AMD - Zen chitchat

[kompukare]

Computerbase went and ran some more meaningful benchmarksthis time with the Meltdown patch and the Spectre microcode updates:
https://www.computerbase.de/2018-01/...el-benchmarks/

They ran these with KabyLake, SandyBridge and Ryzen all under both Win7 and Win10.

Only the Kabylake system had the BIOS microcode update though, so if/when some of the other CPUs get updates these benches will have to be re-run yet again.

For Kabylake the worst performance drop seems to Assassin's Creed Origins which suffered a 9% slowdown:

and disk IO where 4KibQ32T1 seems have suffered a lot.


Sandy Bridge suffered around 16% in AC:O

[Yes it looks like CB have to fix their label there.]

Ryzen didn't suffer any slowdowns from the new patches but then it shouldn't at least for Meltdown. We'll have to see what if anything AMD's microcode update for Spectre brings.

[CAT-THE-FIFTH]

Computerbase went and ran some more meaningful benchmarksthis time with the Meltdown patch and the Spectre microcode updates:
https://www.computerbase.de/2018-01/...el-benchmarks/

They ran these with KabyLake, SandyBridge and Ryzen all under both Win7 and Win10.

Only the Kabylake system had the BIOS microcode update though, so if/when some of the other CPUs get updates these benches will have to be re-run yet again.

For Kabylake the worst performance drop seems to Assassin's Creed Origins which suffered a 9% slowdown:

and disk IO where 4KibQ32T1 seems have suffered a lot.


Sandy Bridge suffered around 16% in AC:O

[Yes it looks like CB have to fix their label there.]

Ryzen didn't suffer any slowdowns from the new patches but then it shouldn't at least for Meltdown. We'll have to see what if anything AMD's microcode update for Spectre brings.

Even the other games tested showed reductions in minimums.

I just tested FO4,which I had not run before the latest update and it does seem a bit more choppier. I tried uninstalling the update,but Windows said I needed to reboot and then it installed it again. FFS.

[kompukare]

Even the other games tested showed reductions in minimums.

I just tested FO4,which I had not run before the latest update and it does seem a bit more choppier.

One way of benching FO4 (and Skyrim of course) is to find something which moves the player in a predictable way. In Skyrim, the intro with the carriage or a mod like the various carriage mods is good for this. Then use something like MSI Afterburner to log the FPSs to a CSV file.

Alternatively just stand somewhere with an on-screen FPS counter but that's not as accurate. A long ride with a carriage also means the background world segment loading part gets benched.

I tried uninstalling the update,but Windows said I needed to reboot and then it installed it again. FFS.

Ah, yes the new helpfulness of Windows 10! Even worse when it tries to do with a (wrong) driver.

[CAT-THE-FIFTH]

One way of benching FO4 (and Skyrim of course) is to find something which moves the player in a predictable way. In Skyrim, the intro with the carriage or a mod like the various carriage mods is good for this. Then use something like MSI Afterburner to log the FPSs to a CSV file.

Alternatively just stand somewhere with an on-screen FPS counter but that's not as accurate. A long ride with a carriage also means the background world segment loading part gets benched.



Ah, yes the new helpfulness of Windows 10! Even worse when it tries to do with a (wrong) driver.

FO4 is only really worth testing in my settlements as that is where performance suffers the most,ie,more NPCs and lots of assets to load.I tried running it off a HDD and it became unplayable. TBH,if you do three to six run throughs,and overlay them its enough to look at the trend in FPS:

https://forums.hexus.net/reader-revi...enchmarks.html

Its what I did with the GTX1080 review,so at least you get a gist of what performance should look like.

I would try and find a way to uninstall the update,and do a proper test,but I apparently have caught the flu,so might have to delay that.

I have downloaded the latest Nvidia driver with the security fix,so will see if that helps.

I mean I do have 16GB of 2400MHZ DDR4 which I got for under £60,but Ryzen seems to love fast RAM,even more than Intel,so I am hoping RR is better in the game than Ryzen MK1 and is less dependent on fast RAM(although as SJ said the lack of cache might not help) - Intel would be better but hardly any site tests settlements,so I am not sure how much performance a Core i5 8400 will lose in the game when it comes to minimums.

If it were not for the RAM cartel fixing prices,I might have upgraded by now,but £200 for 16GB of 3000MHZ/3200MHZ RAM is tacking the piss.

TBH I am just getting annoyed with how much PC parts are just going up in general.

I still remember when a Xeon E3 4C/8T CPU could be had for as low as £170 and RAM was nowhere as expensive!!

[CAT-THE-FIFTH]

It looks like the Ryzen 5 2600 is only a 200MHZ clockspeed bump to the Ryzen 5 1600:

https://videocardz.com/74829/amd-ryz...i-hero-spotted

[scaryjim]

It looks like the Ryzen 5 2600 is only a 200MHZ clockspeed bump to the Ryzen 5 1600 ...

That wouldn't be entirely surprising for the 65W parts - the lower down the power stack you get the better the 14nm LPP process will perform. I think the place we'll see the big improvements are the X parts, particularly the 95W parts.

[Millennium]

Yep. Ryzen 2700 @ up to 4.4 ghz would be shattering, given the 8 core 16 thread accomplishment.

I wouldn't upgrade from 1700 @ 4.0 (top) but many would be swayed.

Still I think Intel has the IPC crown with their 8700k etc. But it doesn't matter long term given the GPU market and the value proposition and so on.,

[scaryjim]

... Ryzen 2700 @ up to 4.4 ghz would be shattering ...

1700 is a 65W part, so I'd expect the 2700 to only get the couple of hundred megahertz boost that the 2600 got.

If the 2700X/2800X (and indeed the 2600X) only get ~ 200MHz, I'll be somewhat disappointed. I mean, they'll still be good processors, but I'd hope for much nearer a 10% speed boost. Precision Boost 2 will add finer-grained clock control in situations where some but not all threads are loaded. IPC is unlikely to change, though, so they will still be behind Intel in pure per-core performance terms.

[DanceswithUnix]

I would expect them to get a 200MHz boost after a year if they stayed on the same 14nm process

[scaryjim]

I would expect them to get a 200MHz boost after a year if they stayed on the same 14nm process

Isn't this 12nm process actually just a development of 14nm anyway? I thought the gate sizes were basically the same?

My understanding was that the voltage profile was a lot more linear (where as the voltage for 14nm LPP starts ramping up exponentially when you go over 3GHz), so you won't hit the voltage wall as early. I still wouldn't expect much more than 10% at the top end, but that should give them up to 4.5GHz to play with on stock boosts.

The difference between 65W 1600 and 95W 1600X was only 400MHz anyway. They've closed half of that gap in the same 65W envelope. I don't think that's too shabby, personally...

[badass]

Isn't this 12nm process actually just a development of 14nm anyway? I thought the gate sizes were basically the same?

My understanding was that the voltage profile was a lot more linear (where as the voltage for 14nm LPP starts ramping up exponentially when you go over 3GHz), so you won't hit the voltage wall as early. I still wouldn't expect much more than 10% at the top end, but that should give them up to 4.5GHz to play with on stock boosts.

The difference between 65W 1600 and 95W 1600X was only 400MHz anyway. They've closed half of that gap in the same 65W envelope. I don't think that's too shabby, personally...

TBH if they could find a way of decoupling memory speed from their performance quite so much with some infinity fabric Tweaks, they would probably gain more performance than a 200 MHz clockspeed bump for users of all but the fastest RAM. However I suspect that we will have to wait for Zen 2 for that.

I'll be keeping a close eye on these. Although still very unlikely to upgrade my old i5-3570k as I can't justify the cost for the performance increase that I will usefully get.

[scaryjim]

... still very unlikely to upgrade my old i5-3570k as I can't justify the cost for the performance increase that I will usefully get.

Erm ... you do know that the Ryzen 3 1200 is roughly the same performance as your i5, yeah? That's the reason my step-son got a full rebuild this Christmas rather than a CPU upgrade. Sure, I could've dropped an i5 into his existing rig for ~ £50, but for longevity it would've made no sense. I suspect the step up to a 1500X or 1600X would be noticable, so definitely worth keeping an eye on how much the second gen bring to the table.

As to infinity fabric improvements, there were rumours of a double-pump setting that was disabled. I always wondered if it was because the 14nm process couldn't support the higher speed for it (at reasonable voltages, anyway), at which point 12nm might allow them to enable it again. For Zen 2, though, I'm absolutely sure they'll be seeking to speed up the fabric and bring down cross-CCX latency...

[CAT-THE-FIFTH]

If the whole series is only getting 200MHZ more with zero IPC improvements and no other performance tweaks,then assigning a whole new chipset to it seems rather pointless. That extra 200MHZ could be down to Ryzen+ just having better boosting mechanisms which the APU does have.

This is the level of another stepping or clock boosted SKU which AMD or Intel launches normally within a generation. I certainly hope that they can at least get 10% more performance on the top bin SKUs,otherwise calling it Ryzen "2000" series and launching a new 400 series chipset is like rebranding the Phenom II X4 980BE as the 1980BE and launching a new chipset.

I sincerely hope Ryzen+ has improved IPC and some fixes for the memory controller,etc to improve overall performance. If it is just a slightly clock boosted Ryzen with a few percent boost in gaming performance,unless my system can last until Ryzen 2,sadly it looks more and more likely I will have to go Intel,unless the Ryzen APUs can show better performance with older engines.

[Xlucine]

Nothing wrong with bringing out a new chipset with new features, especially since the old chips&chipsets are interchangeable. PC sellers like a regular cadence of releases for CPU and chipset, it's easier to shift new stuff

[CAT-THE-FIFTH]

Nothing wrong with bringing out a new chipset with new features, especially since the old chips&chipsets are interchangeable. PC sellers like a regular cadence of releases for CPU and chipset, it's easier to shift new stuff

The problem is AMD does this all the time. They get a good performance jump,and then take yonks to up core performance,ie,they stagnate and then Intel pushes more updates and they start falling behind.

Once Ryzen+ is out it will be probably another year until Ryzen 2.

There are leaks showing Intel is starting to release further CL models,with improved clockspeeds:

https://videocardz.com/74856/intel-c...ta-14-february

Its been a year,and 200MHZ or even 300MHZ is frankly meh if that is all they can muster,especially on a "tweaked" 14NM,ie,12NM and remember the 2000 series will have fixed core boost,so I expect that is partly to explain for the improved clockspeeds. The APUs can already achieve similar clockspeeds on "normal" 14NM.

So for two years,they will have basically improved clockspeeds by 200MHZ or a bit more.

If it is 200MHZ with tweaks to the memory controller,improved inter-CCX bandwidth,and some core tweaks to improve performance oddities,etc which bumps up performance by at least 10% then it might be something.

If not the chances of me getting Ryzen or Ryzen+ are getting fainter and fainter since it performs relatively poorly in games based on older engines and its doubtful AMD will bother pushing devs to improve in such games. Now I live in hope that the single CCX APUs will do better,but OFC no guarantee it will happen.

Remember,I asked AMD about older games,and I get the impression they will only care about newer games used in benchmark suites,but most of these games are not open world ones or the types of games you will put 100s of hours into and tend to not be benchmarked that much(or properly). They really need to understand some of these games are still immensely popular even now(WoW and the Bethesda openworld games,etc being examples) and these are the kind of games people will just buy Intel out of habit for. They CBA even to prod Bethesda who they have a PARTNERSHIP WITH,to patch their games(not the ones they publish) for Ryzen,which is utterly hilarious.

So that means AMD needs to brute force it,so that means me also living in hope for Ryzen 2. However,at that point Intel might have actually had a proper socket change and fixed the security on their CPUs,since they can go and do one with their current fudged CPUs.

I really,really,really want to get AMD for my next CPU upgrade,but sadly I know I will still end up with an Intel CPU despite that. I hope I am proved wrong in this case!!

[DanceswithUnix]

I really,really,really want to get AMD for my next CPU upgrade,but sadly I know I will still end up with an Intel CPU despite that. I hope I am proved wrong in this case!!

I'm not investing effort in tracking leaks of the performance of upcoming parts when Meltdown and Spectre are still somewhat up in the air so I'm not even sure of the performance of the 5 year old CPU I already own

You really don't get much out of process updates these days, the main advantage is more transistors allowing you to fit more functionality on so porting an existing design to 12nm won't get much. OFC AMD aren't that far behind Intel in lots of benchmarks so even another 5% will close a lot of gaps. Hopefully for them it will improve profitability as well.

Still, it is usual for a new line of CPUs to be released with bugs and problems and the second version to be rather sweet. Phenom was a dog of a product, Phenom II was pretty good. Bulldozer missed the mark, Piledriver was way better. That depends on how buggy the original Ryzen was, and how much low hanging fruit they know they can improve, but hopefully Ryzen+ will be interesting.

Leaks are fun, but it isn't over until the fat Ryzen+ Threadripper sings