• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD VEGA confirmed for 2017 H1

Status
Not open for further replies.
Ah wow that is very impressive if that's final TDP.

Got good feelings about Zen, so disappointment is inevitable :P

Na I think this gonna be good..

For me its still up the the air, but when questions are asked those are the answers :p
One glimmer is that AMD have been working on this for 5 years, they have had to deal with an architecture that is very inefficient and try to build some efficiency into it.

They have actually had a lot of success with that, Excavator, which is the last irritation of Bulldozer on 28nm is not far off the same efficiency as Intel on 14nm, what's more the Integer performance is actually slightly higher than Haswell, 'Integer' not floating point, altho that is also up significantly on Bulldozer.

AMD have had a lot of experience making a bad architecture reasonably good, they will have learnt a lot from that.
 
Ah wow that is very impressive if that's final TDP.

Got good feelings about Zen, so disappointment is inevitable :P

Na I think this gonna be good..

It is very important for everyone, even intel users that Zen is very good.

At the moment intel are not really increasing performance with their CPUs. All they are doing is adding extra cores and charging way too much for it.

The 6950X should be no more than £800 and the rest of the Broadwell E range closer to £250 to £300.
 
But we don't care about TDP for a desktop system.

What we need to know is how well the latest GPUs and games run.:)

You asked the questions, those are the answers.

It has a lower TDP than Broadwell, the IPC is a floating point performance in Blender is slightly higher than Broadwell.

Games also depend on Floating Point performance, i agree we need actually game benchmarks, but questions answered; from what we know so far there is no reason to write it off.
 
You asked the questions, those are the answers.

It has a lower TDP than Broadwell, the IPC is a floating point performance in Blender is slightly higher than Broadwell.

Games also depend on Floating Point performance, i agree we need actually game benchmarks, but questions answered; from what we know so far there is no reason to write it off.

I am not writing it off at all.

For desktop use most people don't care about TDP or Blender, they want to know how the latest games will run.

If AMD can get peoples games to run better than intel can, it will do them both a favour. Intel have been guilty of pushing out very poor performance increases for way too long.
 
I am not writing it off at all.

For desktop use most people don't care about TDP or Blender, they want to know how the latest games will run.

If AMD can get peoples games to run better than intel can, it will do them both a favour. Intel have been guilty of pushing out very poor performance increases for way too long.
Intel are just already at the leading edge of things and have thus hit heavy diminishing returns earlier.

They aren't just sitting around twiddling their thumbs. They still pump more into R&D than anybody else in the industry by far.
 
Intel are just already at the leading edge of things and have thus hit heavy diminishing returns earlier.

They aren't just sitting around twiddling their thumbs. They still pump more into R&D than anybody else in the industry by far.

When 3 or 4 year old CPUs perform close to the latest ones I would say intel are sitting on their backsides.

Compare a 3 or 4 year GPU to a Pascal Titan for example and it is a total non contest.

A GTX 680 v a Pascal Titan is such a mismatch not even a wrestling promoter could get away with it.:D
 
Zen 8 Core 16 Thread 95 Watt TDP
Breadwell 6900K 8 Core 16 Thread 140 Watt TDP.

Ah wow that is very impressive if that's final TDP.

Got good feelings about Zen, so disappointment is inevitable :P

Na I think this gonna be good..

As far as I recall from the events they said competing IPC for similar TDP, with Summit Ridge starting at 95W TDP.

So I assume that means the lowest TDP Summit Ridge Zen chip will be 95W, but can get upto 140W as well.

If Zen is more efficient than that it'll be brilliant really, even if they end up being 5-10% slower than Kaby Lake next year IPC wise.
 
As far as I recall from the events they said competing IPC for similar TDP, with Summit Ridge starting at 95W TDP.

So I assume that means the lowest TDP Summit Ridge Zen chip will be 95W, but can get upto 140W as well.

If Zen is more efficient than that it'll be brilliant really, even if they end up being 5-10% slower than Kaby Lake next year IPC wise.

What makes you think Kabylake will see twice the performance uplift to all Intel's previous core step ups?

Actually that is the TDP of the 8 core, scaling from 15 Watts all the way upto 95 Watts.
 
Last edited:
Thats presumably the 3ghz model though, just wonder if their clock speeds scale as well as intels.

That's extremely unlikely. Intel's 14nm process is still slightly superior. But Zen will no longer have a huge process handicap to overcome, and judging by that Blender demo it will be on par in IPC and floating point performance as well.

The whole point about Zen is not high clocks though, but throughput per watt. In server environments high clocks don't matter as much as scaling. Zen seems like it will do well in that market.

For gaming though (and especially DX11 gaming where single thread performance is king), the crown will remain with Intel, no doubt.
 
You mean YOU don't care about TDP. Many do Kaap - it's very ignorant to assume that everyone else has the same desires as you.

Kaap did care until i told him the TDP :D


Throw a Pascal Titan into the motherboard and see how efficient Zen runs.

A 6950X is not a fantastic CPU for gaming because it has 10/20 cores/threads. It is great for gaming because it can trade blows with a 6700k on individual cores/threads for efficiency.

AMDs target has to be to make their CPUs more efficient.

snip
 
Back to the topic at hand.

Vega.

Polaris is just an improved Tonga, per shader clock for clock its 6.5% faster, the RX 470 is like an R9 380X Rev-2, its like Piledriver is to Bulldozer. (FX-8350 is to FX-8150)

AMD could have just made a bigger version of Polaris to replace their Fury-X, the fact that they didn't makes me think Polaris is just a stop gap to keep some money coming in while they work on a ground up redesign architecture. much like Zen,

That would be Vega.
 
Last edited:
What makes you think Kabylake will see twice the performance uplift to all Intel's previous core step ups?

Actually that is the TDP of the 8 core, scaling from 15 Watts all the way upto 95 Watts.

Not so much Intel seeing such an uplift, but AMD falling short due to not being able to match clock speed.

IPC matching Broadwell is all good, but if 14nm from Global Foundry can't get the speed up to similar clock speeds as intel, they'll still be that much slower in cases.

The main thing that concerns me is ole GF, and their process.

Back to the topic at hand.

Vega.

Polaris is just an improved Tonga, per shader clock for clock its 6.5% faster, the RX 470 is like an R9 380X Rev-2, its like Piledriver is to Bulldozer.

AMD could have just made a bigger version of Polaris to replace their Fury-X, the fact that they didn't makes me think Polaris is just a stop gap to keep some money coming in while they work on a ground up redesign architecture, much like Zen, that would be Vega.

That's what I believe as well.

We have to remember that even with NVDIA's original roadmap Pascal wasn't around, it was supposed to go from Maxwell straight to Volta; but 20/22nm didn't pan out.

http://i.imgur.com/Y7QSqV7.jpg

So they put out a die shrink of Maxwell with some architectural improvements.

AMD's new FirePro prototype with with SSD contraption is going to use Fiji, but sadly they didn't get out their version of "Pascal" for Fiji. That's a shame really.

It would have made a decent upper midrange competitor to Pascal, if they managed to shrink and improve Fiji a bit.
 
Last edited:
LOL good point. Really doesn't know what he's saying does he.

Just to clarify

For a gaming PC I don't care about TDP on the CPU, I just want performance.

If anyone is going to worry about a few extra watts on a gaming PC CPU it works out as pennies per week.

And you really should not be so rude in your posts either.
 
Not so much Intel seeing such an uplift, but AMD falling short due to not being able to match clock speed.

IPC matching Broadwell is all good, but if 14nm from Global Foundry can't get the speed up to similar clock speeds as intel, they'll still be that much slower in cases.

The main thing that concerns me is ole GF, and their process.



That's what I believe as well.

We have to remember that even with NVDIA's original roadmap Pascal wasn't around, it was supposed to go from Maxwell straight to Volta; but 20/22nm didn't pan out.

http://i.imgur.com/Y7QSqV7.jpg

So they put out a die shrink of Maxwell with some architectural improvements.

AMD's new FirePro prototype with with SSD contraption is going to use Fiji, but sadly they didn't get out their version of "Pascal" for Fiji. That's a shame really.

It would have made a decent upper midrange competitor to Pascal, if they managed to shrink and improve Fiji a bit.


It costs time and money to die shrink, even up size what you already have, Polaris, i think AMD had to do something to address to question of relevance and the efficiency of their performance end GPU's, thats how Polaris was born, its just a replacement for the 380/X 390/X.
Given thats where AMD's largest share of the market is it was necessary.

Everything else, every penny they have, every designer and engineer available to them is 100% committed to Vega and Zen.

I think that is the right thing to do.

On Glofo, yeah they don't have a great rep for getting mature processing quickly, but once they do they are pretty good, the FX-8150 was really not good at clocking at all unless you got a golden one, once the same process once matured the FX-8350 clocked really well despite the sheer size of it, it is a huge chip clocking regularly to 4.8Ghz, i have seen a few at 5.3Ghz with custom loop AIO's.

Having said all of that, yes Zen is running at 3Ghz, but it is an engineering sample, and while i'm not using this as perspective, i think i'm right in saying Broadwell engineering samples were only running at 2.2Ghz.
 
Last edited:
Status
Not open for further replies.
Back
Top Bottom