AMD Vega Discussion Thread

[xkm1948]

I'd pass then. Those blower coolers have never worked well. AIB cards usually turn out much better.

I'd pass the entire Vega line up. It feels like GCN-2017 variant to me. GCN was good, at the same time GCN may be too old. RTG need some good fresh new design to go with the new 14nm nodes.

 

[FordGT90Concept]

Pretty sure Navi is GCN 6 too so that's gonna be a while.

GCN still is a compute biased GPU.

Considering the clockspeed, GCN5 is a huge change from GCN4. A chip that's almost twice as big as Polaris running at 300 MHz faster than Polaris.

I think what it really comes down to is that Pascal was a better die shrink of Maxwell than Polaris was a better die shrink of Hawaii. Vega may have the architecture right but not the process.

 

[Vya Domus]

AMD may never get the process right as long as they are tied to GlobalFoundries.

 

[ratirt]

AMD may never get the process right as long as they are tied to GlobalFoundries.

Why do you think that? They are still in the market and they are trying. They have the technology maybe they need a bit tweaking and improvements but come on give them some slack bro. They try the Vega card who knows what will happen afterwards.

 

[Vya Domus]

Why do you think that? They are still in the market and they are trying. They have the technology maybe they need a bit tweaking and improvements but come on give them some slack bro. They try the Vega card who knows what will happen afterwards.

AMD is leaking money to GloFo whether or not they use their process. It's holding them back both financially and technologically.

 

[uuuaaaaaa]

I wonder if this is in any way related to Vega:

Terry Makedon‏ @CatalystMaker
I sense a BIG day for Radeon Software tomorrow. Cant wait to let the cat out of the bag......

https://twitter.com/CatalystMaker/status/889946633575780352

 

[ratirt]

AMD is leaking money to GloFlo whether or not they use their process. It's holding them back both financially and technologically.

GloFo is now independent of AMD. They have a relation but not like that. Not sure from where you have taken this. I really don't think AMD is holding GloFo back in any case. Actually i'd go the other way. Both GloFo and AMD would like to get more products and with better quality.

 

[Vya Domus]

GloFo is now independent of AMD. They have a relation but not like that. Not sure from where you have taken this. I really don't think AMD is holding GloFo back in any case. Actually i'd go the other way. Both GloFo and AMD would like to get more products and with better quality.

They are independent now but AMD is still tied to them through a deal , look it up. GloFo is not motivated to improve themselves so much because they still get cash from AMD no matter how bad they are doing. Don't get me wrong they aren't horrible , AMD would have went out of business if that was the case , but they are no where near brilliant. TSMC is way ahead of them.

 

[ratirt]

They are independent now but AMD is still tied to them through a deal , look it up. GloFo is not motivated to improve themselves so much because they still get cash from AMD no matter how bad they are doing. Don't get me wrong they aren't horrible , AMD would have went out of business if that was the case , but they are no where near brilliant. TSMC is way ahead of them.

Yeah I know they are tied to the agreement. But what I understand is that GloFo is being dragged down by AMD and contract they have signed?
Well I know what you are saying but there's the other side here as well. The AMD is giving them an opportunity with stable cash flow. They will need to get more advanced. In my opinion GloFo and AMD are close and they will try to be better together. AMD can always demand more of their process I think it is good for progress. TSMC may be more advanced but that doesn't mean GloFo's process is bad or wrong. Maybe they need some time to progress from what they got already and where they want to be. 7nm is on the way and there's plans to improve 7nm process with additional changes that may help.

BTW: TSMC is not that far far beyond GloFo process. ALso it's hard to compare since those are different processes. TSMC is better maybe but it has had more time to mature quietly when AMD has had problems? Maybe now the tables have been turned and they will get better each year. The announcement for 7nm is there. That's a good sign.

 

[Vya Domus]

But what I understand is that GloFo is being dragged down by AMD and contract they have signed?

They wish. It's the other way around no doubt. Not sure why it is so difficult for you to see , this has always been the case. If you have a deal with a manufacturer that you are going to buy from them a guaranteed amount of goods and not only that but if you want to get goods from someone else you still have to pay them a certain amount of money , I ask you this : Who is dragging down who ?

Also I read some time ago that GloFo has major problems with 7nm , to put it simply their 7nm isn't really 7nm.

BTW: TSMC is not that far far beyond GloFo process.

It doesn't matter how they are doing it. They are ahead , that's all that matters.

 

[ratirt]

They wish. It's the other way around no doubt. Not sure why it is so difficult for you to see , this has always been the case. If you have a deal with a manufacturer that you are going to buy from them a guaranteed amount of goods and not only that but if you want to get goods from someone else you still have to pay them a certain amount of money , I ask you this : Who is dragging down who ?

Also I read some time ago that GloFo has major problems with 7nm , to put it simply their 7nm isn't really 7nm.



It doesn't matter how they are doing it. They are ahead , that's all that matters.

Well i'm a glass kinda full guy. What can I do. TSMC is better but this doesn't mean that GloFo can't do better. it sounds like the agreement they have made drags one of the companies down. Probably you meant GloFo that's being dragged down. Well I don't see that. For me it is a stable agreement which keeps the cash flow which is all that matters. Not sure if other companies ( and there's some) that uses GloFo for manufacturing would state what you have done in your previous posts.

 

[Vya Domus]

Probably you meant GloFo that's being dragged down.

No , I keep making myself clear that AMD is being dragged down by GloFo and that's not subjective at all , I can't understand why you keep thinking I am saying the opposite.

It's a stable agreement that keeps the cash flowing just toward GloFo not AMD. AMD should have never had to sell their fabs and moreover they should have never had to settle down for such a deal. They failed and they did that and it's biting them every step along the way. I don't blame them it was tough to maintain their own fabs , but it's time to make the best out of their situation. Either make GloFo very competitive or move to someone else.

 

[the54thvoid]

I wonder if this is in any way related to Vega:

Terry Makedon‏ @CatalystMaker
I sense a BIG day for Radeon Software tomorrow. Cant wait to let the cat out of the bag......

https://twitter.com/CatalystMaker/status/889946633575780352

File it under hype. Probably the catalyst driver for Vega.

 

[cdawall]

File it under hype. Probably the catalyst driver for Vega.

Maybe this driver half works instead of doesn't work at all?

 

[Vayra86]

AMD may never get the process right as long as they are tied to GlobalFoundries.

Why not? Ryzen is proof that the process is fine; it does perf/watt miles better than even Intel's 14nm in a number of instances, and better than GCN does comparatively to Pascal, and it clocks reliably too.

Improved process - remember the clock bump on RX580 and how it was directly impacting the board TDP? Its still 14nm and that's all she wrote. Meanwhile, we *know* that GCN is a highly inefficient architecture by now and the only way Polaris hides this is because it was the first 14nm GPU at the time.

I don't buy that a smaller node will finally save AMD GPUs from its power consumption woes - and it looks an awful lot like the eternal AMD escape strategy that relies on 'the future' where something will finally do well. AMD needs to do what they've been postponing for too long and that is a completely overhaul of GCN or even starting from the ground up again - and so far what they've tried with it has been only partially successful. That, along with the extremely long time before Vega release, does not point towards brilliant engineers at RTG, but rather a lack of talent and stagnation. Yes, I know, the limitations of R&D budget, but this is what it is.

 

[cdawall]

Why not? Ryzen is proof that the process is fine; it does perf/watt miles better than even Intel's 14nm in a number of instances, and better than GCN does comparatively to Pascal, and it clocks reliably too.

Two completely different designs and performance per watt isn't exactly great the second you want over 4ghz

 

[Vayra86]

Two completely different designs and performance per watt isn't exactly great the second you want over 4ghz

Obviously its a different world. But it still is on the same node, one that is supposedly not power efficient enough because hey, GCN can't keep up with Pascal numbers. Not having the highest clock + high power consumption are closely related, and like you say, are the result of the design used. Not the node.

 

[Vya Domus]

Why not? Ryzen is proof that the process is fine; it does perf/watt miles better than even Intel's 14nm in a number of instances, and better than GCN does comparatively to Pascal, and it clocks reliably too.

Improved process - remember the clock bump on RX580 and how it was directly impacting the board TDP? Its still 14nm and that's all she wrote. Meanwhile, we *know* that GCN is a highly inefficient architecture by now and the only way Polaris hides this is because it was the first 14nm GPU at the time.

I don't buy that a smaller node will finally save AMD GPUs from its power consumption woes - and it looks an awful lot like the eternal AMD escape strategy that relies on 'the future' where something will finally do well. AMD needs to do what they've been postponing for too long and that is a completely overhaul of GCN or even starting from the ground up again - and so far what they've tried with it has been only partially successful. That, along with the extremely long time before Vega release, does not point towards brilliant engineers at RTG, but rather a lack of talent and stagnation. Yes, I know, the limitations of R&D budget, but this is what it is.

But Intel can clock their CPUs higher at the end of the day. Considering the IPC is similar and there are no huge differences in terms of what is present on the silicon the only differences stem from the manufacturing node , which is better on the Intel side. Didn't say it was bad , just not the best. You got to realize that the reason why it is so power efficient it's because those chips simply do not push for a lot of clockspeed.

CGN is a brilliant architecture all-around , the engineers are brilliant too. It is not the best for gaming though which is what bothers most of you , or rather it is not easy to get a lot of performance out of it in terms of gaming. AMD has the cash to make a great gaming card or a great compute card , not both at the same time.

Just to give you an example : CGN has twice as wide SIMD instructions compared to Nvidia , this means it can be more efficient in compute but less stellar in gaming since underutilization gets penalized twice as much.

 

[cdawall]

Obviously its a different world. But it still is on the same node, one that is supposedly not power efficient enough because hey, GCN can't keep up with Pascal numbers. Not having the highest clock + high power consumption are closely related, and like you say, are the result of the design used. Not the node.

The node itself is behind Intel, samsung and tsmc they are so far behind they purchased their next node from IBM.

The node might be capable of lower wattage consumption parts, but yields are terrible and those are heavily limited to who the good gpus go to. Want a 95w board tdp rx480? Sure drop 4k and you can have one. I imagine threadripper vs Ryzen is the same way. If I was amd I sure would be dumping the junk chips to the mainstream.

Basically yields are garbage, having to buy your node is garbage and glofo is garbage.

 

[Vya Domus]

Basically yields are garbage, having to buy your node is garbage and glofo is garbage.

Yields are dictated by die size mostly, not the node itself. Taking into account the low clockspeeds Zen was designed for and I am pretty sure AMD can pump out a ton of CPUs. I wouldn't call their yields garbage. Intel with their astronomically large dies however...

 

[cdawall]

Yields are dictated by die size mostly, not the node itself. Taking into account the low clockspeeds Zen was designed for and I am pretty sure AMD can pump out a ton of CPUs. I wouldn't call their yields garbage. Intel with their astronomically large dies however...

Yet Intel is making money hand over foot.

 

[Vya Domus]

Yet Intel is making money hand over foot.

And spending a lot in the process as well.

There is a very good reson behinde the "CPUs glued together" claim. It pisses them off that AMD has a competing product on a much inferior node and cheaper.

 

[cdawall]

And spending a lot in the process as well.

There is a very good reson behinde the "CPUs glued together" claim. It pisses them off that AMD has a competing product on a much inferior node and cheaper.

You sure it wasn't a comment back to when AMD made the same exact claim against Intel? You know Intel was competing on what was openly referred to as an inferior node as well then.

 

[P4-630]

"amd vega discussion thread"

I'm reading comments about AMD and Intel CPU's now lol!!

 

[Vya Domus]

"amd vega discussion thread"

I'm reading comments about AMD and Intel CPU's now lol!!

It's related more to the quality of the manufacturing process which is relevant to Vega.

You sure it wasn't a comment back to when AMD made the same exact claim against Intel? You know Intel was competing on what was openly referred to as an inferior node as well then.

Exactly and AMD was pissed then as well for investing in something that wouldn't pay off for the most part.