AMD - Zen chitchat

[DanceswithUnix]

Some good IPC improvements in games,and it might be partly down to the improved cache latencies??

That and the new boost system I expect. Games tend to fall into a middle ground where they aren't single threaded but don't use all the cores. Or is that with the boards overclocked to a flat 4GHz all the time? Can't watch vids atm, am working.

However they did it, nice result.

[CAT-THE-FIFTH]

That and the new boost system I expect. Games tend to fall into a middle ground where they aren't single threaded but don't use all the cores. Or is that with the boards overclocked to a flat 4GHz all the time? Can't watch vids atm, am working.

However they did it, nice result.

The CPUs were both clocked at 4GHZ,so it was a test of core performance.

I also added some more charts and more info to the post - it appears the review was done on an X370 based motherboard,and the X470 based ones might be a better fit,and AFAIK the improved boost support is a feature of the newer chipsets.

Edit!!



3200MHZ RAM was used for the stock games tests.

[Firejack]

Significant boost from faster memory speed too. Thinking it might be worth spending the extra for an 8 pack TeamGroup Dark Pro/Xtreem kit on OcUK. Seem to be the cheapest high speed 16GB kits around at the moment.

[DanceswithUnix]

OK that's pretty impressive then. That still limits what's changed to "just about anything" mind
Cache, PCIe buffering, ram interface as well as the cores themselves. Got to keep those puppies fed!

[CAT-THE-FIFTH]

Looks like most of the uplift is down to the latency improvements I suspect.

[Corky34]

OK that's pretty impressive then. That still limits what's changed to "just about anything" mind
Cache, PCIe buffering, ram interface as well as the cores themselves. Got to keep those puppies fed!

Obviously it's by no means a comprehensive list of changes but AFAIK the only thing that's changed from R1 to R2 is the node size and cache latencies, oh and the new versions of XFR and PB (would they be operating on an X370 board?)

It's odd the slide Cat posted is saying they improved latencies on L1 and L3, would that be down to improved cache prefetch?

[DanceswithUnix]

Obviously it's by no means a comprehensive list of changes but AFAIK the only thing that's changed from R1 to R2 is the node size and cache latencies, oh and the new versions of XFR and PB (would they be operating on an X370 board?)

It's odd the slide Cat posted is saying they improved latencies on L1 and L3, would that be down to improved cache prefetch?

I would expect there to be a whole bunch of errata fixed, and some bugs can really hurt performance.

[CAT-THE-FIFTH]

I would expect there to be a whole bunch of errata fixed, and some bugs can really hurt performance.

AFAIK,SMT could lead to performance issues with Ryzen in some games. It could be quite possible,that there are improvements there??

Its just interesting looking at the gaming improvements,clock for clock - if that is replicated by other reviews,those fixes have yield some decent improvements. Then if you add the few hundred mhz in additional clockspeed,that is a good improvement in around 12 months IMHO OFC.

[Biscuit]

Well played AMD


Have their marketing department finally managed to somehow... work?

[DanceswithUnix]

AFAIK,SMT could lead to performance issues with Ryzen in some games. It could be quite possible,that there are improvements there??

Its just interesting looking at the gaming improvements,clock for clock - if that is replicated by other reviews,those fixes have yield some decent improvements. Then if you add the few hundred mhz in additional clockspeed,that is a good improvement in around 12 months IMHO OFC.

I can't see how they could directly improve SMT, but lots of things could help smooth that problem including the cache speed as sharing data between cores can really hurt performance (part of what makes multi threaded code hard to write is keeping threads off each other's active cache lines). The other possibility is the improved boost system may well include better core parking/unparking which again changes cache access patterns.

But the only way we could know what is going on is to measure it. I presume Windows has some equivalent to the Linux kernelshark to measure this sort of thing.

But yeah, it's pretty good progress. If I feel wealthy at some point I will no doubt buy one, though I am still waiting for a decent priced Vega first.

[scaryjim]

... the new versions of XFR and PB (would they be operating on an X370 board?) ...

According to leaked roadmap and slides (e.g. https://hexus.net/tech/news/cpu/1160...ap-specs-leak/), sort of

AFAICT XFR 2 is basically just a marketing name for PB 2 on 2000X processors, and will give it the opportunity to boost above the normal PB2 level for load if there's thermal headroom. However there seems to be an additional 'PB Overdrive' that requires an X processor and 400-series motherboard, and should give better boost clocks with intermediate core loads. None of them seem to allow boosting past the advertised max boost clock, but given you very rarely see a pure single-thread load nowadays it should play out quite well...

[Firejack]

The Ryzen 7 1800X is £229.97 @ Amazon (for prime members) this morning. Do wonder just how low the first Ryzen series will go to clear them out. Looking over the early numbers the extra £50 for the Ryzen 7 2700X looks the better buy. If the £1800X creeps closer to £200 it will give people a headache trying to pick the best.

[Ozaron]

The Ryzen 7 1800X is £229.97 @ Amazon (for prime members) this morning. Do wonder just how low the first Ryzen series will go to clear them out. Looking over the early numbers the extra £50 for the Ryzen 7 2700X looks the better buy. If the £1800X creeps closer to £200 it will give people a headache trying to pick the best.

Just think of the market penetration!

[Corky34]

According to leaked roadmap and slides (e.g. https://hexus.net/tech/news/cpu/1160...ap-specs-leak/), sort of

AFAICT XFR 2 is basically just a marketing name for PB 2 on 2000X processors, and will give it the opportunity to boost above the normal PB2 level for load if there's thermal headroom. However there seems to be an additional 'PB Overdrive' that requires an X processor and 400-series motherboard, and should give better boost clocks with intermediate core loads. None of them seem to allow boosting past the advertised max boost clock, but given you very rarely see a pure single-thread load nowadays it should play out quite well...

Yea it's about as clear as mud for me, i get the need for keeping higher boosts when more cores are lightly loaded but having three different names (Precision Boost 2.0, XFR 2, and Precision Boost Overdrive) sure is confusing, especially when throwing 3 and 4 series motherboards into the mix.

[CAT-THE-FIFTH]

I can't see how they could directly improve SMT, but lots of things could help smooth that problem including the cache speed as sharing data between cores can really hurt performance (part of what makes multi threaded code hard to write is keeping threads off each other's active cache lines). The other possibility is the improved boost system may well include better core parking/unparking which again changes cache access patterns.

But the only way we could know what is going on is to measure it. I presume Windows has some equivalent to the Linux kernelshark to measure this sort of thing.

But yeah, it's pretty good progress. If I feel wealthy at some point I will no doubt buy one, though I am still waiting for a decent priced Vega first.

If the results are replicated,whatever they have done its not bad at all,as its been barely a year since Ryzen MK1 came out.

I also like the fact AMD is trying to keep to yearly cadences - I just hope GF/TSMC can keep to their end of the bargain next year.

The Ryzen 7 1800X is £229.97 @ Amazon (for prime members) this morning. Do wonder just how low the first Ryzen series will go to clear them out. Looking over the early numbers the extra £50 for the Ryzen 7 2700X looks the better buy. If the £1800X creeps closer to £200 it will give people a headache trying to pick the best.

The Ryzen 7 2700X also has a cooler included and it even has RGB!!

Yea it's about as clear as mud for me, i get the need for keeping higher boosts when more cores are lightly loaded but having three different names (Precision Boost 2.0, XFR 2, and Precision Boost Overdrive) sure is confusing, especially when throwing 3 and 4 series motherboards into the mix.

MOAR buzzwords=MOAR sales surely??

[Corky34]

Ahead of the reviews later today WikiChip Fuse has published an article on Zen CPU Complex, Cache, and SMU, i don't mind admitting that a fair amount of what they talk about on page 2 went right over my head but maybe a more technically minded Hexus reader will find it interesting.