videocardzAMD’s Greenland/Vega 10 GPU to feature 4096 Stream Processors

[AtenRa]

Is that how you see bottlenecks as absolutes? 🙂

And I guess you forgot what Raja spoke about at the PR event.

If Fiji was ROP limited it would have the same performance at 4K as Hawaii.

What Raja said at the event ?? link ??

 

[ShintaiDK]

If Fiji was ROP limited it would have the same performance at 4K as Hawaii.

What Raja said at the event ?? link ??

I am sure you can find a video yourself. Specially from an AMD event.

If ROPs are not the limit, why is Fiji performance then not showing this in average?

2816:176:64 vs 4096:256:64

45% more SPs.
45% more TMUs.
33% more bandwidth.
Same ROPs.

So you think 64 ROPs is enough for Fijis potential?

 

[AtenRa]

And I guess you forgot what Raja spoke about at the PR event.

I am sure you can find a video yourself.

😵

So you think 64 ROPs is enough for Fijis potential?

Again, if ROPs were the limiting factor, Fiji would have the same performance at 4K with Hawaii. Since Fiji distance itself at 4K from Hawaii, then its not ROPs the problem at lower resolutions.

 

[ShintaiDK]

Again, if ROPs were the limiting factor, Fiji would have the same performance at 4K with Hawaii. Since Fiji distance itself at 4K from Hawaii, then its not ROPs the problem at lower resolutions.

Maybe because Hawaii isn't ROP limited? And haven't there been cases where a 390X is mighty close to a Fury X?

Again, do you think Fiji is a balanced design that releases all its potential? And remember the numbers I listed before you answer.

Also ROPs isn't just about 4K 😉

 

[AtenRa]

Maybe because Hawaii isn't ROP limited? And haven't there been cases where a 390X is mighty close to a Fury X?

Again, do you think Fiji is a balanced design that releases all its potential? And remember the numbers I listed before you answer.

Fiji is not balanced because it was not meant to be balanced in the first place. Fiji was created for 4K, not 1080p.

 

[KaRLiToS]

Oups, erased by mistake sorry

 

[Schmide]

^ this

 

[rainy]

Sorry Rvenger. But in the pic below, you'll notice why I don't post here anymore. Can't stand trash on a forum. When a pessimistic person post 25% of the post count on a thread to bash a company and kill the hype of the people, I call that a troll and a shill. I have no pleasure in participating on a forums with behavior like this.

Anti-AMD obsession fuelling number of his posts - in just 3 days he has made over 80 posts and majority of them is related to AMD.

Btw, the only reasonable way to deal with such situation is put him on ignore list.

 

[Ma_Deuce]

He actually posts useful and insightful posts every now and again. unlike the blatent trolling of nvgpu and gamedev.

I think more ROPs actually would help Fiji out. It's not like what he is saying doesn't at least make a little sense.

 

[KaRLiToS]

He actually posts useful and insightful posts every now and again. unlike the blatent trolling of nvgpu and gamedev.

I think more ROPs actually would help Fiji out. It's not like what he is saying doesn't at least make a little sense.

I agree. But there are also positive things to post about, which he NEVER NEVER does.

 

[DooKey]

I thought user call outs were against the rules here? Attack the message not the messenger. Anyway, he may consider himself the offset to the AMD echo chamber this forum has become.

 

[Sweepr]

Sorry Rvenger. But in the pic below, you'll notice why I don't post here anymore. Can't stand trash on a forum. When a pessimistic person post 25% of the post count on a thread to bash a company and kill the hype of the people, I call that a troll and a shill. I have no pleasure in participating on a forums with behavior like this.

Would you rather participate on a forum with one-sided opinions instead? I must have missed the troll/shill part.

 

[ShintaiDK]

Kill the hype of the people? Interesting considering what a hype is.

Realistic expectations always gives the better outcome.

And I thought I posted that AMD would get 30%+ market share if they could match or get close to NVidia in performance/watt and deliver low power usage. And that Vega 10 would release unused potential with more ROPs compared to Fiji.

 

[Arachnotronic]

Attack the message not the messenger. Anyway, he may consider himself the offset to the AMD echo chamber this forum has become.

Truth. The pro-AMD posters around here routinely bully those who prefer (or even just use!) products from non-AMD companies. In their minds, you're just a "sheep" or a "sucker" for buying, say, a GTX 970 over an R9 390.

And, as I showed with a recent poll on these forums, pro-AMD posters significantly outnumber the pro-NVIDIA ones.

This has made participation in this forum quite unpleasant. "AMD echo chamber" -- sounds like the perfect characterization of these boards, DooKey.

 

[ShintaiDK]

A little note, 8GB HBM2 would mean either 4 2Hi(Half bandwidth) 2GB stacks or 2 4Hi 4GB stacks.

So if Vega 10 is 8GB, it will be the same bandwidth as Fiji with 512GB/sec.

1TB/sec requires 16 or 32GB HBM2. Its perhaps more likely that Vega 10 will be 16GB than 8GB.

 

[Piroko]

Would you rather participate on a forum with one-sided opinions instead? I must have missed the troll/shill part.

There's a point to be made though that writing 20 posts over a product that you have zero interest in in a thread that has barely 80 posts is far beyond stating your opinion.

And this has been going on for quite a while now, with AMD centric posters routinely attacking Nvidia cards on hardware flaws (970 memory, async debate - we get it, we got it months ago) and Nvidia centric posters routinely attacking - well, everything else, to be honest. I tried to also add two bullet points here, but I just couldn't think of what to boil it down to. Argumentation really got ridiculous.

As for trolling threads, case in point:

Truth. The pro-AMD posters around here routinely bully those who prefer (or even just use!) products from non-AMD companies. In their minds, you're just a "sheep" or a "sucker" for buying, say, a GTX 970 over an R9 390.

And, as I showed with a recent poll on these forums, pro-AMD posters significantly outnumber the pro-NVIDIA ones.

This has made participation in this forum quite unpleasant. "AMD echo chamber" -- sounds like the perfect characterization of these boards, DooKey.

This has made participation in this forum quite unpleasant indeed.

 

[Silverforce11]

If you guys don't want other points of views, particularly from pro-NV folks, then start threads in the AMD sub-section for AMD related news. Otherwise it's a free for all. We're adult enough to realize not to go ad-hominem.

And people who think Fiji is ROP bound are just wrong. It doesn't show the performance profile of a ROP bound GPU at all. It's completely opposite as it's relative performance goes up as the resolution goes up.

It's AMD's same front end, a single Command Processor design made for Hawaii, being tasked to handle a 45% increase in SP. It just can't do it in DX11.

But when examined in Ashes with DX12 Async Compute, it pulls far ahead of the 390X relative to it's SP count. That's because those idling SPs are being fed by the ACEs. If it was ROP limited, more work by the ACE fed SPs would still be bottlenecked by ROPs and thus you will not see more performance.

What's one of the improvements in Polaris? Yup, a new Command Processor, with pre-fetch and longer queue caches. Pretty obvious AMD knows where the bottlenecks are. Throwing more SPs at the problem without improving their hardware scheduler just can't yield decent gains.

 

[digitaldurandal]

A little note, 8GB HBM2 would mean either 4 2Hi(Half bandwidth) 2GB stacks or 2 4Hi 4GB stacks.

So if Vega 10 is 8GB, it will be the same bandwidth as Fiji with 512GB/sec.

1TB/sec requires 16 or 32GB HBM2. Its perhaps more likely that Vega 10 will be 16GB than 8GB.

Will it even be able to utilize that much VRAM effectively or is it due to limitations of the HBM2 architecture?

Seems like this would inflate the price unnecessarily and could be bad for performance\$.

 

[ShintaiDK]

Will it even be able to utilize that much VRAM effectively or is it due to limitations of the HBM2 architecture?

Seems like this would inflate the price unnecessarily and could be bad for performance\$.

It can use it. And for the pro segment its directly needed, including 32GB versions. This is also why Fiji is useless for the pro segment and AMD instead sell Hawaii chips with 16 and 32GB.

I wouldn't be surprised if 4GB is lower end, 8GB mainstream and 16GB high end for the next series from both camps.

Polaris 11 is 4GB and Polaris 10 is 8GB from the looks of it.

Rise of the Tomb Raider in DX12 managed to use 10.3GB VRAM. maybe not the best case for what the average will be. But the bar is going up.

 

[Silverforce11]

Rise of the Tomb Raider in DX12 managed to use 10.3GB VRAM. maybe not the best case for what the average will be. But the bar is going up.

Many games allocate as much vram as they can, but they don't need it. You can play RoTTR with real 4GB GPUs perfectly fine maxed.

Same for Dying Light, CoD Black Ops 3, or any other game that shows high "vram usage".

 

[ShintaiDK]

Yes, if you like 10FPS minimums in Tomb Raider DX12.

 

[Silverforce11]

Yes, if you like 10FPS minimums in Tomb Raider DX12.

You mean beta patch DX12? No thanks, I like 40FPS+ in DX11, with most scenes at 60 fps. 🙂

Even a 970 can achieve that at 1080p, if only it had real 4GB.

http://forums.anandtech.com/showpost.php?p=38126767&postcount=11

Not exactly a fast card, even the 380X does fine maxed out in RoTTR.

Look at that 4GB Fury/X even at 1440p maxed:

Even though the AMD Radeon R9 Fury X has the least amount of dedicated VRAM here it too does not differ much in performance using "Very High" textures. It seems that dynamic VRAM usage is not harming performance here.

While this game may use a lot of VRAM with "Very High" textures, it doesn't suffer in performance because of it.

http://hardocp.com/article/2016/02/29/rise_tomb_raider_graphics_features_performance/4#.Vvhx__l96Uk

^ That's after these guys went into the bench expecting Fury X to tank due to 4GB because the game was found to use a lot of vram.

Again, vram usage does not correlate to vram needed.

 

[KaRLiToS]

....

 

[Silverforce11]

This forum is filled with flame threads...look the threads with OPs asking for help...nobody helps...because members here want to bash, not help.

Hey, we help. In-between the crap talk, there's genuine care for fellow gamers. 🙂

http://forums.anandtech.com/showthread.php?t=2468138

etc. Pretty much all the posts that ask questions and advice, this forum takes care of quite well.

Don't let the negativity get to you. To me, part of the fun of upcoming product launches are to try and get the predictions nailed down or as close to accurate as possible before the launch.

 

[KaRLiToS]

Hey, we help. In-between the crap talk, there's genuine care for fellow gamers. 🙂

http://forums.anandtech.com/showthread.php?t=2468138

etc. Pretty much all the posts that ask questions and advice, this forum takes care of quite well.

Don't let the negativity get to you. To me, part of the fun of upcoming product launches are to try and get the predictions nailed down or as close to accurate as possible before the launch.

Sorry Silver, I agree with you. And I read the thread you linked and I have no doubt some members here like to help. You know I'm not pointing the finger at you bud. And you know who the usual bashers are 😉 .