PDA

View Full Version : 5% to 30% performance hit on cpu renderering



samurai_x
01-03-2018, 03:58 AM
5% to 30% performance hit for computers built within a decade.
'Kernel memory leaking' Intel processor design flaw forces Linux, Windows redesign
Other OSes will need an update, performance hits loom.

https://www.theregister.co.uk/2018/01/02/intel_cpu_design_flaw/

Gpu all the way.

TheLexx
01-03-2018, 05:34 AM
I believe AMD is unaffected, so Threadripper seems a reasonable option for the next PC build... :)

MichaelT
01-03-2018, 05:49 AM
5% to 30% performance hit for computers built within a decade.
'Kernel memory leaking' Intel processor design flaw forces Linux, Windows redesign
Other OSes will need an update, performance hits loom.

https://www.theregister.co.uk/2018/01/02/intel_cpu_design_flaw/

Gpu all the way.

Also, the guy who created minix wrote an open letter to Intel some time ago: http://www.cs.vu.nl/~ast/intel

MichaelT
01-03-2018, 05:54 AM
I believe AMD is unaffected, so Threadripper seems a reasonable option for the next PC build... :)

Wonder what would happen if companies etc.. started demanding Intel to replace these faulty chips. Or even a class action lawsuit. Could get nasty.

rustythe1
01-03-2018, 06:43 AM
yes, for companies like amazon and google a 30% drop could represent billions of business loss, so we get lw2018 and then straight after we get a 30% speed drop, cool, and it took 10 years to spot it?

samurai_x
01-03-2018, 07:03 AM
yes, for companies like amazon and google a 30% drop could represent billions of business loss, so we get lw2018 and then straight after we get a 30% speed drop, cool, and it took 10 years to spot it?

Actually haven't people noticed there's some speed drop with lw 2018's renderer? Going bidirectional pathtracing on cpu would always be slower.
I was using indigo renderer before, which is not so different to lw 2018 renderer, but it supported cpu and gpu . The difference using cpu and gpu was enormous in indigo.

MichaelT
01-03-2018, 07:17 AM
yes, for companies like amazon and google a 30% drop could represent billions of business loss, so we get lw2018 and then straight after we get a 30% speed drop, cool, and it took 10 years to spot it?

And it is a very noticeable drop.. imagine a 3ghz cpu performing like a 2ghz one. Imagine the sh*tstorm that will come when those patches are out. Now I am especially happy I have disabled Microsofts auto updates.

rustythe1
01-03-2018, 08:43 AM
Actually haven't people noticed there's some speed drop with lw 2018's renderer? Going bidirectional pathtracing on cpu would always be slower.
I was using indigo renderer before, which is not so different to lw 2018 renderer, but it supported cpu and gpu . The difference using cpu and gpu was enormous in indigo.

actually with the scenes i was working with just before i got 2018, they are much faster now, for internal renders i think it will be slower, but for external, volumetric and so on once you start tweaking the settings its far faster, and ive not even dropped the settings and started using the denoiser yet,

Mastoy
01-03-2018, 08:44 AM
...

samurai_x
01-03-2018, 09:53 AM
actually with the scenes i was working with just before i got 2018, they are much faster now, for internal renders i think it will be slower, but for external, volumetric and so on once you start tweaking the settings its far faster, and ive not even dropped the settings and started using the denoiser yet,

Good to know. Need further testing.
For interiors, probably the same. Lightwave 2015 was never fast with interiors. I have kray for that.
Volumetric - lw 2018 is faster than lw 2015. Faster than redshift, octane, cycles using a sample vdb? Not tested but I'm sure gpu will be faster.
Exteriors - Is it faster than dpsun+backdrop radiosity? Need to test. Exteriors aren't really problematic with most render engines. Even octane is fast for exteriors.

jwiede
01-03-2018, 11:35 AM
Now I am especially happy I have disabled Microsofts auto updates.

So just to be sure I understand your position: You'd rather continue to operate with the serious security hole in question present (plus all the others you've missed), rather than take a patch that closes the hole (but incurs perf. cost)?

Good luck with that! :rolleyes:

Qexit
01-03-2018, 11:45 AM
So just to be sure I understand your position: You'd rather continue to operate with the serious security hole in question present (plus all the others you've missed), rather than take a patch that closes the hole (but incurs perf. cost)?

Good luck with that! :rolleyes:Microsoft automatic updates are a sore point for me right now after spending around 7 hours yesterday sorting out the mess the latest Win 10 patch caused to my Lightwave PC :devil: I do allow the updates as the security aspect does concern me but when it results in my PC being unusable I have serious second thoughts about it. Yesterday's fun and games were only resolved after swapping in an ancient nvidia graphics card, attaching it to an equally old Dell 4:3 monitor, then swapping in a more recent 1050ti graphics card, then reconnecting my regular monitor and finally reinstating my Quadro graphics card. Oh, and several graphics driver install/unistall/resinstall cycles plus a touch of CCleaner tidying up :) All this because for some reason the latest Win 10 update decided it no longer recognised my monitor so kept switching to an incompatible signal which my monitor could not recognise.

Sekhar
01-03-2018, 11:46 AM
So just to be sure I understand your position: You'd rather continue to operate with the serious security hole in question present (plus all the others you've missed), rather than take a patch that closes the hole (but incurs perf. cost)?

Good luck with that! :rolleyes:

Actually, it might be a good choice because the hole seems to be not too easy to exploit and if you take care to run only trusted programs (i.e., have virus and malware protection on), the chances of being affected may to too low to be worth taking a steady performance hit across all apps.

rustythe1
01-03-2018, 12:30 PM
So just to be sure I understand your position: You'd rather continue to operate with the serious security hole in question present (plus all the others you've missed), rather than take a patch that closes the hole (but incurs perf. cost)?

Good luck with that! :rolleyes:

but at the same time, the hole has been there for 10 years

MichaelT
01-03-2018, 12:56 PM
So just to be sure I understand your position: You'd rather continue to operate with the serious security hole in question present (plus all the others you've missed), rather than take a patch that closes the hole (but incurs perf. cost)?

Good luck with that! :rolleyes:

You're assuming that some nefarious person have physical access to my computer. And even if they (somehow) manage to attack it remotely, I still would avoid a patch that literally caps the cpu performance at the kneecaps. I'd much rather prefer if companies like Microsoft et.al. forced Intel to face up, and replace the cpus. Like if that would ever happen. But yeah... when performance is capped that badly... yeah... I'll pass on that patch. I'd like performance to stay in 2018 thank you very much... not in 2006. I suddenly see AMD stock rising on the horizon.

jwiede
01-03-2018, 08:15 PM
but at the same time, the hole has been there for 10 years

Except that now it's public, and the mechanism is not difficult to discern from the patches and descriptions given (and it'll only get more public from here).

Jarno
01-03-2018, 08:45 PM
The performance hits will be in doing system calls (basically, calling stuff in the kernel of the operating system). That basically doesn't happen during rendering, which is just pure number crunching. So at the moment I'm not expecting to see any significant degradation in render speed.

We will keep an eye on the benchmarks to see if something unexpected pops up.

---JvdL---

OFF
01-03-2018, 08:46 PM
Good to know. Need further testing.
For interiors, probably the same. Lightwave 2015 was never fast with interiors. I have kray for that.
Volumetric - lw 2018 is faster than lw 2015. Faster than redshift, octane, cycles using a sample vdb? Not tested but I'm sure gpu will be faster.
Exteriors - Is it faster than dpsun+backdrop radiosity? Need to test. Exteriors aren't really problematic with most render engines. Even octane is fast for exteriors.

Not faster - but now can give the quality of the picture at the level of the best renderers. Unfortunately - the render speed is not enought high.

samurai_x
01-04-2018, 09:35 AM
So amd cpus are affected as well.
https://techcrunch.com/2018/01/03/kernel-panic-what-are-meltdown-and-spectre-the-bugs-affecting-nearly-every-computer-and-device/

MichaelT
01-04-2018, 11:25 AM
To a lesser degree yes (according to the one I heard it from at least) but it won't matter, as I very much doubt the updates from Linux, Microsoft etc.. will make any distinction. As their primary goal is to close the hole. But I won't install anything at all, until I see how this goes.

MichaelT
01-06-2018, 12:53 PM
This is the list of affected Intel CPUs. As for the other brands, you need to find out for yourselves.
I'm guessing that the older the CPU (or lower spec) the worse that performance cap is going to hit you.
If you never load your CPU you might not even notice, but for those who do (and do the kernel talk) most likely will see a big difference.
I still want to see some ability to disable that fix... because this is potentially so damaging to performance.. I'd like to see some options here.

List:

Intel Core i3 processor (45nm and 32nm)
Intel Core i5 processor (45nm and 32nm)
Intel Core i7 processor (45nm and 32nm)
Intel Core M processor family (45nm and 32nm)
2nd generation Intel Core processors
3rd generation Intel Core processors
4th generation Intel Core processors
5th generation Intel Core processors
6th generation Intel Core processors
7th generation Intel Core processors
8th generation Intel Core processors
Intel Core X-series Processor Family for Intel X99 platforms
Intel Core X-series Processor Family for Intel X299 platforms
Intel Xeon processor 3400 series
Intel Xeon processor 3600 series
Intel Xeon processor 5500 series
Intel Xeon processor 5600 series
Intel Xeon processor 6500 series
Intel Xeon processor 7500 series
Intel Xeon Processor E3 Family
Intel Xeon Processor E3 v2 Family
Intel Xeon Processor E3 v3 Family
Intel Xeon Processor E3 v4 Family
Intel Xeon Processor E3 v5 Family
Intel Xeon Processor E3 v6 Family
Intel Xeon Processor E5 Family
Intel Xeon Processor E5 v2 Family
Intel Xeon Processor E5 v3 Family
Intel Xeon Processor E5 v4 Family
Intel Xeon Processor E7 Family
Intel Xeon Processor E7 v2 Family
Intel Xeon Processor E7 v3 Family
Intel Xeon Processor E7 v4 Family
Intel Xeon Processor Scalable Family
Intel Xeon Phi Processor 3200, 5200, 7200 Series
Intel Atom Processor C Series
Intel Atom Processor E Series
Intel Atom Processor A Series
Intel Atom Processor x3 Series
Intel Atom Processor Z Series
Intel Celeron Processor J Series
Intel Celeron Processor N Series
Intel Pentium Processor J Series
Intel Pentium Processor N Series

Amerelium
01-06-2018, 01:33 PM
Don't understand why they published this in the first place. This is an obscure exploit that no one knew about 6 months ago; now everyone does. Expect attacks to follow.

In any event, I'm not sweating it. I'm not gonna do any of the potentially slowing updates. Keep the system clean and only online when needed, and you're in the clear. My really important stuff is in any event dependent on security items not connected to anything relevant

MichaelT
01-06-2018, 02:05 PM
Obscure or not... I find that all IHVs share the same bug.. very interesting. .. in any case. This fix should not be obligatory... it is far to damaging for that.

Qexit
01-06-2018, 02:27 PM
Don't understand why they published this in the first place. This is an obscure exploit that no one knew about 6 months ago; now everyone does. Expect attacks to follow.

In any event, I'm not sweating it. I'm not gonna do any of the potentially slowing updates. Keep the system clean and only online when needed, and you're in the clear. My really important stuff is in any event dependent on security items not connected to anything relevantWell, the information about the exploit was first published by a well-known hacker. That was picked up by The Register who published an article which got widely distributed, so everyone else had little choice but to put their side of the story (so to speak). At least that is the gist of how things seem to have gone from what I have read on the subject.

Amerelium
01-06-2018, 06:38 PM
...methinks it's all being blown a bit out of proportion though; as I understand it it takes quite a bit of work and access to actually make use of the exploit - unless you're big or famous you're a highly unlikely target for it. I'm certainly not going to compromise speed for a less-likely-that-being-attacked-by-shark-whilst-struck-by-lightning scenario. I can see bitcoin hubs, Trump and Big Evil Corporation being targets for something like this, not you or me. No sweat.

MichaelT
01-06-2018, 07:17 PM
...methinks it's all being blown a bit out of proportion though; as I understand it it takes quite a bit of work and access to actually make use of the exploit - unless you're big or famous you're a highly unlikely target for it. I'm certainly not going to compromise speed for a less-likely-that-being-attacked-by-shark-whilst-struck-by-lightning scenario. I can see bitcoin hubs, Trump and Big Evil Corporation being targets for something like this, not you or me. No sweat.

The problem isn't the potential target. The problem are the fixes... it is those who drag down the performance by as much as 30% (lower cpu performance)

jwiede
01-06-2018, 08:22 PM
Obscure or not... I find that all IHVs share the same bug

If that's your interpretation of the multiple issues now being discussed, you've misunderstood some rather significant details in your study of it.

- - - Updated - - -


So amd cpus are affected as well.
https://techcrunch.com/2018/01/03/kernel-panic-what-are-meltdown-and-spectre-the-bugs-affecting-nearly-every-computer-and-device/

You're conflating multiple issues together as if they're all the same, and they are not.

samurai_x
01-06-2018, 08:38 PM
If that's your interpretation of the multiple issues now being discussed, you've misunderstood some rather significant details in your study of it.

- - - Updated - - -



You're conflating multiple issues together as if they're all the same, and they are not.

I didn't say by the same thing. That's why the link is there. Lol.

MichaelT
01-07-2018, 03:11 AM
If that's your interpretation of the multiple issues now being discussed, you've misunderstood some rather significant details in your study of it.

- - - Updated - - -



You're conflating multiple issues together as if they're all the same, and they are not.

When a patch applied covers more than one brand of CPU.. it is at the very least similar. In any case.. I have only what's out there to go on :) If that is wrong or incorrect, I am always willing to change. My main point however is still the rather significant drop in performance.

Amerelium
01-07-2018, 04:15 AM
The problem isn't the potential target. The problem are the fixes... it is those who drag down the performance by as much as 30% (lower cpu performance)

Which is why I'm not gonna do any software based fixes. If they redesign all future chips so that they work slower; well we cannot do much about that - I'll fall off that cliff when we get to it.

MichaelT
01-07-2018, 08:14 AM
I honestly think they should focus more on securing communication.. because let's face it.. this is the pipe all those attacks go through.

Cageman
01-07-2018, 03:49 PM
5% to 30% performance hit for computers built within a decade.
'Kernel memory leaking' Intel processor design flaw forces Linux, Windows redesign
Other OSes will need an update, performance hits loom.

https://www.theregister.co.uk/2018/01/02/intel_cpu_design_flaw/

Gpu all the way.

One of the most ill-informed and stupid topics I've ever read on this forum.

This is about the kernels of the Operating systems; not about Render engines or any software running ontop of these Operating systems...

Educate yourself before creating a "drama" thread.

samurai_x
01-07-2018, 07:41 PM
One of the most ill-informed and stupid topics I've ever read on this forum.

This is about the kernels of the Operating systems; not about Render engines or any software running ontop of these Operating systems...

Educate yourself before creating a "drama" thread.

The news says overall performance.
Why don't you stick with your fabulous joystick renders. Lol!

MichaelT
01-08-2018, 01:15 AM
Very "ill-informed":

139287
(Fortnight servers. Which mostly only crunch numbers btw.)

We know it is about kernels. I think you underestimate the impact.
And it is not about rendering engines either. It is about a patch that can have a very significant effect on the overall performance on your computer.

And I can think of one thing that affects performance.. since we are talking about kernel operations.... DRM.

samurai_x
01-08-2018, 01:27 AM
That's from Epic games right?
20 percent increase in CPU utilization after installing the patch.

MichaelT
01-08-2018, 01:30 AM
That's from Epic games right?
20 percent increase in CPU utilization after installing the patch.

Yup. Fortnight servers.

Lito
01-08-2018, 11:29 AM
I got the patch on Saturday evening for Win 10 and and probably most Win 10 users have it by now. Well I can't give any LW2018 benchmark figures on the patch because I didn't write down any prepatch numbers. But blender wise the difference is negligible for the gooseberry one on my system. My old Gooseberry benchmark was 22m 10.19s with my AMD 1950x OCed to 4.0Ghz pre patch, the current one post patch is 22m 47.13s but that is also with a drop in my OC setting to 3.8Ghz (I lowered the OC because the CPU ran much hotter given the amount of performance an extra 200mhz offered). So I am guessing that the patch on the CPU render side is going to only really cost you a lot of performance if the renderer has to constantly ask for or swap out things from memory during rendering. So low memory situations, or multitasking will lower performance much more than pre-patch but for anything that basically fits easily into memory the amount of extra processing is negligible. Anything constantly allocating/deallocating memory will probably hit that higher level performance penalty percentage.

Qexit
01-08-2018, 02:16 PM
I got the patch on Saturday evening for Win 10 and and probably most Win 10 users have it by now. Well I can't give any LW2018 benchmark figures on the patch because I didn't write down any prepatch numbers. But blender wise the difference is negligible for the gooseberry one on my system. My old Gooseberry benchmark was 22m 10.19s with my AMD 1950x OCed to 4.0Ghz pre patch, the current one post patch is 22m 47.13s but that is also with a drop in my OC setting to 3.8Ghz (I lowered the OC because the CPU ran much hotter given the amount of performance an extra 200mhz offered). So I am guessing that the patch on the CPU render side is going to only really cost you a lot of performance if the renderer has to constantly ask for or swap out things from memory during rendering. So low memory situations, or multitasking will lower performance much more than pre-patch but for anything that basically fits easily into memory the amount of extra processing is negligible. Anything constantly allocating/deallocating memory will probably hit that higher level performance penalty percentage.Just out of interest/curiosity, what O/S version and build number are you now running in Wondows 10 ? I did get a patch on my LW PC last week that caused me a whole load of problems but nothing since then. My LW PC is Windows 10 Pro while my internet and other stuff PC is Windows 10 Home. Both are showing:

Version 1709
Build 16299.125

MichaelT
01-08-2018, 02:29 PM
I got the patch on Saturday evening for Win 10 and and probably most Win 10 users have it by now. Well I can't give any LW2018 benchmark figures on the patch because I didn't write down any prepatch numbers. But blender wise the difference is negligible for the gooseberry one on my system. My old Gooseberry benchmark was 22m 10.19s with my AMD 1950x OCed to 4.0Ghz pre patch, the current one post patch is 22m 47.13s but that is also with a drop in my OC setting to 3.8Ghz (I lowered the OC because the CPU ran much hotter given the amount of performance an extra 200mhz offered). So I am guessing that the patch on the CPU render side is going to only really cost you a lot of performance if the renderer has to constantly ask for or swap out things from memory during rendering. So low memory situations, or multitasking will lower performance much more than pre-patch but for anything that basically fits easily into memory the amount of extra processing is negligible. Anything constantly allocating/deallocating memory will probably hit that higher level performance penalty percentage.

:) Good good .. happy to hear that. It will be interesting to see more reports from people. I've heard good ones (like yours) and awful ones.. like older AMD chips ending up in more or less locked state. So, we'll see.

nvme is hit quite badly it seems... I see as much as 40% reduction in performance (mainly very small files), SSDs are affected but not as badly.. So disk activity indeed seems to be affected the most (so far)
Games are affected, but not very much.

Personally I hope the disk thing is addressed.. because I will be affected by this.

Lito
01-08-2018, 05:32 PM
The patch brought me up to 1709 build 16299.192.

Qexit
01-10-2018, 12:49 PM
The patch brought me up to 1709 build 16299.192.Dang, just left this PC for a while to have my evening meal...and now it is on Build 16299.192 as well. Oh well :beta::D

MichaelT
01-10-2018, 12:59 PM
Well, at least you're not one of the "lucky" ones, who got their computer bricked by this update (AMD chips.. older ones) because Microsoft apparently disabled update for those computers.
I blocked updates.. because I'm terrified of the nvme drives being hit so badly with up to 40% drop in performance. With that, my music program would tank. It would be impossible to write music as it streams off the drive.
Don't know if they'll fix that any time soon either :(

Edit: Seems to be different brands mostly. I might get away with only a few % at worst. So that is some relief.

I so do hate this forced patching nonsense. It really must stop.

Anyway.. as it looks now.. I'll probably upgrade.

rustythe1
01-13-2018, 05:20 AM
just to point out, i had this patch install about a week ago, no it dosnt affect things like rendering, however it has broken a lot of software, specifically things that monitor or scan your computer, Spacesniffer, Aisuite, that sort of thing, although these are only the first ones ive seen,

Amerelium
01-13-2018, 02:49 PM
Well, at least you're not one of the "lucky" ones, who got their computer bricked by this update (AMD chips.. older ones) because Microsoft apparently disabled update for those computers.
I blocked updates.. because I'm terrified of the nvme drives being hit so badly with up to 40% drop in performance. With that, my music program would tank. It would be impossible to write music as it streams off the drive.
Don't know if they'll fix that any time soon either :(

Edit: Seems to be different brands mostly. I might get away with only a few % at worst. So that is some relief.

I so do hate this forced patching nonsense. It really must stop.

Anyway.. as it looks now.. I'll probably upgrade.

You don't really need to - there's a snowballs chance in hell of you falling victim from an attack like this.

MichaelT
01-14-2018, 03:30 AM
Technically no, don't *need* to :) But if I plan on getting any other upgrades I must.

genesis1
01-15-2018, 02:21 AM
A PC engineer was talking online about this and he says for home users the effect will be minimal. He says its more servers for big corporations etc that will see a hit on performance not the home user. If you think about it, its not exactly new. Patches for performance and security have been going on for years but the press dramatise everything and blow it out of proportion. Anyone remember the year 2000 computer scare of every computer going into meltdown? Nothing happened of course. I would say just chill and carry on doing what you do best with LW ;-)