AMD's Mark Papermaster Discusses 7nm Process Technology

In case you missed the original presentation, AMD just uploaded the part of their Next Horizon event where they discussed the upcoming use of 7nm process technology. AMD has traditionally stuck with their former internal fab, GlobalFoundries, for most chip designs, and they're reportedly still relying on GloFo for certain chips, like 2nd gen Epyc's 14nm I/O die. But for the most part, AMD is using Taiwan Semiconductor's bleeding edge 7nm process in their major upcoming products, like Ryzen CPUs, the recently released Radeon VII, and the long rumored Navi GPUs.

Check out AMD's talk on 7nm here.

AMD notes that they expected to achieve process parity with Intel this generation, but thanks to Intel's 10nm woes and a conspicuous lack of problems at TSMC's 7nm node, they've "unexpectedly" ended up ahead of their x86 competitors.

And now AMD, do not take the foot off that throttle - mash it down. This is an opportunity you've not had in a very long time.

PhaseNoise said:

And now AMD, do not take the foot off that throttle - mash it down. This is an opportunity you've not had in a very long time.

Click to expand...

Honestly whether you are team red or team blue, I hope this gen they beat Intel and mash their faces into the ground, Intel and AMD seem to advance CPU design when they are in a heated competitive environment, It was a shame when AMD fell behind as consumers we all paid the price, I have both Intel and AMD now as well as before in yesteryears I want both companies to succeed and thrive as the industry progresses and is better for the consumers when they are both at their best. As long as AMD can hit that 5ghz IPC parity with multiple cores we will all be good.

What AMD doesn't tell you is that the gates the data must past thru is larger than nVidia's / Intel's .. someone can of course correct me here.

1Nocturnal101 said:

Honestly whether you are team red or team blue, I hope this gen they beat Intel and mash their faces into the ground, Intel and AMD seem to advance CPU design when they are in a heated competitive environment, It was a shame when AMD fell behind as consumers we all paid the price, I have both Intel and AMD now as well as before in yesteryears I want both companies to succeed and thrive as the industry progresses and is better for the consumers when they are both at their best. As long as AMD can hit that 5ghz IPC parity with multiple cores we will all be good.

Click to expand...

Well, i think Intel views x86 as legacy, and they will slow walk it as much as possible as to keep margins as high as possible. For AMD everything is growth/ growth based strategy since they are relatively puny. Intel will slow walk and milk x86 to the end of time if allowed, meanwhile they will keep looking for the next growth opportunities. To be honest it makes sense for Intel to do this specially if competition is weakened. Its sucky, Intels has not been a good stalwart of x86... unless competition. They certainly could have been doing better, even if on their own.

SixFootDuo said:

What AMD doesn't tell you is that the gates the data must past thru is larger than nVidia's / Intel's .. someone can of course correct me here.

Click to expand...

Meaning, the I/O chip uses 14nm? That was covered, above. I like AdoredTV's explanation of why using that I/O and the 7nm chiplet design for the processor makes economic sense...

Would a 7nm I/O chip give better performance? I don't know, but if 7nm I/O is better, I'm sure a chiplet design would enable one to be installed. Ryzen 3?

The big question is whether 7nm I/O would make a faster cpu than a 14nm I/O on a chiplet design. And, by how much, and at what cost.

Am hoping AMD makes a meaningful improvement on clock speed with zen 2. Many users do not have a current use for more than 8 core/16 thread.
Give me 6-8 cores at 4.6Ghz plus and I will have a reason to upgrade my 2600X gaming rig

7nm Process didn't seem to do that much for Radeon 7, though AMD is well behind in graphics core tech.

It should be quite nice on CPUs where AMD is already close to parity except on clock speed.

Uvaman2 said:

Well, i think Intel views x86 as legacy, and they will slow walk it as much as possible as to keep margins as high as possible. For AMD everything is growth/ growth based strategy since they are relatively puny. Intel will slow walk and milk x86 to the end of time if allowed, meanwhile they will keep looking for the next growth opportunities. To be honest it makes sense for Intel to do this specially if competition is weakened. Its sucky, Intels has not been a good stalwart of x86... unless competition. They certainly could have been doing better, even if on their own.

Click to expand...

Honestly, x86 and x64 are probably all we need a x128 CPU structure isn't needed as they are very little types of programs(like calculating Pi and such, although GPUs might benefit significantly more, but now you are way beyond out of league for me, you would have to ask Ian Cutress, Kyle Bennet, Jim(adoredtv), maybe Gordon Ung or one of those guys that go farther down the hardware arch and programming rabbit hole and can explain it barney style for us normal people) that would benefit from the 128bit address, Unless Intel can nail something like 3dXpoint to the point we have non-volatile RAM as fast as Volatile RAM I really can't see any reason for moving beyond the x86/x64 arches as there is little benefit.

I know the community would probably throw a bitch fit for me saying this, but it would be interesting to see x86 go away completely with no legacy support for more future modern computers, yes I do understand the implications of doing it, x86 is legacy, but only because to support older programs and software.

sleepybp said:

Am hoping AMD makes a meaningful improvement on clock speed with zen 2. Many users do not have a current use for more than 8 core/16 thread.
Give me 6-8 cores at 4.6Ghz plus and I will have a reason to upgrade my 2600X gaming rig

Click to expand...

I think you might be disappointed, just in the fact we probably are going to get upsold to the 12 core model, even according to leaks I am not sure why as it doesn't make a lot of sense, but the 12 core is using a 12/14nm IO from Gloflo while 2 Chiplets at 7nm from TSMC are 6C/12T a piece there was a 5.0Ghz turbo boost on that model, I to want just an 8C/16T at 5.0Ghz but to be honest we just need to wait and I honestly want to see the performance metrics on the Chiplet design as we seen what "glued together" design is capable of on both ends with the Q6600 Intel and these Ryzen series, and it is impressive but how will Chiplets effect latency? at the cost of having a better TDP and Higher clocks with ipc.

Snowdog said:

7nm Process didn't seem to do that much for Radeon 7, though AMD is well behind in graphics core tech.

It should be quite nice on CPUs where AMD is already close to parity except on clock speed.

Click to expand...

Seems to me it helped Vega quite a bit, just that Vega was already reaching it's own limits and a die shrink can only do so much. I mean, ipc may not have changed much with VII (it did go up, though), however clocks went way up. That would not have been possible on 14/12nm.

Nobu said:

Seems to me it helped Vega quite a bit, just that Vega was already reaching it's own limits and a die shrink can only do so much. I mean, ipc may not have changed much with VII (it did go up, though), however clocks went way up. That would not have been possible on 14/12nm.

Click to expand...

Well yes and no about generalized a 25% increase at the same TDP, the lithography was 7nm yes but it was double/triple patterned, not EUV/UVL so it lacked precision, and without going to far into it, the pitch as well(don't ask). Vega 7 is a cut down Instinct card, it shouldn't exist, in fact the guy who wanted to see it left even before they started manufacturing it possibly because 7nm EUV/UVL isn't here fast enough, or they needed some kind of filer until Navi. Keep in mind the Chiplet is 7nm from TSMC while the I/O is GLOFLO 12/14nm, Honestly we know what a "Glued Together" die looks like Intel Q6600 and Ryzen both did excellent in their relative time, but this is new and we don't know about what the rabbit hole entails till we see it Latency could be an issue who knows, but Intel's 10nm I somehow feel is having problems due to the problems you get when you try shrinking the I/O down that introduces a different set of issues, so AMD may have the right solution here.

Either way speculation is speculation and we won't know how it is till we see it, Zen as an Arch I feel is damn good so whatever benefits we see may scale more positively for Zen over Vega(never mind the fact you are trying to correlate performance from Apples to Oranges), personally I never liked Vega, while I can admit the 2200/2400G and Intel variant are amazing the rest to me is just not compelling as a product for anything more than cheap compute cards or for those Die hard team red fanboys, otherwise they serve no market purpose.

Nobu said:

Seems to me it helped Vega quite a bit, just that Vega was already reaching it's own limits and a die shrink can only do so much. I mean, ipc may not have changed much with VII (it did go up, though), however clocks went way up. That would not have been possible on 14/12nm.

Click to expand...

The point was, that even on 7nm AMD is using more power than NVidia at 16nm. Process is not a panacea. AMD drastically needs a big architecture overhaul.