Vega/Navi Rumors (Updated)

[french toast]

Did you bother reading the context? He said he could put it in a small case with a 450w PSU. I was pointing out that the heat/power usage is very similar.

Yes you are correct, in that context.

 

[Tweak155]

Wow. Just noticed this thread is a year and nine months old. Terrible execution by AMD in this segment. Maybe ryzen diverted away a lot of their resources. After all this time with no high end product, they really need a home run, not just a decent competitor for the more than a year old 1080.

How is a thread started last year in Sept a year AND nine months old? Not saying it hasn't taken a long time, but no need to exaggerate.

 

[Malogeek]

Wow. Just noticed this thread is a year and nine months old. Terrible execution by AMD in this segment. Maybe ryzen diverted away a lot of their resources. After all this time with no high end product, they really need a home run, not just a decent competitor for the more than a year old 1080.

Rumor threads can often start 2+ years before actual products. Irrelevant.

 

[swilli89]

Considering how they hyped Polaris... I wouldn't just trust what they are hinting.

I am not trusting/hyping/doomsaying or any other hyperbolic action. I simply am withholding judgement until RX Vega is revealed.

 

[OatisCampbell]

I am not trusting/hyping/doomsaying or any other hyperbolic action. I simply am withholding judgement until RX Vega is revealed.

It's pretty obvious Rx will be within 10% of FE clock for clock.

If not, FE should not have been released unless AMDs goal was to anger their buyers and sell Titans.

 

[Topweasel]

AMD Navi will be manufactured using 7 nm, but will it still use GCN? Multi-chip architecture using the familiar 4096-core, 256-TMU, 64-ROP structure that does not look like it is working well?

I am not sure about 7nm but Navi is NCU not GCN.

 

[Malogeek]

I am not sure about 7nm but Navi is NCU not GCN.

NCU is the new name for the CU's, which are part of GCN. Literally "Next-Gen Compute Unit".

 

[tential]

It's pretty obvious Rx will be within 10% of FE clock for clock.

If not, FE should not have been released unless AMDs goal was to anger their buyers and sell Titans.

Some people no matter what will not make any judgements, guesses, or remote estimates of the rx Vega chip no matter what data we have today at our hands.
They will simply not hold any opinion regarding the performance range of the chip in any regard until its actually out.

Which I just don't understand then why they're in a rumor / tech theory discussion forum and thread.

If all you want to do is wait to see what a company has to say or waiting for a product to be released then I don't understand why you're doing in a thread like this.

Let us make our guesses as to where performance lies based on what we know and you can patiently wait for amd to release the product or release a press statement.

 

[Veradun]

Some people no matter what will not make any judgements, guesses, or remote estimates of the rx Vega chip no matter what data we have today at our hands.
They will simply not hold any opinion regarding the performance range of the chip in any regard until its actually out.

Which I just don't understand then why they're in a rumor / tech theory discussion forum and thread.

If all you want to do is wait to see what a company has to say or waiting for a product to be released then I don't understand why you're doing in a thread like this.

Let us make our guesses as to where performance lies based on what we know and you can patiently wait for amd to release the product or release a press statement.

Ok, let's play this game.

RX Vega will obviously be in a 9% ball around 1080Ti in 4K (average TPU's review chart)

 

[french toast]

Ok, let's play this game.

RX Vega will obviously be in a 9% ball around 1080Ti in 4K (average TPU's review chart)

Im going for 15℅ below in the same metrics.

 

[Topweasel]

NCU is the new name for the CU's, which are part of GCN. Literally "Next-Gen Compute Unit".

You are right it have been confusing NCU with the new arch that will be debuting with Navi. But specifically Navi will not be GCN.

 

[tential]

Is there any good reason to suspect amd will release Navi top down full lineup? I can't imagine so. Looks like they're high end/ low end cycle releases for awhile to come which means Navi would hold no interest to me at all.

 

[Magee_MC]

Is there any good reason to suspect amd will release Navi top down full lineup? I can't imagine so. Looks like they're high end/ low end cycle releases for awhile to come which means Navi would hold no interest to me at all.

If Navi is, as we believe, a modular system where smaller components can be put together on an interposer, then there isn't any reason for AMD not to release the whole stack at once. As far as I can tell, the only thing that would be holding back AMD setting whatever performance level that they want is the total power consumption. Actually they would be able to pretty much dial in any level of performance that they want to match NV from the smallest to the largest of the Volta offerings.

 

[tential]

If Navi is, as we believe, a modular system where smaller components can be put together on an interposer, then there isn't any reason for AMD not to release the whole stack at once. As far as I can tell, the only thing that would be holding back AMD setting whatever performance level that they want is the total power consumption. Actually they would be able to pretty much dial in any level of performance that they want to match NV from the smallest to the largest of the Volta offerings.

Ya you're right I focused a little too much on the current high end low end alternation I forgot that he's navi is supposed to be modular and should be quite interesting.
Is it feasible financially for amd to do so though? Or even worth it to abandon what they're currently doing?

Navi 11 releases first focusing on modular navi bringing all of the power savings side of the story (Polaris 11, what's clearly going to be Vega 11,etc.) and then navi 10 further improve and deliver the best and highest performance to the business consumer who will pay the largest?

May be possible for amd to release the whole lineup, but still could be good reasons why they don't such as the scenario I laid out.

Fiji and Vega could just be two duds before the modular chip similar to Ryzen in modularity is out.

 

[Bacon1]

If Navi is, as we believe, a modular system where smaller components can be put together on an interposer, then there isn't any reason for AMD not to release the whole stack at once. As far as I can tell, the only thing that would be holding back AMD setting whatever performance level that they want is the total power consumption. Actually they would be able to pretty much dial in any level of performance that they want to match NV from the smallest to the largest of the Volta offerings.

Yep, the Navi rumor is its going to be similar to Ryzen where its multiple smaller dies stacked to make larger ones / high end graphics. So might be through multiple physical GPUs ("seemless CFX") or maybe they'll have multiple dies per gpu, but without the normal headaches from single board CFX/SLI.

 

[Kenmitch]

Yep, the Navi rumor is its going to be similar to Ryzen where its multiple smaller dies stacked to make larger ones / high end graphics. So might be through multiple physical GPUs ("seemless CFX") or maybe they'll have multiple dies per gpu, but without the normal headaches from single board CFX/SLI.

When I suggested that years ago it was said to be impossible. Guess Zen opens the door for impossibilities.

 

[EXCellR8]

AMD stating the heat FE is getting doesn't reflect the gaming RX products... and I'm inclined to agree but we're still a month out.

 

[swilli89]

Vega is 484mm2. https://twitter.com/GFXChipTweeter/status/881370308694822913

 

[Bacon1]

Still waiting to see a reviewer kicking up the power target and just let the clock reach the 1600Mhz. Probably thats whats going to happen on the AIB boards.

We've seen a few running @ 1600. What I want to see is one downclocked to Fury X @ 1050 so we can see just how much (or not at all) has changed.

 

[.vodka]

We've seen a few running @ 1600. What I want to see is one downclocked to Fury X @ 1050 so we can see just how much (or not at all) has changed.

https://forums.anandtech.com/thread...updated-june-30.2508928/page-30#post-38965661

Fury X @Stock vs Vega FE @1052MHz and 800MHz HBM2. Less bandwidth (don't know why his HBM2 is only running at 800MHz) but same core clock. There is definitely something not right with Vega, even with less bandwidth the difference between the Fury X and Vega FE at the same core clock is way too big.

https://www.forum-3dcenter.org/vbulletin/showpost.php?p=11419198&postcount=7679
http://www.planet3dnow.de/vbulletin...giemessungen?p=5162951&viewfull=1#post5162951

As it is, it's a regression vs Fiji because there clearly is something VERY wrong somewhere. 484mm² isn't as outrageous for the performance we're seeing, it just leaves a lot to be desired now.

Maybe RX Vega changes this. I'll reserve further judgment until it launches, at the very least it deserves the benefit of the doubt.

 

[crisium]

Vega is 484mm2. https://twitter.com/GFXChipTweeter/status/881370308694822913

Still no reason it should not be trouncing GP104 (314mm2). Polaris 10 is only 16% larger than GP106 and the two trade blows. Can they harness whatever the heck they packed in there come RX Vega launch? They must or they die.

 

[tential]

Still waiting to see a reviewer kicking up the power target and just let the clock reach the 1600Mhz. Probably thats whats going to happen on the AIB boards.

I already posted this in the frontier edition thread. Results looked promising. Someone will hopefully repost for you if you can't find the YouTube video posted, it was literally labeled oc results or something so you'll be able to see and judge the results yourself

 

[swilli89]

Still no reason it should not be trouncing GP104 (314mm2). Polaris 10 is only 16% larger than GP106 and the two trade blows. Can they harness whatever the heck they packed in there come RX Vega launch? They must or they die.

1. Exactly. People just need to put 2+2 together. AMD and Nvidia run full simulations of chips before they are ever taped out. They know to a fairly high level of confidence how their chip will perform. They did not regress so extremely from Polaris.

2. Eh.. not really. If Vega cores themselves are solid then AMD will still be the only company on the planet that can provide a high performance APU. 1024 down clocked Vega cores with a 4C zen cluster will be an absolute monster.

 

[Bacon1]

https://forums.anandtech.com/thread...updated-june-30.2508928/page-30#post-38965661

As it is, it's a regression vs Fiji because there clearly is something VERY wrong somewhere. 484mm² isn't as outrageous for the performance we're seeing, it just leaves a lot to be desired now.

Thanks, unfortunately those were done with two different systems so it's not 100% accurate, but it's something at least. If Ryzen wasn't so finicky about its RAM memory speeds it would be easier to say it's purely GPU differences but without knowing what speed ram was in each machine it's hard to rule it out.

 

[3DVagabond]

When I suggested that years ago it was said to be impossible. Guess Zen opens the door for impossibilities.

Right. Years ago it was impossible. Nobody can say something is never going to be possible. People were predicting years ago that AMD was going to do it years ago.