GLOBALFOUNDRIES Introduces New 12nm FinFET Technology for High-Performance Applications

[krumme]

Same Zen core for Summit and Pinnacle. The only differences will be in frequency and possibly out of the box memory support.

They could use a bd arch for my sake as long there is a "performance uplift".

 

[SarahKerrigan]

Same Zen core for Summit and Pinnacle. The only differences will be in frequency and possibly out of the box memory support.

So we're supposed to believe a slide that says Bristol Ridge has Polaris when it makes claims about Pinnacle Ridge?

 

[Arachnotronic]

So we're supposed to believe a slide that says Bristol Ridge has Polaris when it makes claims about Pinnacle Ridge?

You can believe the official AMD roadmap that I retrieved from AMD's website above it if you doubt the other one.

 

[krumme]

Yes its the ppt or beliewe the chief of technology.

 

[Arachnotronic]

Yes its the ppt or beliewe the chief of technology.

You mean the PPT presented by Mark Papermaster himself?

 

[krumme]

You mean the PPT presented by Mark Papermaster himself?

Yes that ppt with same arch and performance uplift next to each other. Ppt language.
Perhaps he forgot what tock means? Cant blame him if he looks out the window.

 

[Arachnotronic]

Yes that ppt with same arch and performance uplift next to each other. Ppt language.
Perhaps he forgot what tock means? Cant blame him if he looks out the window.

All that matters in the end is performance increase and not whether it's delivered by perf/MHz increases or by clock speed increases. That's what he was talking about.

 

[krumme]

All that matters in the end is performance increase and not whether it's delivered by perf/MHz increases or by clock speed increases. That's what he was talking about.

I dont think we disagree is near impossible to get any meaningfull ipc increase due to limited time learning from zen tapeout.
Its semantic. If eg max uops altered slightly and we have a new and better imc and it all adds up to eg 1-2% ipc increase is it then a tock?
I am pretty sure he thinks its a tock. And tock as arch change miniscule or not.
When he says tock its tock. Its also on a clearly new node not a slight rev so it makes sense to make a few alterations at the same time.

 

[krumme]

And i agree its performance that matters and the real news here is we get a performance node at last. Its incredible this 14nm lpp even worked for desktop freq.
Secondly its february. 1 month before anticipated if it goes as rumored.
All very good news imo.

 

[Arachnotronic]

I dont think we disagree is near impossible to get any meaningfull ipc increase due to limited time learning from zen tapeout.
Its semantic. If eg max uops altered slightly and we have a new and better imc and it all adds up to eg 1-2% ipc increase is it then a tock?
I am pretty sure he thinks its a tock. And tock as arch change miniscule or not.
When he says tock its tock. Its also on a clearly new node not a slight rev so it makes sense to make a few alterations at the same time.

Whatever you say.

 

[krumme]

Whatever you say.

Yeaa it wasnt me who said tock you know.

 

[CatMerc]

Same Zen core for Summit and Pinnacle. The only differences will be in frequency and possibly out of the box memory support.

It is entirely possible there are small IPC gains to be had from small tweaks that wouldn't warrant calling it a Zen 2.
I'm not saying it will happen, I'm currently in the camp of clockspeed increases and IMC, but it could still be.

 

[Abwx]

All that matters in the end is performance increase and not whether it's delivered by perf/MHz increases or by clock speed increases. That's what he was talking about.

He said that there will be an improvement of Zen uarch even before Zen 2 is released, his wording is explicit that they wont only shrink Zen, or have you some other explaination for what he implied in the quote i posted :

“We’re not going tick-tock,” he said. “Zen is going to be tock, tock, tock.”

In recent history they never released a new CPU without a uarch improvement, wether on a new node or a half node, FTR PD was released one year after BD...

 

[moinmoin]

Did we ever get that slide without the watermark?

 

[Arachnotronic]

In recent history they never released a new CPU without a uarch improvement, wether on a new node or a half node, FTR PD was released one year after BD...

Bristol Ridge.

 

[raghu78]

It is entirely possible there are small IPC gains to be had from small tweaks that wouldn't warrant calling it a Zen 2.
I'm not saying it will happen, I'm currently in the camp of clockspeed increases and IMC, but it could still be.

Well said. Summit Ridge is the first iteration of Zen and there are some easy improvements to be made which would not warrant calling it a new core. IMC tweaks to reduce latency and to improve high speed DDR4 support for speeds above DDR 4000 are few which come to mind. IMC tweaks could increase perf/Mhz even if they are small 2-3%.Lets wait and see what AMD has done.

 

[krumme]

Bristol Ridge.

Yep. Do you think zen+ comes without any uarch changes?

 

[rainy]

Bristol Ridge.

Not true - Bristol Ridge is on the same 28nm as Carrizo and he clearly has mentioned new node or half node.
https://en.wikipedia.org/wiki/List_...icroprocessors#.22Bristol_Ridge.22_.282016.29

 

[Ajay]

You can believe the official AMD roadmap that I retrieved from AMD's website above it if you doubt the other one.

The same architecture doesn't mean the same implementation, clearly. I'm real curious to see what AMD comes up with. I expect some minor improvements (and obviously bug fixes) + a frequency bump. I think 10% more performance is the floor for Ryzen on 12LP and for all our sake, hope they coax a bit more than that out. The 8700K seems like it's a bit pricey, but only because of performance gains over Ryzen.

 

[IRobot23]

The same architecture doesn't mean the same implementation, clearly. I'm real curious to see what AMD comes up with. I expect some minor improvements (and obviously bug fixes) + a frequency bump. I think 10% more performance is the floor for Ryzen on 12LP and for all our sake, hope they coax a bit more than that out. The 8700K seems like it's a bit pricey, but only because of performance gains over Ryzen.

Like you said, ryzen needs for gaming is little higher frequencies and less memory latency.
DF at 2000Mhz or higher would give ryzen some boost.

 

[DrMrLordX]

Memory latency is ultimately tied to IF latency. They either need to tweak IF clockspeed, or they need to give the end-user the ability to change it somehow. Because right now, 1/2 DRAM speed is holding back the platform.

 

[CatMerc]

Memory latency is ultimately tied to IF latency. They either need to tweak IF clockspeed, or they need to give the end-user the ability to change it somehow. Because right now, 1/2 DRAM speed is holding back the platform.

It is tied to IF, but it's not the sole reason I don't think. Considering AMD's history with IMC's, I'm betting they can make it lower latency with a revision.

Remember, any CPU needs to pipe data from the IMC, not just AMD's with IF. They all have latency induced from that.

 

[Ajay]

Like you said, ryzen needs for gaming is little higher frequencies and less memory latency.
DF at 2000Mhz or higher would give ryzen some boost.

Changing the speed of the data fabric would likely break the architecture. Improving speed and latency supported by the IMC wouldn't.

 

[IRobot23]

Changing the speed of the data fabric would likely break the architecture. Improving speed and latency supported by the IMC wouldn't.

IMC is great. DF is great, only low clocks.

 

[DrMrLordX]

Changing the speed of the data fabric would likely break the architecture.

I doubt it. R7s have already shown the ability to run IF speeds up to 2000 MHz, stable. The problem is the IMC and/or RAM ICs can't handle DDR4-4000 at the present. There may be some bizarre technical reason why they are unable to choose different ratios, but in terms of raw speed, the present IF implementation CAN handle such speeds. Plus there's no reason why they couldn't fix that technical limitation, if they so chose.