AMD - Zen chitchat

[kompukare]

So GF 12NM is 15% denser due to the libraries used:

https://www.anandtech.com/show/12233...-vega-on-7nm/8

Hope that isn't a bad thing... More density should mean cheaper to make chips, but if that results in lower clocks, more heat or similar, those cheaper to make chips will have to be sold for less. Hawaii's impressive density compared to something like GK100 didn't really do it too much good - a bigger less dense Hawaii chip which either had higher clocks or clocked slower but wider might have gotten far better reviews. Instead, AMD were forced to drop the prices compared to 780, 780Ti and later GTX970.

[CAT-THE-FIFTH]

Hope that isn't a bad thing... More density should mean cheaper to make chips, but if that results in lower clocks, more heat or similar, those cheaper to make chips will have to be sold for less. Hawaii's impressive density compared to something like GK100 didn't really do it too much good - a bigger less dense Hawaii chip which either had higher clocks or clocked slower but wider might have gotten far better reviews. Instead, AMD were forced to drop the prices compared to 780, 780Ti and later GTX970.

AMD seems to be saying that Zen+ has improved performance/watt over Zen though,but there isn't much noise about performance or clockspeeds.

[kompukare]

AMD seems to be saying that Zen+ has improved performance/watt over Zen though,but there isn't much noise about performance or clockspeeds.

Yes, read in the AT article.

Thing is, it's know without context as Ryzen already has incredible efficient at most frequencies below about 3.8GHz. Plus 10% in perf/watt for a 1700 could mean as little as a reduction down to 59W from 65W which is pretty pointless. Still, AMD have been very reluctant to pay for multiple masks recently, so for them to be doing a 12nm respin there must be more in it that a few watts.

4.3GHz with cherry picked 'golden samples' able to reach 4.6GHz or so. Although those might all get reserved for Threadripper2 assuming Zen+ is not going to be used for EPYC. So jump from the 4.2Ghz max of lucky 1950X to a luck 4.6 for a hypothetical 2950X. If they can tweak the IMC, inter-CCX or similar that's an added bonus.

[CAT-THE-FIFTH]

Yes, read in the AT article.

Thing is, it's know without context as Ryzen already has incredible efficient at most frequencies below about 3.8GHz. Plus 10% in perf/watt for a 1700 could mean as little as a reduction down to 59W from 65W which is pretty pointless. Still, AMD have been very reluctant to pay for multiple masks recently, so for them to be doing a 12nm respin there must be more in it that a few watts.

4.3GHz with cherry picked 'golden samples' able to reach 4.6GHz or so. Although those might all get reserved for Threadripper2 assuming Zen+ is not going to be used for EPYC. So jump from the 4.2Ghz max of lucky 1950X to a luck 4.6 for a hypothetical 2950X. If they can tweak the IMC, inter-CCX or similar that's an added bonus.

Well apparently the Zen APUs were hitting 4.2GHZ when overclocked,so if these don't hit at least 4.5GHZ,it would be a touch meh when it comes to clockspeeds.

[Xlucine]

Be silly not to upgrade the EPYC chips. For one, with OG zen leaving production there's no opportunity to harvest the best chips anymore. The other big benefit is that 10% higher perf/watt is a couple hundred MHz more speed for a consumer, but for a server it's 10% lower running costs

[scaryjim]

Be silly not to upgrade the EPYC chips. For one, with OG zen leaving production there's no opportunity to harvest the best chips anymore. The other big benefit is that 10% higher perf/watt is a couple hundred MHz more speed for a consumer, but for a server it's 10% lower running costs

Who says 14nm Zen is leaving production? If 12nm LP is a performance oriented node rather than a power-oriented node, then at the TDPs the server chips play in (as low as 30W per die) 12nm might make negligible difference. I rather suspect the 10% perf/watt is at the higher 65W and 95W per die levels that the desktop chips utilise. And after a year, 14nm should be yielding better, so they should be able to bin more parts for EPYC at potentially lower power draw or higher performance.

That's not to say that they won't transition EPYC to 12nm processors, of course, but from what we know about the refresh and the characteristics of the two nodes I think it would be understandable if they didn't switch EPYC to the new dies straight away.

[Xlucine]

Who says 14nm Zen is leaving production? If 12nm LP is a performance oriented node rather than a power-oriented node, then at the TDPs the server chips play in (as low as 30W per die) 12nm might make negligible difference. I rather suspect the 10% perf/watt is at the higher 65W and 95W per die levels that the desktop chips utilise. And after a year, 14nm should be yielding better, so they should be able to bin more parts for EPYC at potentially lower power draw or higher performance.

That's not to say that they won't transition EPYC to 12nm processors, of course, but from what we know about the refresh and the characteristics of the two nodes I think it would be understandable if they didn't switch EPYC to the new dies straight away.

They could keep it in production, but EPYC is claimed to be the top 5% bin (or whatever). Without 1XXX ryzen CPUs to use the other chips, yields would have to get staggeringly good for them to produce 14nm EPYC parts without scrapping huge amounts of workable (but not quite good enough for EPYC) chips

[CAT-THE-FIFTH]

I suspect 12NM Epyc will follow but these parts probably need extra validation and hence this is why you are only seeing the actual large scale rollout of Epyc systems happening in the last few months. IIRC,AMD is targeting between 5% to 10% server share over the next few years,IIRC.

[watercooled]

Binning isn't just a scalar process - there's more than one quality assessed when testing dies and most EPYC parts won't be chasing the high clock speeds desired for HEDT. Power vs clock speed isn't linear or equal between dies either. A difference is that they probably bin server parts more conservatively than those for the desktop market.

IIRC the 'top 5%' comment was about Threadripper where it makes sense - at similar clocks to Ryzen (i.e. pushing it before TDP runs away) they want to keep TDP down for obvious reasons, and 5% is likely enough volume for that market. EPYC's clocks generally aren't that high in comparison, no sane datacentre user is going to attempt an overclock, and I'm not sure what sort of volume they're shipping but I doubt even 5% of the dies sold would be sufficient for the demand.

And as others have said, the qualities that make 12nm desirable for Ryzen doesn't necessarily make it suitable (or worthwhile) for EPYC.

[CAT-THE-FIFTH]

I think its more down to validation. The Zen MK1 cores used in Epyc are only now starting to enter any volume deployments,with companies like Baidu. I assume Zen+ will need validation before being deployed,so we might see it at a later date,or AMD might just skip it for Zen2.

[watercooled]

Agreed - assuming it's not somehow inferior at EPYC's clocks and the finished wafer costs aren't substantially higher after accounting for yield, it makes little sense not to use the 12nm version after it's ramped.

[DanceswithUnix]

Just been looking for a work laptop on the Dell website. No AMD Ryzen options to be seen

[Numpti]

The new Ryzen 2 (zen+) processors won't be showing up in devices until around April.

[CAT-THE-FIFTH]

Just been looking for a work laptop on the Dell website. No AMD Ryzen options to be seen

Apparently Dell is launching one and I did see a mention on a techsite recently about it.

[kompukare]

Apparently Dell is launching one and I did see a mention on a techsite recently about it.

Well, NBC had this story about a business-class Latitude 5495:
https://www.notebookcheck.net/AMD-Ra....264272.0.html
but that was posted in November.
Over Reddit it also comes up along with mention of a ThinkPad A485 and A285:
https://www.reddit.com/r/Dell/commen...ith_amd_ryzen/

So it looks like some business-class stuff is getting released, but a Latitude 5495 sounds like a lesser model compared to the E7480 or E7490; and those ThinkPad A series are also likely to slot in below the ThinkPad T series. Guess if HP were releasing them, they would be called ProBooks (although HP previously did do EliteBook 745 (or similar) which used quite low end AMD processors - but were 'proper' EliteBooks. Proper in this case meaning some metal in the case and a maintenance hatch.

[CAT-THE-FIFTH]

Well, NBC had this story about a business-class Latitude 5495:
https://www.notebookcheck.net/AMD-Ra....264272.0.html
but that was posted in November.
Over Reddit it also comes up along with mention of a ThinkPad A485 and A285:
https://www.reddit.com/r/Dell/commen...ith_amd_ryzen/

So it looks like some business-class stuff is getting released, but a Latitude 5495 sounds like a lesser model compared to the E7480 or E7490; and those ThinkPad A series are also likely to slot in below the ThinkPad T series. Guess if HP were releasing them, they would be called ProBooks (although HP previously did do EliteBook 745 (or similar) which used quite low end AMD processors - but were 'proper' EliteBooks. Proper in this case meaning some metal in the case and a maintenance hatch.

Its going to take time for AMD to get traction with OEMs,especially since the Intel designs have in theory not changed much from Skylake,so are probably quicker to implement.

However,once AMD Ryzen based models start showing up,I suspect the second generation ones will enter the market much quicker.