, and more to push these GPUs. Also you can buy a Dell 8k monitor. Next year I expect that price to be sub 2K and VEGA to be within its refresh window.
 
2880x1440@90, ouch...

That's over 4MP, about half of 4k total pixels, and you're going to want to maintain 90FPS synced.
 
IdiotInCharge said:
2880x1440@90, ouch...

That's over 4MP, about half of 4k total pixels, and you're going to want to maintain 90FPS synced.
Click to expand...

Considering a card like the GTX1080 can get ~50 FPS at 4k in a game like Battlefield 1, I don't think hitting 90+ FPS consistently pushing half as many pixels in a less demanding VR game would be hard for that kind of hardware.

Lower end stuff... probably not so much.
 
It's the consistent 90FPS that concerns me more; 50FPS at 4k should mean 100FPS at half-4k just in terms of fillrate, but keeping consistently short (<11ms) frametimes will be key to the experience and a margin of just 10FPS on average isn't promising!
 
geok1ng said:
Why so much obsession with 16GB?
there is around ZERO evidence that 8GB is too little for today games, or even new games for the lifetime of Vega at its intended target resolution range of 1440p-2160p.
The 2 resolutions more relevant for vega are 2560x1440 120hz and 4k 60hz, where Freesync price matters most. It needs better performance than 1070 in both resolutions, which even Shintai believes it will achieve. the biggest risk for Vega right now is not showing up for the fight.
Click to expand...
It just people think there is a 16 gig HBM variant of Vega and while others think there isn't a 16 gig HBM variant. Personally I rather AMD release the damn thing so everyone can shut up.
 
razor1 said:
If the 1200 mhz scores are true, @1500 its still at 1080 performance levels, there is a 30% difference between the 1070 to the 1080 in performance, and 1200mhz to 1500 mhz is 25% difference lol.
Click to expand...

Sounds about right. Although I suspect we'll see a 1600 MHz part. Been hearing that number often enough to suspect there's some truth to it.

So maybe something like this for a lineup?

Vega light - 1200 MHz-ish. 1070 competitor
Vega regular - 1500 MHz-ish. 1080 competitor
Vega heavy - 1600 MHz-ish. Not really a 1080 Ti competitor... but a way for AMD to offer a bit extra for less money than a Ti. Cherry-picked silicon.

Which goes back to Kyle's guess some time ago that the top Vega will probably be between a 1080 and a 1080 Ti.

Don't think we're seeing 16GB on a consumer card, though. Maybe for the professional cards (or a quasi-professional card offered for the 'extreme' segment, like Titan Xp). I'd guess 8GB is our top end for consumer parts, at least for now.
 
The 1600MHz came about from people calculating the required clocks to hit 12.5 TFLOPs based upon what was known at the time and the 4096 cores, compared to Fiji.

However this was applied to MI25, and the MI25 is not watercooled but passive, meaning cooling from the server-node-blade chassis structure.
It is a lot to ask especially when Nvidia is more conservative with their Boost clocks and the limits they are at with HPC Pascal on the larger dies (even GV102 that is smaller than Vega).

If AMD manage 1600MHz and raw performance actually scales to the theoretical peak limit and importantly sustained, then that is one great engineering feat from AMD IMO.
Personally I am expecting performance around 1080 level.
CHeers
 
You must log in or register to reply here.
Back
Top