Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
CoolCab
Apr 17, 2005

glem
my e cores story is a friend of mine who is honestly pretty technical finally replaced his dinosaur rig this year and moved up to i think a 12700k or something, real beast of a machine. and he was using it for months and was ecstatic with the performance and routinely would marvel at what a CPU monster it was. then someone taught me the copy and paste in task manager trick and i was showing it off and he pasted his and as near as we can tell he had accidentally turned the e cores off at some point during setup and never noticed.

to be totally fair that's definitely more the generational improvement than anything else but it made me lol.

Adbot
ADBOT LOVES YOU

BlankSystemDaemon
Mar 13, 2009



Paul MaudDib posted:

hot take: what if you put big/little on the server platform? even in, eg, web services, you've still got threads and functions that are "hotspots" and could benefit from being moved to a high-performance core. Something like zen3 (in the epyc clock ranges) is fine as a little core, but, instead of a zen3 ccx you could roughly have 3-4 intel golden cove cores, ish, iirc. Obviously, again, there are a lot of things where twice the cores at 70% the potency really adds up, but, you'd also have some fat cores to deal with hotspots.

you could of course do things like launching particular threads or types of delegates onto particular threadpools, but, you could also do it at a method level and annotate particular methods as being "hotspots" and count time spent inside those, or similar, and preferentially allocate those threads onto the p-cores. I bet you could squeeze some extra architectural PPA with either heuristics or annotation like that.

yeah lol at the support involved, but, moving a couple big enterprise applications to it might actually get a lot of bang. There's a lot of applications where java or rdbms etc are the particular bottleneck and you really wish you could run that one part faster.
The problem is, if you're doing CPU heavy workloads on a server, it's almost certainly either multithreaded or it benefits heavily from CPU caches (or, more likely, both), meaning you can't move the threads around as that causes the high-performance cache-line to be invalidated.

Also, modifications to the scheduler to add the heuristics to properly take advantage of the energy savings of using heterogenus multi-processing would add many thousands of lines of code to the hot path in the scheduler, likely offsetting any benefit gained in terms of added cputime used by the server during execution.

mdxi
Mar 13, 2006

to JERK OFF is to be close to GOD... only with SPURTING

BlankSystemDaemon posted:

The problem is, if you're doing CPU heavy workloads on a server, it's almost certainly either multithreaded or it benefits heavily from CPU caches (or, more likely, both), meaning you can't move the threads around as that causes the high-performance cache-line to be invalidated.

Also, modifications to the scheduler to add the heuristics to properly take advantage of the energy savings of using heterogenus multi-processing would add many thousands of lines of code to the hot path in the scheduler, likely offsetting any benefit gained in terms of added cputime used by the server during execution.

And if you're edging into HPC (even the weird, DIY-grade pseudo-HPC that I do) you very quickly learn to take direct control of things in a lot of ways that are contrary to general, modern OS use.

  • Don't even get the scheduler involved so far as your application is concerned. The rest of this list is mostly how you do that.
  • Is your application not FPU-bound? Then restrict it to running exactly as many instances as you have threads of execution.
  • Is it FPU-bound (AKA AVX)? Now you're down to running as many copies as you have physical cores
  • Oh, but is it multi-core aware within a single process? Then your work-queue manager needs to consider that and only run as many concurrent processes as can be supported on the machine
  • Don't enable swap at all. It's better to have a process OOM-killed and restarted from a checkpoint than to have many processes grind to a halt due to paging
  • Do you have a program which actively uses L3 rather than having things passively end up there due to management by the CPU/OS? Better know how much that program wants, and don't oversubscribe your cache either, because that gets nasty real fast (looking at you, Rosetta suite, wanting 4MB of L3 and running 1/3 as fast when you can't have it all to yourself)
  • Are you also doing GPU compute? Remember to leave a thread available for the process which is feeding the GPU

So yeah, heavy duty compute is a fundamentally different regime than human-scale, user-task oriented compute is. I believe heterogenous systems are fantastic for computing machinery which will be used by people. I also believe it'll be a minute before we work out how to make them a good and efficient fit for more backend things, even if we ignore HPC.

As a much more concise example, think about trying to set up meaningful auto-scaling rules for a K8S pod whose CPUs are not homogenous.

AMD's answer at the moment is to make Epycs in two variants: CPUs with full-fat cores for performance; and CPUs with smaller "C" core for things that will benefit from density rather than power. The only difference between the two is physical size due to amount of L3 included, and so far there are no products which blend the two. I don't know what Intel's plan is.

Potato Salad
Oct 23, 2014

nobody cares


Dr. Video Games 0031 posted:

I hope you're excited for even more e-cores for the 14th and 15-gen too by the way. Intel apparently plans not adding any more P-cores for a while and just pushing more and more E-cores instead.

I'm honestly sold, I'm on a 8+8 12th gen and....it works. It loving works.

Got some e cores at work too, it works.

Rinkles
Oct 24, 2010

What I'm getting at is...
Do you feel the same way?
Are the Quicksync (QSV) presets not available because I'm using my dGPU? This is Handbrake.



I have a 11600K.

Hasturtium
May 19, 2020

And that year, for his birthday, he got six pink ping pong balls in a little pink backpack.

Rinkles posted:

Are the Quicksync (QSV) presets not available because I'm using my dGPU? This is Handbrake.



I have a 11600K.

That appears to be the case, if your setup prevents the IGP from being used while a dGPU is in use. Some motherboards and configurations are more accommodating than others.

lih
May 15, 2013

Just a friendly reminder of what it looks like.

We'll do punctuation later.
oh those prices actually are pretty good since they're only raising the 13600K a little bit and nothing else

redeyes
Sep 14, 2002

by Fluffdaddy
Not a performance post, but switching from an X570 amd system to Intel 12th gen x690 was really nice. Small glitches that had been happening just completely stopped, not the least of which was my USB freaking out multiple times a day requiring me to add in a PCIe USB card so I could have 2 working ports. Fun stuff. The Intel is hotter but oddly not much. 3900x vs 12700k with just a max boost on all core OC.

https://www.youtube.com/watch?v=ZrnWII_XQlw

drat glitches keep happening.

redeyes fucked around with this message at 02:37 on Sep 28, 2022

PC LOAD LETTER
May 23, 2005
WTF?!

redeyes posted:

drat glitches keep happening.

Supposedly that glitch can effect Intel systems too. Its some sort've issue with the Geforce driver and that MS update and not AMD's software or the CPU.

He mentions in the vid that if you don't install that Microsoft update the issue mostly goes away.

Ihmemies
Oct 6, 2012

lih posted:

oh those prices actually are pretty good since they're only raising the 13600K a little bit and nothing else

It is very good prices stay the same. I can upgrade later from my 8+4 to 8+16, like in s year or two, if I need more multicore processing power. 13900K is very fast in games and multithreading. And you can still cool it on air, unlike AMD, lol.

Arzachel
May 12, 2012

Ihmemies posted:

It is very good prices stay the same. I can upgrade later from my 8+4 to 8+16, like in s year or two, if I need more multicore processing power. 13900K is very fast in games and multithreading. And you can still cool it on air, unlike AMD, lol.

You'll have a harder time aircooling the 13900k than any AMD CPU.

BlankSystemDaemon
Mar 13, 2009



Arzachel posted:

You'll have a harder time aircooling the 13900k than any AMD CPU.
That's why I'm going AMD when I eventually get enough money saved up - I absolutely do not want to have a pump creating more noise than any even half-good fan, let alone a case full of Noctuas.

Kibner
Oct 21, 2008

Acguy Supremacy

BlankSystemDaemon posted:

That's why I'm going AMD when I eventually get enough money saved up - I absolutely do not want to have a pump creating more noise than any even half-good fan, let alone a case full of Noctuas.

TBF, it is nearly trivial now to set power limits in the BIOS for both Intel and AMD chips. Of course, you would also need to test performance to see where you want to fall on the power consumption vs performance curve, which takes a not-insignificant amount of time.

Twerk from Home
Jan 17, 2009

This avatar brought to you by the 'save our dead gay forums' foundation.
Both of these specific fireballs of chips apparently happily run with 65W power limits that an Intel stock cooler from 10 years ago can dissipate without problems. The first real reviews of the 7950X note that running it at 1/3 the stock power limit still gets more than 80% of peak performance at full wattage, and Intel themselves called out that the 13900K at 65W performs the same as the 12900K at 241W.

If you want a huge cooler on these, that's your call, but a $30 single tower cooler can easily handle ~140W that these CPUs will want to perform well.

Arzachel
May 12, 2012
I wonder if we're going to see a shift in reviews to account for the fact that running CPUs full tilt produces numbers that are irrelevant for the vast majority of users.

Ihmemies
Oct 6, 2012

I have a 150W limit on my 12700K and I don't think it hits even that very often. Definitely not when gaming at least. When gaming CPU temps are around 55C depending on game with the big Noctua dual Tower cooler.

gradenko_2000
Oct 5, 2010

HELL SERPENT
Lipstick Apathy

Arzachel posted:

I wonder if we're going to see a shift in reviews to account for the fact that running CPUs full tilt produces numbers that are irrelevant for the vast majority of users.

CPUs have been able to outrun their own stock coolers for a while now, to the point where Intel doesn't bother giving you one with their K SKUs, and AMD has only stopped recently because running their chips full-tilt is the assumed behavior all the time and they have no "non-K" or "T" model that'll run "cooler" out of the box (anymore).

some enterprising reviewers might add a new section that's like, "power normalized performance" against a set/flat power target, or two targets, with one being particularly low like 65w or less, but they're always going to headline the redline, so to speak.

BlankSystemDaemon
Mar 13, 2009



Kibner posted:

TBF, it is nearly trivial now to set power limits in the BIOS for both Intel and AMD chips. Of course, you would also need to test performance to see where you want to fall on the power consumption vs performance curve, which takes a not-insignificant amount of time.

Twerk from Home posted:

Both of these specific fireballs of chips apparently happily run with 65W power limits that an Intel stock cooler from 10 years ago can dissipate without problems. The first real reviews of the 7950X note that running it at 1/3 the stock power limit still gets more than 80% of peak performance at full wattage, and Intel themselves called out that the 13900K at 65W performs the same as the 12900K at 241W.

If you want a huge cooler on these, that's your call, but a $30 single tower cooler can easily handle ~140W that these CPUs will want to perform well.
The thing is, paying out of the nose for something and then not being able to use it because you're forced to set heavy limits on it - that's not something I want to do.

Ihmemies posted:

I have a 150W limit on my 12700K and I don't think it hits even that very often. Definitely not when gaming at least. When gaming CPU temps are around 55C depending on game with the big Noctua dual Tower cooler.
I just wish Noctua would make something that can fit in a 3U, but I guess I'll need to buy a 4U chassis - shouldn't be too expensive, hopefully.

Twerk from Home
Jan 17, 2009

This avatar brought to you by the 'save our dead gay forums' foundation.

BlankSystemDaemon posted:

The thing is, paying out of the nose for something and then not being able to use it because you're forced to set heavy limits on it - that's not something I want to do.

A little fiddling with PBO curves and you can reduce power usage by 50W without giving up a lick of performance. If you're willing to sacrifice 3-5% you can probably cut power usage in half.

I'm absolutely planning to turn a high-end part down a bit to make less heat.
https://www.youtube.com/watch?v=FaOYYHNGlLs

Hasturtium
May 19, 2020

And that year, for his birthday, he got six pink ping pong balls in a little pink backpack.

Twerk from Home posted:

A little fiddling with PBO curves and you can reduce power usage by 50W without giving up a lick of performance. If you're willing to sacrifice 3-5% you can probably cut power usage in half.

I'm absolutely planning to turn a high-end part down a bit to make less heat.
https://www.youtube.com/watch?v=FaOYYHNGlLs

Amen. I’m pretty well computationally set for a while on my desktops, but if I decide to get ambitious again I would never run one of these chips at fireball stock settings. Cutting the stock max voltage in half, still getting 90% of the performance, and not outrageously heating up my Texas house is a no-brainer for me.

BlankSystemDaemon
Mar 13, 2009



Twerk from Home posted:

A little fiddling with PBO curves and you can reduce power usage by 50W without giving up a lick of performance. If you're willing to sacrifice 3-5% you can probably cut power usage in half.

I'm absolutely planning to turn a high-end part down a bit to make less heat.
https://www.youtube.com/watch?v=FaOYYHNGlLs
Well, maybe I'll look into it - I just don't want to futz around too much with overclocking et al.

I'm certain any modern CPU is gonna be a fair bit faster than my i7-2600 (non-k) overclocked to 3.7GHz.
Especially considering all the instructions-per-clock improvements that's been made since 2011, and that basically any CPU I'm looking at will turbo-boost to at least 700MHz above, which is one of the breakpoints for a single-threaded compute task feeling noticably faster.

It should also be said that the power usage is only part of the reason I want to avoid Intel.
I'd much rather have a chiplet-with-NUMA design, because those are by-and-large solved for any modern OS, whereas there's still many unanswered questions about HMP, up to and including if it's even an idea that'll stick around long-term.

BlankSystemDaemon fucked around with this message at 16:14 on Sep 28, 2022

VorpalFish
Mar 22, 2007
reasonably awesometm

BlankSystemDaemon posted:

The thing is, paying out of the nose for something and then not being able to use it because you're forced to set heavy limits on it - that's not something I want to do.

I just wish Noctua would make something that can fit in a 3U, but I guess I'll need to buy a 4U chassis - shouldn't be too expensive, hopefully.

You are able to use it though? You can choose where you want to be on the spectrum of raw performance to efficiency - just like you could in years past.

Just previously, the processors would have shipped with defaults closer to the efficiency side and you could overclock to choose raw performance instead. Now they ship dialed to 11 and you can power limit to choose efficiency instead.

Giving up some raw performance to gain efficiency is often a good thing.

BlankSystemDaemon
Mar 13, 2009



VorpalFish posted:

You are able to use it though? You can choose where you want to be on the spectrum of raw performance to efficiency - just like you could in years past.

Just previously, the processors would have shipped with defaults closer to the efficiency side and you could overclock to choose raw performance instead. Now they ship dialed to 11 and you can power limit to choose efficiency instead.

Giving up some raw performance to gain efficiency is often a good thing.
I don't know if i'm able to use it; probably not while gaming (except for a very small subset of games, since most don't thread-scale linearly), and almost certainly while building the FreeBSD base system since it scales pretty well up to 512 threads thanks to some fairly recent optimizations.

The thing is, I'm on a very tight budget because my health is absolute poo poo so I can't work, and with electricity prices being absolutely nuts in Denmark (even more-so than they are in the rest of the world, I mean), that alone is gonna be a considerable cost during the lifetime of the system.

So is 3-5% potential performance that ups both the cost of the CPU as well as the total lifetime ownership cost worth it? I don't think so, but I don't seem to have much choice.

Dr. Video Games 0031
Jul 17, 2004

You can undervolt and do power limits to cut down power usage on Intel's chips too, though. So really just go with who gives the best performance at the price you want to pay.

BlankSystemDaemon
Mar 13, 2009



Dr. Video Games 0031 posted:

You can undervolt and do power limits to cut down power usage on Intel's chips too, though. So really just go with who gives the best performance at the price you want to pay.
Intel is going hard on HMP, an unproven and unfinished technology, so that doesn't give me a whole lot of choice.

VorpalFish
Mar 22, 2007
reasonably awesometm

BlankSystemDaemon posted:

I don't know if i'm able to use it; probably not while gaming (except for a very small subset of games, since most don't thread-scale linearly), and almost certainly while building the FreeBSD base system since it scales pretty well up to 512 threads thanks to some fairly recent optimizations.

The thing is, I'm on a very tight budget because my health is absolute poo poo so I can't work, and with electricity prices being absolutely nuts in Denmark (even more-so than they are in the rest of the world, I mean), that alone is gonna be a considerable cost during the lifetime of the system.

So is 3-5% potential performance that ups both the cost of the CPU as well as the total lifetime ownership cost worth it? I don't think so, but I don't seem to have much choice.

So it sounds like setting power limits makes sense for you- you trade maybe 25% of your all core performance for like a 50% reduction in power consumption. Chip is easier to cool and cheaper to run, while gaming workloads are mostly unaffected since they don't draw much power comparatively even without limits.

A modern CPU limited to 125w is going to be plenty fast for most people.

Twerk from Home
Jan 17, 2009

This avatar brought to you by the 'save our dead gay forums' foundation.

VorpalFish posted:

So it sounds like setting power limits makes sense for you- you trade maybe 25% of your all core performance for like a 50% reduction in power consumption. Chip is easier to cool and cheaper to run, while gaming workloads are mostly unaffected since they don't draw much power comparatively even without limits.

A modern CPU limited to 125w is going to be plenty fast for most people.

The initial numbers that we've seen are much better than that. Ars Technica saw an 18% performance drop from a 62% reduction in power consumption for the 7950X, and Intel said that the 13900K at 65W outperforms the 12900K at 241W, which is really promising.

Saukkis
May 16, 2003

Unless I'm on the inside curve pointing straight at oncoming traffic the high beams stay on and I laugh at your puny protest flashes.
I am Most Important Man. Most Important Man in the World.
What is the price and power consumption of a chip that runs at -18% performance by default?

Kibner
Oct 21, 2008

Acguy Supremacy

Saukkis posted:

What is the price and power consumption of a chip that runs at -18% performance by default?

Look at chips before self boosting algos became a thing.

Dr. Video Games 0031
Jul 17, 2004

You really don't have to lose any performance at all in order to cut back power consumption and reduce temps significantly. This has already been established.

Palladium
May 8, 2012

Very Good
✔️✔️✔️✔️

Dr. Video Games 0031 posted:

You really don't have to lose any performance at all in order to cut back power consumption and reduce temps significantly. This has already been established.

Dunno about y'all but once I experienced the wonders of a <30W 5600 in games I will never go back to the space heaters.

VorpalFish
Mar 22, 2007
reasonably awesometm

Saukkis posted:

What is the price and power consumption of a chip that runs at -18% performance by default?

Uh I guess look at Intel's non k or even T skus? Hard to say for sure though because of mobo fuckery. What even is a default anymore?

Alarbus
Mar 31, 2010
Hey is there a decent guide or site to undervolting or capping the power on the 11700k? Some cursory googling returned a lot of results on how to shove more power through it, not less. I did it to my 3080 to great success, so I'd be interested in trying the cpu as well.

Broken Machine
Oct 22, 2010

Alarbus posted:

Hey is there a decent guide or site to undervolting or capping the power on the 11700k? Some cursory googling returned a lot of results on how to shove more power through it, not less. I did it to my 3080 to great success, so I'd be interested in trying the cpu as well.

Probably the easiest is to just use Intel's tuning util:

https://www.intel.com/content/www/us/en/download/17881/intel-extreme-tuning-utility-intel-xtu.html

There's a slider there to lower the voltage, or you can set it to max tdp. It's pretty simple.

Alarbus
Mar 31, 2010
Oh. That's remarkably easier than I expected. I've used XTU before, I'll have to see if I still have it installed. Thanks!

Rinkles
Oct 24, 2010

What I'm getting at is...
Do you feel the same way?
For that one other person interested in RPCS3, there's now a dev maintained CPU tier list.



I guess Intel's "waste of sand" found its niche (unfortunately my 11600K is a big step below).

Danhenge
Dec 16, 2005
I cannot believe how well my purchase of a 4790K is holding up, even without buying a Z board. It's low-key the 2600k of its generation.

wargames
Mar 16, 2008

official yospos cat censor

Danhenge posted:

I cannot believe how well my purchase of a 4790K is holding up, even without buying a Z board. It's low-key the 2600k of its generation.

The 12400 looks to be the same for the current gen.

Palladium
May 8, 2012

Very Good
✔️✔️✔️✔️

Danhenge posted:

I cannot believe how well my purchase of a 4790K is holding up, even without buying a Z board. It's low-key the 2600k of its generation.

i ran a 4790K on a H81 in 2014 and every 'OC = free performance' bigbrained meme parrot (when it's clearly not) mocked me at being an idiot

Adbot
ADBOT LOVES YOU

Danhenge
Dec 16, 2005
I didn't feel too dumb because it was only like $10 more than the 4790 non-k and it's legitimately faster.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply