• 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: We have very strong roadmap for GCN architecture

Gregster

Gregster

Gregster

Caporegime
Joined
24 Sep 2008
Posts
38,284
Location
Essex innit!
Despite of ongoing problems and lowering research and development budget, Advanced Micro Devices maintains optimism and claims that it will keep financing development of key products.
In the recent months, Advanced Micro Devices has been criticized heavily for re-introducing its previous-generation graphics processing units under new names. The release of one new GPU per year is clearly a result of multiple business decisions, massive layoffs from the company several years ago as well as limited research and development budget. Nonetheless, AMD seems to remain rather optimistic and claims that it has strong APU, CPU and GPU roadmaps. What is particularly interesting is that the company intends to continue improving its GCN [graphics core next] architecture, not introduce something brand new.
Click to expand...

http://www.kitguru.net/components/g...ave-very-strong-roadmap-for-gcn-architecture/

Good that AMD are sticking with GCN or a sign that they just don't have the funds to spend on R&D?
 
Ooops!

In this slide, I miss read, FX CPUs as FIX CPU's. :p

amd_graphics_leadership_1.png
 
layte said:
No point in tossing the baby out with the bathwater. Performance from GCN is competitive, but they do need to get a handle on power efficiency, before that runs away from them.
Click to expand...

Yer agreed. As I was doing this thread, I imagined what Maxwell would have been like if it was on 14nm and had the great power efficiency that it has and it just made me look forward to Volta. I am sure AMD will address power and refine it as well....
 
As consumers, we so need AMD to stay in business and be competitive both on the GPU and CPU front.

I still remember the bad old days of Intel monopolization, and it wasn't pretty.
 
Hopefully as part of their roadmap they will spend a bit of time on product availability to market. There is no point in having a decent product (Fury) if no-one can buy the damn thing!
 
Fiji isn't bad for power consumption and a significant improvement on Hawaii, @ 4K its a solid 30% faster while using the same power, actually a bit less according to TPU, so real a improvement there and a 175 Watt Fury-Nano would have about the same efficiency as Maxwell.

GCN is a good architecture and solid for the modern age, if they can evolve it (Which they already are) i don't see a problem.
 
Are they consolidating CPU and APU socket sets? That's good, makes sense.

Speculating about the current state of the GPU lineup, It seems to me that both companies expected to be on the smaller processes for this gen and AMD was hit harder with it's depleted R&D resources, hoping they are back on track in the coming year.

They really should be in an advantageous position with priority HBM supply, a years lead in retail production and being a development partner in the first place.

I'm very interested in what's going to happen in the laptop/tablet side of the market.
 
There's not much point improving GCN if they aren't going to implement it in GPU's, GCN1.2 first appeared a year ago and still only Tonga/Fiji have it. The majority of their 'newest' 3xx series are GCN1.1 or older (GCN1.1 dates back 2 years, longer if GCN1.0).

They definitely need to do some work on tessellation going forward, they can't rely on driver cheats forever.
 
Last edited:
Gregster said:
http://www.kitguru.net/components/g...ave-very-strong-roadmap-for-gcn-architecture/

Good that AMD are sticking with GCN or a sign that they just don't have the funds to spend on R&D?
Click to expand...

I'm sure an earlier AMD message hinted that Fiji is end the line for GCN but it my have been end of the line for GCN1.2, and there will be a new GCN version 2.0.
Something about 4 shader engines being the limit for GCN, 1 geometry engine per shader engine etc.


Fingers crossed that AMD do have soemthign up their sleeve. The worry is a cut in R&D has a delayed onset in showing itself, and has a long delay in recovery (you cant suddenly throw 500million at R&D and get results in 6 months).
 
humbug said:
Fiji isn't bad for power consumption and a significant improvement on Hawaii, @ 4K its a solid 30% faster while using the same power, actually a bit less according to TPU, so real a improvement there and a 175 Watt Fury-Nano would have about the same efficiency as Maxwell.

GCN is a good architecture and solid for the modern age, if they can evolve it (Which they already are) i don't see a problem.
Click to expand...

But it gets the power efficiency form using HBM, and the FuryX being water cooled.

The HBM supposedly saves 30-50watts according to AMD PR.


AMD and Nvidia both increased performance per watt by a factor of 1.5X, but NVidia did that entirely within the GPU while AMD did that with the help of HBM. That is fine and irrelevant form current consuemr eprspective, but going to the next gen Nvidia will gain the 40 odd watts from HBM while AMD will have to focus on the GPU efficiency, nothing they wont be capable of achieving but Nividia's job is certainly much easier. They will both gain from the node shrink at least.
 
D.P. said:
But it gets the power efficiency form using HBM, and the FuryX being water cooled.

The HBM supposedly saves 30-50watts according to AMD PR.


AMD and Nvidia both increased performance per watt by a factor of 1.5X, but NVidia did that entirely within the GPU while AMD did that with the help of HBM. That is fine and irrelevant form current consuemr eprspective, but going to the next gen Nvidia will gain the 40 odd watts from HBM while AMD will have to focus on the GPU efficiency, nothing they wont be capable of achieving but Nividia's job is certainly much easier. They will both gain from the node shrink at least.
Click to expand...

Maxwell will gain less from HBM as they have a much smaller IMC and use 8 Memory IC's on GM204 as opposed to 16 on Hawaii, often Maxwell also runs the very efficient Samsung IC's while AMD tend to have the very inefficient Elpida IC's, tho i think the 390/X now use Hynix more than Elpida.

I agree AMD need to work on GPU efficiency but i think they will, we may see an improvement once we see Fury-Nano.
 
Hopefully Arctic Islands will put up more of a fight than Fiji. I really wanted to buy a Fury X but I just can't justify it. Luckily I can wait until next year so it's not too bad for me.
 
I think i read somewhere that with Greenland and 14NM, they will be releasing a full line up of new cards across all segments.
 
mmj_uk said:
They definitely need to do some work on tessellation going forward, they can't rely on driver cheats forever.
Click to expand...

:rolleyes:

Still spouting that crap. How is a user defined option a cheat? I've asked you this numerous times but it seems all you do is come in to spout your "driver cheat" bs then run out of the thread never to be seen again. If it was something in the driver that users had no control over then you might have a point, but its something the user can totally disable.

Its not as if its going to gain them some wins in reviews as reviewers are well aware of it, so if the user wants better performance they can enable it. Who are they cheating exactly? Oh right, nobody. :rolleyes:
 
I hope 16nm GPU's come earlier in 2016 rather than later, there has been no real performance gain from either side around the £300 mark in more than 2 years, i'm sick to the back teeth of 28nm, i have had 4 of them.
 
You must log in or register to reply here.
Back
Top Bottom