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

[moinmoin]

To be honest I have a hard time finding anything concrete on 14nm+ for Zen chips. Are we sure that it isn't just design improvements on the same process for Raven Ridge/Ryzen Mobile/Ryzen APU as those will contain an 1x CCX die unlike the 2x CCX Zeppelin die used everywhere up to now?

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.

If tape out for Zen 2 is this year (which Lisa Su stated at the AMD Financial Analyst Day Q&A), risk production will definitely be within next year, unless you expect a slower turn around with Zen 2 than happened with the original Zen (which was a whole new design to boot). And I personally expect AMD to be able to start volume production earlier than normal as even with low yield at the foundries the Zen design and product line contains plenty redundancy to repurpose "bad" dies.

 

[formulav8]

So, we have : 14nm, 14nm+

LPU could increase transistor switching speed some within still the 14nm low power process allowing AMD to increase (potentially) clocks higher than the current 14nm LPP process. But just because the transistors can handle faster switching doesn't mean the cpu design can. But if AMD uses the Samsung LPU process they should be be to do a remask/layout and benefit from the better 14nm lpu process. Thus improving performance with (and maybe very low hanging fruit IPC improvements) increased clock-speeds.

Of course I'm not an expert,

 

[french toast]

To be honest I have a hard time finding anything concrete on 14nm+ for Zen chips. Are we sure that it isn't just design improvements on the same process for Raven Ridge/Ryzen Mobile/Ryzen APU as those will contain an 1x CCX die unlike the 2x CCX Zeppelin die used everywhere up to now?


If tape out for Zen 2 is this year (which Lisa Su stated at the AMD Financial Analyst Day Q&A), risk production will definitely be within next year, unless you expect a slower turn around with Zen 2 than happened with the original Zen (which was a whole new design to boot). And I personally expect AMD to be able to start volume production earlier than normal as even with low yield at the foundries the Zen design and product line contains plenty redundancy to repurpose "bad" dies.

We pretty much know Zen 2 is coming for desktop in 2019, I'm not sure about server though, could hit server earlier in late 2018?
Zen + pinnacle ridge Q1 2018.
Im really hoping for 14nm LPU or IBM 14HP.

 

[.vodka]

LPU could increase transistor switching speed some within still the 14nm low power process allowing AMD to increase (potentially) clocks higher than the current 14nm LPP process. But just because the transistors can handle faster switching doesn't mean the cpu design can. But if AMD uses the Samsung LPU process they should be be to do a remask/layout and benefit from the better 14nm lpu process. Thus improving performance with (and maybe very low hanging fruit IPC improvements) increased clock-speeds.

Of course I'm not an expert,

Zen can and does go above 4GHz without much issue apart from CPU safety. Current Ryzen products are process limited. Keep feeding it voltage on a robust board like the C6H or Taichi and you'll hit 4200-4500MHz, but then you're at 1.5-1.6v and you're about to fry your CPU into a crisp. Add LN2 to the mix and you'll hit 5.8GHz. If Zen couldn't scale further than that, well, it wouldn't nudge much further even under LN2. Their actual 14nm process is dirt cheap, so at least AMD is enjoying some great margins right now and earning some much needed cash.

If Pinnacle Ridge can move clocks up 200-300Mhz in the same generational node, it would be a great performance bump, nevermind if they've decided to do little tweaks here and there to bring a little IPC bump too. This could be a Phenom II C2 -> C3 or i7 920 C0 -> D0 style bump, or a Zambezi -> Vishera. Who knows. What we do know is that the actual 14nm process in use targeted 3GHz operation and AMD does a practical limit of 4GHz on it!

7nm Zen 2 should not have these problems scaling above 4GHz (since the process targets 5GHz operation) and should also bring a decent IPC bump... but that's for 2019.

 

[Eric1987]

I would love for the next ryzen to be an 8 core clockable to 5GHz. Would be sex.

 

[Tee9000]

Zen can and does go above 4GHz without much issue apart from CPU safety. Current Ryzen products are process limited. Keep feeding it voltage on a robust board like the C6H or Taichi and you'll hit 4200-4500MHz

Whhhaaaaat?! Can I please see a couple of examples of Zen 4.2-4.5 Ghz not on LN2?

 

[.vodka]

https://www.reddit.com/r/Amd/comments/6d1b33/let_me_see_your_ryzen_cpuz_validations_new_ones/

4.1GHz 1700x 8C16T
https://valid.x86.fr/rsrsdg

4.275GHz 1800x 8C16T
https://valid.x86.fr/t2n1ac

http://www.legitreviews.com/amd-ryzen-5-1500x-overclocked-benchmark-results-at-4-2ghz_193599


etc. google is your friend. Ryzen's current limits aren't architecture related. Those clocks can be done with enough luck at the silicon lottery and good enough cooling (not LN2) disregarding your CPU's safety at the long term since anything higher than 1.4v is dangerous according to AMD. If Pinnacle Ridge brings a better physical implementation on a better 14nm node, then those 4.2-4.3GHz golden samples could become the norm, and then you're 500-700MHz behind Kabylake max practical OC, not 1-1.2GHz behind. Every improvement is welcome.

Trinity -> Richland brought 300-400MHz extra clocks across the board, Kaveri -> Godavari had a similar clock bump... so did Carrizo -> Bristol Ridge, they got A LOT out of that horrible glofo 28nm process. It's possible, AMD has done this fine tuning of their designs before with excellent results relative to the original.

 

[formulav8]

Current Ryzen products are process limited.

While the current lpp process could be a factor, its not necessarily the problem when it comes to clocks. There could very well be a layout issue. Unless AMD releases info, it is what it is, speculation.

I'd still expect an LPU 14nm Zen+ part with the PRidge core. Clocks should improve and they may be able to do a BD to PD IPC improvement. (nil to 7% increase)

 

[Tee9000]

https://www.reddit.com/r/Amd/comments/6d1b33/let_me_see_your_ryzen_cpuz_validations_new_ones/

4.1GHz 1700x 8C16T
https://valid.x86.fr/rsrsdg

4.275GHz 1800x 8C16T
https://valid.x86.fr/t2n1ac

http://www.legitreviews.com/amd-ryzen-5-1500x-overclocked-benchmark-results-at-4-2ghz_193599

4.1 Ghz? I was asking 4.2-4.5. CPU-Z validations is one thing, stable OC is another. Legitreviews is the only example I'll take.

I was asking because I've seen renowned overclockers struggle to go above 4.2 on water, chilled water, let alone reach 4.5, and they had many samples at their disposal to cherry pick. And I've also seen a few seasoned overclockers struggle to reach stable 4.0 Ghz. So I can't say as you that 4.0 Ghz is easily doable or that OC can reach 4.5 Ghz.

 

[.vodka]

I know CPUz validations are one thing, and stable clocks are a whole different stuff. I remember seeing someone push their 1800x all the way to 4.4GHz but at something insane like 1.6v. The thing keeps scaling if pushed and that points at the process used and the physical implementation being the liming factor. For all intents and purposes Ryzen in its current form tops at 4-4.1GHz, usable, 24/7.

My point was that the CPU will work at those higher clocks in some form, and those could become the norm with Pinnacle Ridge. Hopefully.

 

[dlerious]

I know CPUz validations are one thing, and stable clocks are a whole different stuff. I remember seeing someone push their 1800x all the way to 4.4GHz but at something insane like 1.6v. The thing keeps scaling if pushed and that points at the process used and the physical implementation being the liming factor. For all intents and purposes Ryzen in its current form tops at 4-4.1GHz, usable, 24/7.

My point was that the CPU will work at those higher clocks in some form, and those could become the norm with Pinnacle Ridge. Hopefully.

I took my 1800X up as high as 4150 all cores at 1.59v with a Noctua NH-D15. I only kept it there for a minute or two before I bumped it back, running 3.9 at 1.37v now. Maybe I'll see what I can get again when I move the system from the test bench to a case and my ambients drop (80 F now).

 

[wahdangun]

http://www.tomshardware.com/reviews/overclocking-amd-ryzen-ln2,5116-4.html

look at tomshardware finding, its look like ryzen truly scale above 4,2 Ghz, with voltage and lower temp.

maybe it was because ryzen is dense and need better cooling to prevent electron migration with higher voltage to be stable, just like GCN.

maybe in 7nm or 14nm++ they will make it less dense.

 

[moinmoin]

While the current lpp process could be a factor, its not necessarily the problem when it comes to clocks.

The 14LPP process is optimized for around 3Ghz, 7LP will be around 5Ghz. Process is the one main factor frequencies above 4.1Ghz currently are not feasible for Ryzen chips and why Epyc chips max out at 3.3Ghz, the frequency where 14LPP is at its most efficient.

 

[csbin]

ARCTIC announces Liquid Freezer compatibility with Ryzen Threadripper

https://videocardz.com/newz/arctic-announces-liquid-freezer-compatibility-with-ryzen-threadripper

 

[csbin]

ASUS, Gigabyte, MSI and ASRock to detail X399 motherboards on July 25th

https://videocardz.com/71002/asus-gigabyte-msi-and-asrock-to-detail-x399-motherboards-on-july-25th

 

[french toast]

It certainly is process imo, just imagine ryzen being fabbed on Intel's 14nm ++?? I would expect 30℅ higher clocks (really).

 

[ddogg]

ASUS, Gigabyte, MSI and ASRock to detail X399 motherboards on July 25th

https://videocardz.com/71002/asus-gigabyte-msi-and-asrock-to-detail-x399-motherboards-on-july-25th

I really like that zenith board but it's probably going to cost an arm and leg

 

[ajc9988]

I really like that zenith board but it's probably going to cost an arm and leg

So, the crosshair VI extreme was announced. $350. That means the cost will be comparable to, but possibly lower than, the X299 boards for the X399 boards.

http://www.tweaktown.com/news/58486/asus-rog-crosshair-vi-extreme-announced-ready-ryzen/index.html

 

[ddogg]

So, the crosshair VI extreme was announced. $350. That means the cost will be comparable to, but possibly lower than, the X299 boards for the X399 boards.

http://www.tweaktown.com/news/58486/asus-rog-crosshair-vi-extreme-announced-ready-ryzen/index.html

I'll probably buy it at 350 but if it's 500+ I'm going to pass.

 

[formulav8]

It certainly is process imo

Sure it could be strictly process, but there could be something else core-wise that a layout change could help give more clocking headroom. For instance, when AMD released TBred on their .13 process, it was not any better than .18 clocks. Turns out it wasn't the process and AMD added a 9th metal layer to the core so they could do some re-routing and it ended up as TBred B which greatly increased the clockspeed on the same process.

 

[formulav8]

Has any kind of TR mobo pricing data surfaced?

 

[ajc9988]

I'll probably buy it at 350 but if it's 500+ I'm going to pass.

That is for the AM4 socket. The Zenith board is for TR4. So, it could be $500.

 

[tamz_msc]

I think that it was speculated that the Zenith Extreme would be around 500-600$ when it was first revealed during Computex.

 

[ajc9988]

I think that it was speculated that the Zenith Extreme would be around 500-600$ when it was first revealed during Computex.

It makes sense. Whatever features it lacks from x299, it makes up for in trying to add traces for the extra pcie. The VRM are the same, as is a lot of I/O

 

[scannall]

It certainly is process imo, just imagine ryzen being fabbed on Intel's 14nm ++?? I would expect 30℅ higher clocks (really).

For the 6 and 8 core parts I'd agree. But considering the thermal problems Intel is having, it's fairly likely that 16 core Threadrippers will clock higher than 16 core Intel parts.