AMD CPUs, SoC Rumors and Speculations Temp. thread 2

[juanrga]

I didn't say that AMD chose 14LPP over 14LPE due to costs. My point was that "The only reason why AMD chose 14LPP for all the chips is because AMD doesn't have enough money to select different process nodes", with "nodes" in plural. Instead implementing each chip in the more adequate process node (mobile on 14LPE, enthusiast on 14HP) or some other combination, AMD is using a single process node for all the chips. And this is not a decision taken by technical reasons but by economic reasons.

There are different kind of servers. IBM is using 14HP because their servers target very high clocks. But I agree that AMD server CPUs will target lower clocks, probably the 32 core Zen could be clocked around 2GHz.

About LPP clocks, I am hearing from people as Thevenin that:

The 14nm LPP performs best (efficiency wise) up to 2.5GHz, and takes a significantly nose dive at frequencies above 3GHz. [...] Samsung 14nm LPE/P is a truly mobile / embedded / specialized ASIC optimized process.

Just to point out, you are quoting the same comment from a year ago...which was speculation then...and is speculation now.

As for IBM, as far as I am aware, Power9 top core count part runs at 2.5 GHz like Power8. As Yuka pointed out, they are after efficiency...not clockspeeds.

Additionally...LPP will have much better efficiency than that of 14HP, not to mention, GloFo does not even carry 14HP, either.

Yes, the quote is from November, but 14LPP is the same node then than now.

Indeed 14LPP is a more efficient node than 14HP. The node selected by AMD is optimized for power and low clocks, whereas the node selected by IBM is optimized for performance and highest clocks. I don't know what "GloFo does not even carry 14HP, either" means.

The clocks reported above for Power8 are incorrect. IBM is shipping the top core count processor at frequencies of 4.024 GHz, 4.190 GHz, and 4.350 GHz, albeit the Power8 core itself supports up to 5GHz. And IBM hinted that Power9 is designed to hit similar frequencies than its predecessor, and that is why IBM engineers are using 14HP Globalfoundries process.

 

[-Fran-]

Yes, current Power8 CPUs from IBM and not sourced from the OpenPower thingy, max out at 4.2-ish Ghz. What I've read, is they will keep the 4Ghz ceiling if possible.

In that same note... Then the 14LPP node is a denser node than 14HP? Are there other key differences between them?

Cheers!

 

[salgado18]

Don't forget that AMD wants Zen to serve desktop and servers, but also mobile and low power. I believe they want to replace Jaguar with Zen. So it must use low power process to achieve those targets. They probably went with 14LPP instead of 14HP or other high-power because of those mobile targets, not because of servers.

I also remember a slide from them showing they expected a growth in every market segment except desktop, which was declining in size year after year. They are targeting efficiency for mobile and servers first, and then desktop with that they can get out of it.

Yes Zen replaces both Jaguar and Piledriver/Excavator. But my point wasn't that AMD would chose 14HP instead 14LPP. My point was other:

"The only reason why AMD chose 14LPP for all the chips is because AMD doesn't have enough money to select different process nodes", with "nodes" in plural. Instead implementing each chip in the more adequate process node (mobile on 14LPE, enthusiast on 14HP or some other combination), AMD is using a single process node for all the chips. And this is not a decision taken by technical reasons but by economic reasons.

Yes, I got your point, it's definitely that. I'm just answering this other post of you:

The 14LPP node cannot compete in high performance with 14HP with 16FF+ or with 14P1271. The 14LPP cannot compete with 16FC or 14LPE in mobile. Thus AMD choice is going to hurt them sure, how much remains to see.

They choose LPP over HP because they need the scalability over the performance. Since they had to choose one, it would be the efficient one. Yes, it's going to hurt them, but I don't think it's that much, if they play the right marketing and product targeting. They also have the flexibility to expand, with just one arch and one process, into many markets. They a) won't need to branch development and products, and b) have no other option.

 

[Cazalan]

Samsung doesn't make 14HP. So if AMD wanted a backup 14LPP was their only option.

 

[8350rocks]

Samsung doesn't make 14HP. So if AMD wanted a backup 14LPP was their only option.

This is very much true.

Hopefully the node gets some clocking headroom and there will be some nice refresh chips a la Richland.

 

[ComputerSecurityGuy]

My guess is that in clockspeed the Zen chips will be on par with Broadwell-E, but it's just my guess.

 

[juanrga]

My guess is that in clockspeed the Zen chips will be on par with Broadwell-E, but it's just my guess.

My guess is octo-core Zen will have lower clocks than octo-core Broadwell. I expect 3GHz base for Zen.

 

[cdrkf]

My guess is that in clockspeed the Zen chips will be on par with Broadwell-E, but it's just my guess.

My guess is octo-core Zen will have lower clocks than octo-core Broadwell. I expect 3GHz base for Zen.

I agree this looks possible- although my personal prediction is (based on what AMD have been doing recently with their other cpu designs) that Zen will feature a higher turbo range than Broadwell, I think up to 1ghz range on Turbo is possible- provided they can achieve a reasonable Turbo clock with at least half the cores (~3.7ghz +) then the processor will scale nicely between loads.

Ideally I'd like to see 4ghz+ on single core, 3.7+ up 4 cores, 3ghz base on the whole chip (and even then it might turbo slightly, although I guess that'd be silicon dependent so wouldn't be guaranteed, maybe 3.2ghz on all 8 cores active).

This is the thing with modern processors, base clock isn't that relevant. What is more important is what clocks are sustainable under different scenarios. A well designed turbo system, with built in current and temp monitoring in silicone itself, can boost performance nicely depending on situation. If the chip is stuck at 3ghz all the time, that'd be a problem. 3ghz when running all 16 threads, less so. AMD have gotten pretty good with Turbo lately (just look at the huge ranges they're getting out of their mobile APU's). What might be an issue is if the turbo performance turns into a bit of a 'Silicone lottery'- which is looking somewhat evident with the Polaris 10 gpu (with some parts purportedly being clock-able in excess of 1.5ghz on air, others being stuck at 1.26ghz). Then again, if that is an issue AMD can probably solve that with good binning- offering parts with lower sustainable turbo clocks as a lower model.

I remain quietly optimistic for Zen- no I'm not expecting it to unseat Intel in the high end, but I think we'll see something competitive that sits comfortably above the incumbent quad core i7 parts in threaded workloads, and is close in lower thread software (e.g. games) thanks to good tubro scaling. That would also suggest pricing in the middle range of Intels Broadwell-E line up for the octo core Zen part (but a lower overall platform price due to dual channel memory and resulting cheaper motherboards which should be more akin to Intels mainstream platform). I think that'd present consumers with a worthwhile alternative.

 

[juanrga]

I expect 3.5GHz turbo.

 

[cdrkf]

I expect 3.5GHz turbo.

On how many cores?

 

[juanrga]

I expect 3.5GHz turbo.

On how many cores?

One, I am reporting maximum turbo that I expect.

 

[ComputerSecurityGuy]

Considering AMD historical clock boosts I would think 3.0/3.3/3.6 or 3.2/3.5/3.8 would be reasonable expectations, though 14nmLPP is a new node. Also remember that Intel has been trying some Turbo Boost tricks as well.

 

[jimmysmitty]

Considering AMD historical clock boosts I would think 3.0/3.3/3.6 or 3.2/3.5/3.8 would be reasonable expectations, though 14nmLPP is a new node. Also remember that Intel has been trying some Turbo Boost tricks as well.

Intel was the first to use Turbo Boosts with Nehalem. Of course they will try to get the most out of it.

As you said though the 14nm LPP is a new node and it is also a node designed for a vastly different purpose than high powered desktop CPUs so how it will react to that is anyones guess.

 

[ComputerSecurityGuy]

Yup, especially if you have been watching some tricks they've been playing with on Broadwell-E.

 

[Cazalan]

Are we using Price is Right rules? Closest one without going over wins? 3.8Ghz

 

[cdrkf]

Cazlan, I already called 3.8... 😛

Edit: doh! I called 4 (assuming we're talking Max turbo). Computerscienceguy called 3.8 😛

Juan, ever the pessimist, has 3.5...

 

[gamerk316]

If we're talking max Turbo, then I'll go with infinity; thermals are the limiting factor, after all.

If we're talking base clock, I'll go with 3.2.

 

[Cazalan]

Cazlan, I already called 3.8... 😛

Edit: doh! I called 4 (assuming we're talking Max turbo). Computerscienceguy called 3.8 😛

Juan, ever the pessimist, has 3.5...

Yeah max turbo 1 core. Technically Computerscienceguy called 2 different numbers, but I'll say 3.9 anyway.

I'm just not seeing a reason for a significant regression on the 28nm to 14nm change when the GPUs aren't regressing.

 

[jimmysmitty]

Are we using Price is Right rules? Closest one without going over wins? 3.8Ghz

"I'll do 3.79GHz Bob."

 

[cdrkf]

@gamerk by 'max turbo' we're taking at stock settings on the fastest sku of the desktop first gen zen processor. Infinity is probably a little high 😛 but if your feeling that optimistic should I put you down for 4.5 ghz? Winner gets 100 kudos points 😛

 

[Krnt]

4.5GHz for max turbo seems a little high, for the first releases of Zen, but seems like a the max frequency that we could achieve by overclocking it without using LN2 or phase changing, I even doubt 4.0GHz, 3.7GHz is what I could expect based on the conversations I've read.

However for default clocks around 3GHz sounds reasonable, less than that would be a huge letdown for an enthusiast CPU.

 

[Vogner16]

Cazlan, I already called 3.8... 😛

Edit: doh! I called 4 (assuming we're talking Max turbo). Computerscienceguy called 3.8 😛

Juan, ever the pessimist, has 3.5...

what were the other choices??? Ill take 1!

 

[cdrkf]

Cazlan, I already called 3.8... 😛

Edit: doh! I called 4 (assuming we're talking Max turbo). Computerscienceguy called 3.8 😛

Juan, ever the pessimist, has 3.5...

what were the other choices??? Ill take 1!

3.6 and 3.9 are both up for grabs I believe 😛

 

[Vogner16]

Cazlan, I already called 3.8... 😛

Edit: doh! I called 4 (assuming we're talking Max turbo). Computerscienceguy called 3.8 😛

Juan, ever the pessimist, has 3.5...

what were the other choices??? Ill take 1!

3.6 and 3.9 are both up for grabs I believe 😛

price is right rules! I want those internet points!

 

[esrever]

Put me down for 3.2 base 4ghz turbo for top end part.