The Ryzen "ThreadRipper"... 16 cores of awesome

[Lodix]

AMD has stated publically that Zen+ will have at least a 10% IPC gain thanks to that 'low hanging fruit' you speak of. I've heard rumors that internally they are expecting closer to 20%. Furthermore, the node change to 14nm+ is going to allow them to have much higher clock speeds. Intel had better already be working on something to counter. Otherwise, when next year rolls around, we are in for a repeat of the Athlon 64.

Source for that AMD statement of Zen+ having at least 10% ipc increase ? Just curious.

 

[krumme]

I'm actually quite surprised there turned out to be no Threadripper SKU that would have been 2x Ryzen 1700 "glued together". I mean you would have got a 3.0 Base to about 3.7 Boost 16 core processor with a 130W TDP. That would certainly be air-coolable. Definitely not for everybody, but I could certainly see some uses for it

As for The Zen+ I'm holding my finger crossed for it to be the same that Piledriver was to Bulldozer. E.g. up to 200-300 MHz frequency improvements (no 4 GHz wall anymore) and between 5-10% IPC increase, with a bit more in a couple corner cases. The IPC part might be a bit optimistic, considering that even though Zen is the first implementation of a new architecture, it has way less low-hanging fruit than bulldozer had, but I'm still hopeful. It certainly wouldn't hurt against the coming 6-core Coffee-lake.

A noctua d15 can easily cool a tr 180w tdp. With capacity left for oc.
I think skl x have messed up understanding a bit here because the bad efficiency and high power usage clouds the tim is actually a bad solution even at stock for 7900x.
A soldered tr will not have that airgap to fight.

 

[Tee9000]

AMD has stated publically that Zen+ will have at least a 10% IPC gain thanks to that 'low hanging fruit' you speak of. I've heard rumors that internally they are expecting closer to 20%. Furthermore, the node change to 14nm+ is going to allow them to have much higher clock speeds.

Zen+ is called Zen 2 now, built on GF's 7nm.

But I think you're making a confusion and you're actually talking about the 14nm+ Zen (Pinnacle Ridge, a refresh of Summit Ridge, similar to Godavari). I don't recall AMD ever publicly stating that Pinnacle Ridge or Zen 2 will have 10% IPC increase. Can we have a source for that please?

 

[beginner99]

A noctua d15 can easily cool a tr 180w tdp. With capacity left for oc.

Theoretically going by TDP. Same applies for Skylake-X. But my main point to reply is that no, the D15 won't cool a TR because it's not compatible with the huge IHS and socket of TR.

 

[Shivansps]

I played the original Crysis back in the day barely breaking 25 fps. So I can't imagine why you'd base a purchase decision on whether a CPU can run an ancient game at 5 fps or 7 fps. I mean, you'd have a point if you were trying to push hundreds of FPS in CS:GO, but sub-10 fps is unplayable, regardless of the CPU.

Because im still playing them because they are good games? Whats wrong with wanting to have some improvement over my ancient hardware with all the money im paying?

Also Stellaris came out in 2016 thats no ancient, just bad programing.

Anyway, i just brought a 1700.

 

[moinmoin]

Anyway, i just brought a 1700.

With your focus on ST I was expecting you to go 1800X or at least 1600X.

 

[Shivansps]

With your focus on ST I was expecting you to go 1800X or at least 1600X.

1800X is just too expensive, if i was willing to pay that whould have waited for CFL-S instead.

And the 1600X plus the cooler, that was more than the 1700 alone.
Im going to go with the 1700 and wait to see if Zen2 is better in clock/ipc.

 

[LTC8K6]

A noctua d15 can easily cool a tr 180w tdp. With capacity left for oc.
I think skl x have messed up understanding a bit here because the bad efficiency and high power usage clouds the tim is actually a bad solution even at stock for 7900x.
A soldered tr will not have that airgap to fight.

http://noctua.at/en/tdp-guide

 

[ajc9988]

TR has its own sinks. https://www.techpowerup.com/233969/...nk-protoypes-for-threadripper-tr4-sp3-sockets

 

[french toast]

Just fyi, GloFo has estimated 40% transistor improvement over current 14nm process. That isn't IPC, but worth noting. Where the 20% came from is early ads for zen+. It disappeared once the actual process info came out. So, somewhere between Zen+ and Zen 2, we should see IPC improvement.

I agree, but it's the summit ridge to pinnacle ridge IPC improvement I'm interested in, 5℅ with 10℅ frequency is what I'm going with, I don't have confidence in 14nm lpp though, unless they use 14nm LPU, which is what over been speculating since last year.

 

[ajc9988]

I agree, but it's the summit ridge to pinnacle ridge IPC improvement I'm interested in, 5℅ with 10℅ frequency is what I'm going with, I don't have confidence in 14nm lpp though, unless they use 14nm LPU, which is what over been speculating since last year.

I plan on TR, so 14nm+ doesn't interest me in the slightest.

 

[maddie]

Because im still playing them because they are good games? Whats wrong with wanting to have some improvement over my ancient hardware with all the money im paying?

Also Stellaris came out in 2016 thats no ancient, just bad programing.

Anyway, i just brought a 1700.

My response to your earlier post is that if you're getting 5 fps with an overclocked 2500k, then you'll never find a CPU with enough single thread power, so why even bring such an argument to the discussion.

You're wishing for a fantasy world.

 

[raghu78]

I agree, but it's the summit ridge to pinnacle ridge IPC improvement I'm interested in, 5℅ with 10℅ frequency is what I'm going with, I don't have confidence in 14nm lpp though, unless they use 14nm LPU, which is what over been speculating since last year.

Summit Ridge to Pinnacle Ridge is most likely the same Zen core on a higher performance process with a physical design optimized for higher frequencies. AMD calls it Zen on 14nm+

http://hexus.net/tech/news/cpu/105772-amd-unveils-roadmap-ryzen-threadripper-ryzen-3-apus/

The process node is likely GF 14 HPP .

 

[Topweasel]

Summit Ridge to Pinnacle Ridge is most likely the same Zen core on a higher performance process with a physical design optimized for higher frequencies. AMD calls it Zen on 14nm+

http://hexus.net/tech/news/cpu/105772-amd-unveils-roadmap-ryzen-threadripper-ryzen-3-apus/

The process node is likely GF 14 HPP .

I thought it was LP for Lead Process. Where we are on LPP for Low Power Process.

 

[raghu78]

I thought it was LP for Lead Process. Where we are on LPP for Low Power Process.

Leading Performance (LP) is for GF 7LP .

https://www.globalfoundries.com/new...ver-leading-performance-7nm-finfet-technology

For 14nm GF has 14LPP, 14HPP and 14HP (Silicon on Insulator) . 14 HP is a IBM node.

http://www.eetimes.com/document.asp?doc_id=1324343

High Performance 14nm SOI FinFET CMOS Technology with 0.0174µm2 Embedded DRAM and 15 Levels of Cu Metallization.

 

[Topweasel]

Leading Performance (LP) is for GF 7LP .

https://www.globalfoundries.com/new...ver-leading-performance-7nm-finfet-technology

For 14nm GF has 14LPP, 14HPP and 14HP (Silicon on Insulator) . 14 HP is a IBM node.

http://www.eetimes.com/document.asp?doc_id=1324343

High Performance 14nm SOI FinFET CMOS Technology with 0.0174µm2 Embedded DRAM and 15 Levels of Cu Metallization.

Okay yeah I must have been confusing the 7LP with the move to 14nm+.

 

[french toast]

Leading Performance (LP) is for GF 7LP .

https://www.globalfoundries.com/new...ver-leading-performance-7nm-finfet-technology

For 14nm GF has 14LPP, 14HPP and 14HP (Silicon on Insulator) . 14 HP is a IBM node.

http://www.eetimes.com/document.asp?doc_id=1324343

High Performance 14nm SOI FinFET CMOS Technology with 0.0174µm2 Embedded DRAM and 15 Levels of Cu Metallization.

cool, do you have any info on the improvement of 14nm HP to 14nm LPP?

 

[formulav8]

I'm still wondering if Z+ is going to be 14nm LPU.

 

[french toast]

I'm still wondering if Z+ is going to be 14nm LPU.

Thats what I've been saying, if ryzen is already 14nm lpp, then an easier upgrade would be 14nm LPU I would think as it's still a Samsung variant, but glofo already has 14nm HP in use with IBM so who knows?

Which one is better? 14nm LPU vs 14nm HP?

 

[ub4ty]

Can you guys discussing the manufacturing processes give the lay version to some of those viewing this thread?
So, we have : 14nm, 14nm+, 7nm, 7nm+

Were on 14nm heading towards 14nm+.
When do you expect the manufacturing process to go to 14nm+?, 7nm?, 7nm+
What is the fundamental difference between 14nm and 14nm+? What can be expected from it w.r.t to performance/power draw?
Given the obvious big difference between 7nm and 14nm, this is seemingly the bigger jump to look forward to. That seems somewhat down the road. Thus, as it has already been detailed, the 14nm results look good. Buy 14nm now and wait to 7nm or 7nm+ seems like a solid strategy? 14nm+ still is going to be nowhere near 7nm and meanwhile you're skipping out on having it (now) to get 14nm+ which will be seriously overshadowed by 7nm.

So : 14nm buy -> 14nm sell -> 7nm+ (~2 years from now?)
Thoughts?

 

[moinmoin]

Can you guys discussing the manufacturing processes give the lay version to some of those viewing this thread?
So, we have : 14nm, 14nm+, 7nm, 7nm+

Were on 14nm heading towards 14nm+.
When do you expect the manufacturing process to go to 14nm+?, 7nm?, 7nm+
What is the fundamental difference between 14nm and 14nm+? What can be expected from it w.r.t to performance/power draw?
Given the obvious big difference between 7nm and 14nm, this is seemingly the bigger jump to look forward to. That seems somewhat down the road. Thus, as it has already been detailed, the 14nm results look good. Buy 14nm now and wait to 7nm or 7nm+ seems like a solid strategy? 14nm+ still is going to be nowhere near 7nm and meanwhile you're skipping out on having it (now) to get 14nm+ which will be seriously overshadowed by 7nm.

So : 14nm buy -> 14nm sell -> 7nm+ (~2 years from now?)
Thoughts?

This is the roadmap:

We know that tape out for Zen 2 is planned for the end of this year, with production starting end of next year and high-volume production in early 2018 respectively. Details on Zen+/14nm+ are scarce in comparison.

 

[french toast]

We have different forks of 14nm process technology developed by different companies.
Samsung fork which is used by global foundries and by extension AMD- is the 14nm LPP, the series of that process timeline in strict samsung nomenclature is 14nm LPE - 14nm Lpp- 14nm Lpc?- 14nm LPU. (1st to 4th generation)
To confuse things AMD uses second gen 14nm LPP for its processors, but upgraded it to "3rd gen" for Rx 580, no one knows if that's their own revision of 14nm lpp or 3rd gen in strict samsung parlance (14nm Lpc? )

To confuse things even more glofo uses IBM designed 14nm HP SOI finfet for IBM server processors, this is a separate and completely different process to Samsung derived 14nm LPP, and is more suited to higher clocked/ higher power asics, more in line ( it still different) to Samsung's 4th gen 14nm LPU.

So when AMD uses the term 14nm+ on its roadmaps it could conceivably mean one of 5 things, 1) 14nm LPU 2)14nm LPC? 3) IBM 14nm HP. 4) AMD own revision of 14nm LPP. 5) AMD own revision of 14nm LPC.

As for performance I think 10-15℅ speed improvement with 14nmLPU and perhaps IBM14nm HP, less for 14nm LPC and AMD own revision of 14nm LPP/LPC
7nm IBM/glofo should be 25-30℅ faster than that in H2 2018/2019.

Just my thoughts, not sure about correct naming of 3rd gen 14nm LPC, just from memory.
Edited.
https://www.extremetech.com/extreme...ans-10nm-improvements-shows-off-7nm-euv-wafer

 

[ub4ty]

We have different forks of 14nm process technology developed by different companies.
Samsung fork which is used by global foundries and by extension AMD- is the 14nm LPP, the series of that process timeline in strict samsung nomenclature is 14nm LPE - 14nm Lpp- 14nm Lpc?- 14nm LPU. (1st to 4th generation)
To confuse things AMD uses second gen 14nm LPP for its processors, but upgraded it to "3rd gen" for Rx 580, no one knows if that's their own revision of 14nm lpp or 3rd gen in strict samsung parlance (14nm Lpc? )

To confuse things even more glofo uses IBM designed 14nm HP SOI finfet for IBM server processors, this is a separate and completely different process to Samsung derived 14nm LPP, and is more suited to higher clocked/ higher power asics, more in line ( it still different) to Samsung's 4th gen 14nm LPU.

So when AMD uses the term 14nm+ on its roadmaps it could conceivably mean one of 5 things, 1) 14nm LPU 2)14nm LPC? 3) IBM 14nm HP. 4) AMD own revision of 14nm LPP. 5) AMD own revision of 14nm LPC.

As for performance I think 10-15℅ speed improvement with 14nmLPU and perhaps IBM14nm HP, less for 14nm LPC and AMD own revision of 14nm LPP/LPC
7nm IBM/glofo should be 25-30℅ faster than that in H2 2018/2019.

Just my thoughts, not sure about correct naming of 3rd gen 14nm LPC, just from memory.
Edited.
https://www.extremetech.com/extreme...ans-10nm-improvements-shows-off-7nm-euv-wafer

Thank you for this information. This is the kind of detail I was looking for.
So, 7nm is likely coming This time or later next year.
Given the bigger step coming with 7nm in approx. a year, it seems it will just be purchase timing, if one intends to buy now, as to whether or not they get 14nm/14nm+. I don't see a big reason to hold out for 14nm+. I do so a slightly pronounced one for 7nm. However, if pressed, one can just get 14nm now and wait 2+ years to sell it and go towards 7nm+

Your extended detailed helped affirm some of my thinking. So, thank you very much.

 

[DrMrLordX]

A noctua d15 can easily cool a tr 180w tdp.

I've forced mine to cool a 250W socket load. Yowza! Granted it has really loud, really powerful fans on it.

I would like to see some kind of Zen CPU on GF 14nm HP, but I doubt that IBM will allow that by contract.

 

[french toast]

Thank you for this information. This is the kind of detail I was looking for.
So, 7nm is likely coming This time or later next year.
Given the bigger step coming with 7nm in approx. a year, it seems it will just be purchase timing, if one intends to buy now, as to whether or not they get 14nm/14nm+. I don't see a big reason to hold out for 14nm+. I do so a slightly pronounced one for 7nm. However, if pressed, one can just get 14nm now and wait 2+ years to sell it and go towards 7nm+

Your extended detailed helped affirm some of my thinking. So, thank you very much.

No problem. I think 7nm is risk production in H1 2018 and volume production in H2 2018, but in all likelihood we would probably be seeing some kind of pipe cleaner from AMD in late 2018 for 7nm, with the real serious stuff pumped out in H1 2019.
No-one knows for sure though.