[Various] AMD Ryzen Reviews (Official)

[Motiv]

isn't this the same problem Intel had with SMT in the beginning. Took them a year or two to sort out.

[StAndrew]

isn't this the same problem Intel had with SMT in the beginning. Took them a year or two to sort out.

I thought the issue with Intel's HT was the OS wasn't disseminating between HT and actual cores and double booking one core while another was dormant. The software fix to prioritize hardware cores over "HT" cores should work on AMD's hardware, no? It should be pretty easy to see if this is actually the case by observing the CPU utilization during light thread loads. If you observe a skewed load on one or two processors while the others are running at low utilization then software will probably need to be updated for AMD's SMT. Given the large disparity between different benchmarks, I have a feeling there is some optimization to be done on the software side to take advantage of AMD's architecture.

[Greg83]

http://hexus.net/tech/reviews/cpu/10...nm-zen/?page=9
Also all the AMD fanboys can now safely say bulldozer sucks.

the only reference to bulldoser well vishera in all the reviews i saw in gaming, was linus of techtips, and that shown instead, at 4k gaming 8350 and everyone else, are hardly any different at all. and no one would notice. even with its pci-e 2.0 bottleneck.

making me sure of my choice, in my form of light gaming , web browsing. With goal being only end of system bottlenecks and feeling the lag, never frame rates and bullet candy. playing no first person shooters, only 4x and rts games. naples is the only thing that is gonna impress me, unless maybe zen+ is 16 cores already in 2018-2019.
Not that the benchmarks do not impress me.
But i guess in real world. to me bulldoser 8 interger units and 4 floating point units, was a real 8 core in my usages.
I still find my 8120 faster then my 8320. When i have not used the AMD patch to "fix" the issues in certain software, by using the same fpu twice for 2 threads.

[Ket]

I wouldn't cancel my processor pre-order but I would cancel my moboard. I'm sure there will be much better offerings in the not too distant future. However, on the Anandtech review (IIRC) AMD already mentioned they are working on some short fall in the design and are coming out with a Re-Ryzen CPU... Yeah, I'll wait

Those improvements aren't exactly right around the corner, you won't see them until Zen+ which likely won't happen until mid-end 2018. Sure you can wait another 12-18 months?

[chew*]

While i ran prime 95 and watched a live stream, on skype i ran cinebench. Not only did it barely impact results. There was not one hiccup in the video stream. I was on a crappy netgear usb wireless stick.

Anyway thats something you have to experience in person to grasp. I think back when amd showed a gamer running a livestream on same system he was gaming on they were trying to show this.

I took it to extremes and tbh while it was running had i installed steam i probably would have been playing a game np whatsoever.

[KiSUAN]

I thought the issue with Intel's HT was the OS wasn't disseminating between HT and actual cores and double booking one core while another was dormant. The software fix to prioritize hardware cores over "HT" cores should work on AMD's hardware, no? It should be pretty easy to see if this is actually the case by observing the CPU utilization during light thread loads. If you observe a skewed load on one or two processors while the others are running at low utilization then software will probably need to be updated for AMD's SMT. Given the large disparity between different benchmarks, I have a feeling there is some optimization to be done on the software side to take advantage of AMD's architecture.

From https://www.reddit.com/r/Amd/comment...lv&sh=b217adad

Why there is huge discrepancy is gaming benchmarks for reviewers today? Is this something related to BIOS update?

Early motherboard BIOSes were certainly troubled: disabling unrelated features would turn off cores. Setting memory overclocks on some motherboards would disable boost. Some BIOS revisions would plain produce universally suppressed performance.

Ryzen benefits from disabling High Precision Event Timers (HPET). The timer resolution of HPET can cause an observer effect that can subtract performance. This is a BIOS option, or a function that can be disabled from the Windows command shell.

Ryzen benefits from enabling the High Performance power profile. This overrides core parking. Eventually we will have a driver that allows people to stay on balanced and disable core parking anyways. Gamers have been doing this for a while, too.

There are some games that are using code optimized for our competitor... we are confident that we can work through these issues with the game developers who are actively engaging with our engineering teams, this is strictly CPU scheduling within the game.

[StAndrew]

Those improvements aren't exactly right around the corner, you won't see them until Zen+ which likely won't happen until mid-end 2018. Sure you can wait another 12-18 months?

Haha, I JUST "upgraded" to a used i7 980x. If I buy a Ryzen, it will be the first new and up to date piece of hardware I've purchased since I bought a Conroe X6800 and some 9800 GTX's. From there it was an i7 975 and some GTX285's (all used). My sig is my latest "update." I can wait for 12-18 months

[AliG]

Haha, I JUST "upgraded" to a used i7 980x. If I buy a Ryzen, it will be the first new and up to date piece of hardware I've purchased since I bought a Conroe X6800 and some 9800 GTX's. From there it was an i7 975 and some GTX285's (all used). My sig is my latest "update." I can wait for 12-18 months

Haha I was running an AM2 Athlon x2 4000 for years before I finally succumbed and bought a 2500k. Ran on a ATI 4850 for years as well until I "upgraded" to a used R9 280X LOL.

Thankfully, becoming an "adult" and having a "job" with "responsibility" has made it easier to not worry about playing the latest games .

[NEOAethyr]

In the review Linus did he said he couldn't get past 2666:
https://www.youtube.com/watch?v=9wJQEHNYE7M

I thought the review on Paul's hardware was decent:
https://www.youtube.com/watch?v=OXbOC_OyvG4


I can't load some of the pages from this review:
http://www.phoronix.com/scan.php?pag...0x-linux&num=1

However I can get access to some of the info from here:
http://openbenchmarking.org/system/1...en%207%201800X
Not much really...

In the forums there's some talk about smt and bios issues.
https://www.phoronix.com/forums/foru...nux-benchmarks

The gigabyte auros or whatever it's called supposedly the acpi tables are causing linux not to boot.
Also the audio chip isn't supported in the current repo'd kernel, you have to build your own.
It's in v4.11, just not in the repo yet, I think ubuntu variants of 17.04 get 4.10 if I remember right.
Just so people know when they goto plug in there usb stick or dvd with linux on it.
It can be done, however it will be a little easier in a month or so.
I haven't checked the repo in a while, I haven't updated to 4.10 yet becauase I didn't feel like setting up grub again and adding a patch for exfat to build with it (I'm just lazy).
(Hmm I'm not seeing 4.10 in my repo..., I'm on 14.04 though)

And as for smt...
It could be scheduling software, or the sse/avx implementation maybe that accounts for the gaming perf.
That's what I'm thinking.

Edit:
Oh and no sensors in aida I heard, and nor linux yet.

One more edit:
Huh?!
http://openbenchmarking.org/system/1...le_frequencies
This can't be right... (There's only 3 base freq steppings...)

[BenchZowner]

From https://www.reddit.com/r/Amd/comment...lv&sh=b217adad

HPET is disabled in Win10 by default...

[NEOAethyr]

Some random bcdedit cmd's for win7-win10'ish.

Helps for win8-10 for getting more consistent dpc latency, there's a similar kernel setup in linux called no hz (ie tickless).
bcdedit /set {current} disabledynamictick yes

You probably want a tick, a constant clock...

This enabled hpet, if you wanted to force it disabled then it would be false, or off possibly might work too.
bcdedit /set {current} useplatformclock true
I recommend it ON..., hpet is a fast timer... ^^, it's called high perf for a reason.. (Edit: duh it stands for precision not perf lol...)
It will get you lower dpc latency with it ON... ^^

Never execute blah blah
bcdedit /set {current} nx AlwaysOff

PAE 48bit addressing...
bcdedit /set {current} pae ForceEnable

This is the std for userspace/kernal space (ignore the kernel warning crap from the msdn)
bcdedit /set {current} IncreaseUserVa 4096
This is if you had 8gig in your system and you want your app to beable to use all of it:
bcdedit /set {current} IncreaseUserVa 8192
However there's also limits on apps, large address aware I think, it needs to be enabled in the admin security policies or something I forget (Edit: this one was lock memory pages I think), and lots of apps need patching blah blah (Edit: lol there could be a patch for this for windows it's self in theory since there is a patch for wine...).
Windows is kinda lame in that respect.
Lol, just ignore what I just posted lol, seriously... :\

Some other options that don't actually work I think lol (It'll say success but not have any effect):
bcdedit /set {current} loadoptions DDISABLE_INTEGRITY_CHECKS
bcdedit /set {current} nointegritychecks yes

Other misc ones:
enable avx
bcdedit /set {current} xsavedisable 0

disable avx
bcdedit /set xsavedisable 1

...
bcdedit /set {current} kernel ntkr128g.exe

bcdedit /set {current} quietboot on
bcdedit /set {current} quietboot off

bcdedit /set {current} bootmenupolicy legacy
bcdedit /set {current} bootmenupolicy standard


Anyways to the point...
Hpet should be on.
Core parking should be off.
If you want to take a 50% hit on mt and want a measly 5% improvement in gaming perf, disable smt.

In the power management settings in windows.
Set the min perf lv to 0%, and max to 100% and apply that, use a custom power profile baised on perf.
Windows even has this prob on my board/cpu, without changing power management for min and max cpu, I get a huge lose in perf in some areas.

Also, if your seriously gonna use win10 ^^, disable dynamic tick, it does help a little, it's just that you'll never get the low dpc latency of win7, and in that sense, win7 will never get the dpc latency of winxp... ^^

Some people are so confused over hpet, there's whole threads on why you should disable it lmao.
It's a faster tick then any other timer in the system lol.

Edit:
Looks like TastyPC is making vids again, here's another video for you guys :
https://www.youtube.com/watch?v=2brYnTGTSR8

[Andi64]

There is a Patch for Wine because Wine is an aplication. And you are recommending to mess with Windows memory management without knowing how it works (that as far as I know will apply only on 32bit aplications based on your recommendations). If I'm not mistaken, Windows has 8TB (terabytes) of user Virtual Address Space on Windows 64b.

If you have 8GB RAM and set the User Space or Virtual Address Space or whatever you wanna call it to 8GB, how much RAM do you reserve for Kernel space? Anyway, it is impossible to increase it to 8GB. "IncreaseUserVa" is a Windows 32b only value, and can be between 2GB and 3GB MAX (and will set 1GB remaining for Kernel space).

PAE on 64-bit Windows does not exist. Period.

Please, try to avoid throwing numbers and recommendations that you dont understand and can cause problems to an unexperienced user.

[StyM]

[StyM]

[NEOAethyr]

There is a Patch for Wine because Wine is an aplication. And you are recommending to mess with Windows memory management without knowing how it works (that as far as I know will apply only on 32bit aplications based on your recommendations). If I'm not mistaken, Windows has 8TB (terabytes) of user Virtual Address Space on Windows 64b.

If you have 8GB RAM and set the User Space or Virtual Address Space or whatever you wanna call it to 8GB, how much RAM do you reserve for Kernel space? Anyway, it is impossible to increase it to 8GB. "IncreaseUserVa" is a Windows 32b only value, and can be between 2GB and 3GB MAX (and will set 1GB remaining for Kernel space).

PAE on 64-bit Windows does not exist. Period.

Please, try to avoid throwing numbers and recommendations that you dont understand and can cause problems to an unexperienced user.

Patch wine, is not an app, it's the HAL... (Edit: sorry it was the kernel: https://bugs.winehq.org/attachment.c...56&action=edit )
And for 2 I told you to ignore that crap I posted...
For 3, it works on 64bit just as well as 32bit it's only 2 gb diff in config for user space.
And for 4, again told ya to ignore that section, it was mainly about hpet and dynamic tick since people starting bringing it up.

;; Edit:
There are 32bit apps that you run on 64bit windows you know right?
Directshow maybe... ^^ (not everything is 64bit good to go...)
And you do realize that there's tons of apps that don't use squat for ram and keep swapping for no reason?
This helped with my final decision of swap..., why bother swapping out when I'm just gonna read it from the drive again anyways? (Other then having low amounts of physical ram, like a vm and compiling with it or video editing or something)
Also, pae does too exist on 64bit lol, otherwise you wouldn't beable to access more then 3.5/4gb, it's the same exact thing that 32bit os'es "can" use to get access to all that ram...



I'm am so ticked off at amd.
I made a fool of myself promoting that pos.
There should class action lawsuit, for them lying about the l2 and l3 cache sizes, and for not giving us the spec for the interconnect, the darned chip is out and those specs are still unavail.
I'm so pissed off, I wanted this to work, but amd just had to design another freaking fail arc.
Now I gotta make a freaking phone call and warn people about this garbage .
Smt fail, not just in the normal way, but in a colossal way.
I wouldn't buy this crap for $100 now, I wouldn't even get it for my brother that's on a 1090t.
Why even bother with workstation cpu when it can't handle workstation loads (64 f'ing bit mother *...)

That's my 2 cents, and it's the last time I'm gonna post about this pos.
Amd, shove it.

Edit:
Sorry guys for going off the deep end for a bit.
Just I was up all night with a buddy discussing this.
I'm just disappointment that's all, I never expected a separate l3 cache on an interconnect fabric thing, with no bit rate specs.
Supposedly it's locked to 1/2 of the mem clock.
I knew avx might not be as good as intel, but this is just something else.
Amd will have to invest money on scheduler updates to keep code running in a single ccx.
And maybe invest code in general scheduler improvements as well.

This modular design was the cheap way to go in my opinion, great for business maybe but I think it was a mistake and after all this time..., that's why I went off an a rant :\.

[Monstru]

Ok, Ok, I know... the language is Klingon...but the graphs are universal and with english subtitles... so...

7350K vs 7600K vs 7700K vs 6800K vs 6900K vs 6950X vs 1800X


7700K vs 6800K vs 6900K vs 1700 vs 1700X vs 1800X vs 1700 OC'ed + Gaming comparison in UHS, WQHD and Full HD

[chispy]

Ok, Ok, I know... the language is Klingon...but the graphs are universal and with english subtitles... so...

7350K vs 7600K vs 7700K vs 6800K vs 6900K vs 6950X vs 1800X


7700K vs 6800K vs 6900K vs 1700 vs 1700X vs 1800X vs 1700 OC'ed + Gaming comparison in UHS, WQHD and Full HD

Nice , well written articles Monstru. Great to see you back

[JAWS]

Maybe it will help me feel better about not being able to upgrade right away anyway lol

I feel the same way. Leave it to AMD to release a brand new product (that competes) when I have no money.

[techjesse]

AMD is back Now chew can start benching YEAH!

[Monstru]

Nice , well written articles Monstru. Great to see you back

Thx brother!

[alpha0ne]

Ok, Ok, I know... the language is Klingon...but the graphs are universal and with english subtitles... so...

Thanks Monstru for the Klingon Ryzen review and especially for Adobe Photoshop CC 2017 which is the only bench I'm interested in these days

[StyM]

[StyM]

Premiere Pro CC 2017 AMD Ryzen 7 1700X & 1800X Performance

[Monstru]

Thanks Monstru for the Klingon Ryzen review and especially for Adobe Photoshop CC 2017 which is the only bench I'm interested in these days

Thx man!

[chew*]

AMD is back Now chew can start benching YEAH!

Is it me or do her boo err I mean shirt say chew* forever I did not photo shop that lol...

After a learning curve and some preparations I think we will be seeing some of this again