• 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.
D.P. said:
The worrying thing for me is that with the RX580 AMD and/or GF has shown there is basically zero improvement in the process. Then again, GF even states that the latest 14nmFF+ has improvements in production costs but not in performance. And with the RX580 there eems to be different ASIC qualities with only some chips making the highest speeds which suggests to em there are sitll process problems.

So then it comes down to how likely a bigger, more complex chip than Polaris 20 going to hit the clock speeds? P20 runs at 1257MHz base clock and is often pulling about 220W during gaming and power increases rapidly with over-clocks. Vega is getting built on the same process. the MI25 was reportedly using 300W for 1.5GHz according to AMD specs, I'm really doubtful AMD want to put a 300w gaming GPU out.
Click to expand...
Architectural improvements.

Maxwell saw a huge reduction in necessary power draw on the same process as Kepler. Not saying we'll see the same leap here, but efficiency improvements could easily give them room for going bigger and faster without running into power ceilings.
 
alec said:
It isn't best. Ie Ryzen is not getting to 5GHz on air, and Polaris clocks are nowhere near Pascal. But it is good.
Click to expand...

Well Intel is leading the way in process technology, and you can't compare clockspeeds of different architectures really.
 
Seanspeed said:
Architectural improvements.

Maxwell saw a huge reduction in necessary power draw on the same process as Kepler. Not saying we'll see the same leap here, but efficiency improvements could easily give them room for going bigger and faster without running into power ceilings.
Click to expand...


There is a limit to how much architectural improvements are possible on the same process. Maxwell achieved a lot by massively cutting down Fp64 support to 1/32, Polaris (and Fiji) has already done that. Maxwell chips also got larger, so the 980ti chip was bigger than the 780ti while the chip size of the vega as picture is similar to the 1080ti (475 vs 471mm^2). Furthermore, the 28nm process used by Maxwell had undergone a lot of refinements and improvements but ti doesn't appear to be the case withe the GF 14nm process.

But yes, the real change will be architectural improvements. Maxwell achieved about a 30-35% boost but that was really quite remarkable. 35% can be achieved by cutting support for FP64, increasing die size 13%, decent process improvements and architectural improvements. AMD have the latter in their favor.
 
D.P. said:
There is a limit to how much architectural improvements are possible on the same process. Maxwell achieved a lot by massively cutting down Fp64 support to 1/32, Polaris (and Fiji) has already done that. Maxwell chips also got larger, so the 980ti chip was bigger than the 780ti while the chip size of the vega as picture is similar to the 1080ti (475 vs 471mm^2). Furthermore, the 28nm process used by Maxwell had undergone a lot of refinements and improvements but ti doesn't appear to be the case withe the GF 14nm process.

But yes, the real change will be architectural improvements. Maxwell achieved about a 30-35% boost but that was really quite remarkable. 35% can be achieved by cutting support for FP64, increasing die size 13%, decent process improvements and architectural improvements. AMD have the latter in their favor.
Click to expand...
If process shrinks stopped, there'd still be improvements in architectures for quite a while. Obviously diminishing returns would come into play pretty quickly, but there is not some strict limit on how much architectures can improve.

And you're right, Maxwell chips got larger. Yet drew even less power compared to the smaller Kepler chips, especially at the mid/upper mid range level, really showing that Maxwell's architectural efficiency improvements were even bigger than many really think.

I also agree Maxwell's improvements were remarkable, but my point is that you cant just look at Polaris and say, "That's it, that's what we have to scale TDP estimates on when talking about Vega". I'm not making any predictions myself, I dont know what will happen, but theoretically, notable efficiency gains are totally possible.
 
Seanspeed said:
If process shrinks stopped, there'd still be improvements in architectures for quite a while. Obviously diminishing returns would come into play pretty quickly, but there is not some strict limit on how much architectures can improve.

And you're right, Maxwell chips got larger. Yet drew even less power compared to the smaller Kepler chips, especially at the mid/upper mid range level, really showing that Maxwell's architectural efficiency improvements were even bigger than many really think.

I also agree Maxwell's improvements were remarkable, but my point is that you cant just look at Polaris and say, "That's it, that's what we have to scale TDP estimates on when talking about Vega". I'm not making any predictions myself, I dont know what will happen, but theoretically, notable efficiency gains are totally possible.
Click to expand...


I totally agree that efficiency can increase but Polaris 20 serves as a useful benchmark to indicate that the process hasn't improved thus it is a baseline that can still be extrapolated form. 150Mhz should be possible at 300w but I also think it is worth bearing in mind that the mroe mainstream parts might be clocked nearer 1200Mhz due to issues at GF. Clock speed is a combination of the process and the architecture, but architectural differences tht can increase clock speed typically reduce IPC.
 
As there is no sound news on vega just yet cant we do what we did in the Ryzen thread and make up a load of random rubbish so that WCCFT read it and report it again ? :D that was epic haha
 
D.P. said:
If it's clocked at 1200MHZ, then yeah 1070 performance would be realistic
Click to expand...

Ideal would be yea 1070 performance, good retail price for that level and it overclocks to 1080ti performance with some additional cooling :D
I dont mind where its placed exactly if development post retail is very good, its what they do with these new features that seems to matter most. Why even bother with these touted highlights like HBCC if its not going to leapfrog current tech.
If it turns out the old dx11 focus and plain brute force, the normal DDR memory and that whole current meta was the superior choice it's disappointing. But if in development they have pushed boundaries in some way then its a successful product
 
Seanspeed said:
Being negative or positive about AMD doesn't have any effect whatsoever on what AMD actually puts out if we're just talking us regular folks. No amount of praise and positive thinking will make Vega any better than it will be.

It's also very possible to want AMD to do well yet still be negative about how you'll think they'll do. Hopes vs realistic expectations - they are not the same thing.
Click to expand...

It does though. a lot were very negative about AMD drivers and the general consensus was AMD drivers bad and nvidia drivers great. When in fact couldn't be more wrong nowadays. But the negativity had people believe avoid AMD because of bad drivers. It does have a impact. For example, that could sway people from going for example 480/580 to a 1060 because the age old AMD drivers bad. Probably not as bad now as it was back a year ago. When i worked for a ISP for half a year there was so many misinformed people working there i was actually stunned. And i came to see they wasn't very uptodate on tech just what they see and hear from other people. I heard some one recommend a friend there a GTX 780 over a AMD 390 and i asked why he just said 780 is way better so i was like really why is that? He was silent for few seconds then said well one better drivers and more frequent driver updates. I was like nope drivers are not better maybe more frequent but not better. I then said for one 390 is a better performer, it performs better in DX12 and is more future proofed with a extra 5Gb of GDDR5 memory. But the 390 was a no brainer vs something slower with a lot less memory. He then told me he had a 780 and it was faster than his friend's 390. So i said could be anything he could have a slower CPU than you bottlenecking things. Does not change the fact the 390 is better. And he brought up the 390 being very hot and noisy. I was like it's just that negativity AMD seems to have nvidia is always better AMD is bad always hot and power hungry and bad drivers when yea AMD cards may draw slightly more power and be a couple degrees hotter than their nvidia counter parts but nothing that makes a considerable difference.

He then said well 980 is faster than 390 so i agreed but i said he would be better off looking at a 970 for similar money but still personally i would recommend getting a 390 for future proofing and especially as looking at 1440p and freesync monitors are cheaper.
He was so sure AMD cards were bad just because of the negativity around them at the time but didn't seem to know much because i would have recommended the 970 over the 780 at the time but for longevity i would suggest 390 at the price range he was looking at.
 
labbby said:
As there is no sound news on vega just yet cant we do what we did in the Ryzen thread and make up a load of random rubbish so that WCCFT read it and report it again ? :D that was epic haha
Click to expand...

A source who has seen the future of Vega has told me that it will run at 1.21 Giga.....hertz :)
 
LoadsaMoney said:
1gxCHM3.gif
Click to expand...

Lol, it doesn't quite work like that. On paper looks like roughly double the specs of the 480 so lets double the FPS of the 480. double the computer power of the 480 is 11.6Tflops which is around the compute power of the 1080ti at stock clocks but we all know it boosts way higher than that on its own meaning its faster. Every 1080ti will boost higher than this meaning a 1080ti is more like high end 13 to 14 Tflops of performance.
This guy has 480x2 (Vega) at either same or slightly faster than 1080ti.
It's going to be more likley slightly faster than a 1080 but falling short of a 1080ti.
 
r7slayer said:
Lol, it doesn't quite work like that. On paper looks like roughly double the specs of the 480 so lets double the FPS of the 480. double the computer power of the 480 is 11.6Tflops which is around the compute power of the 1080ti at stock clocks but we all know it boosts way higher than that on its own meaning its faster. Every 1080ti will boost higher than this meaning a 1080ti is more like high end 13 to 14 Tflops of performance.
This guy has 480x2 (Vega) at either same or slightly faster than 1080ti.
It's going to be more likley slightly faster than a 1080 but falling short of a 1080ti.
Click to expand...

Nah mate, this AMD we are talking about here, it will be 1070 performance at best innit. Don't get your hopes up bruv, AMD don't have da R&d money's so Vega is doomed. 1080ti is just sick blud, best drivers and performance!





















:p

On a more serious note, bet at the very least it will beat the 1080ti in DX12 and AMD games like doom. Oh and I agree, if anything driver wise AMD is ahead of Nvidia in my experience, nvidia driver take much longer to load and ui is very old. The only two things AMD drivers lack feature wise that I can see is adaptive v-sync and fastsync.
 
Status
Not open for further replies.
Back
Top Bottom