A comparison of stock vs efficient gaming

[eidairaman1]

They did. games.
Furmark used *less* since it doesnt surge, it's a constant sustained load


That 12VO test showing the reduced consumption at low load/idle sitatuations really is a game changer, now i dont wanna buy a new PSU til they're more readily available

Spikes/Surges are inrush current which can occur from voltage dips and brownouts, blackouts. Solution is the BBU/UPS.

No, I just don't have time to waste on LTT, that's the honest truth.

Other sources I may consider, and as you have provided, I will look into it. Stand by.

EDIT: Ok, that certainly is interesting. My immediate question is simple: Why? Because from a electrical stand point this makes zero sense. Are the mobos conversion electronics far lower wattage than the PSUs? That might make sense.


My evga SuperNova T2 is nearly 8 years old now, hard to believe. I will probably replace it with something when all this PSU standards dust settles.

I'd only replace it if there is a problem.

 

[R-T-B]

I'd only replace it if there is a problem.

That sounds boring. Logical, but boring.

 

[The red spirit]

EDIT: Ok, that certainly is interesting. My immediate question is simple: Why? Because from a electrical stand point this makes zero sense. Are the mobos conversion electronics far lower wattage than the PSUs? That might make sense.

Well, I'm complete dumbass when it comes to electronics, but Linus said that converting from AC to DC is inefficient and conventional PSU do it several tiems for each voltage (12V, 5V, 3.3V), then there's less conduction loss of transferring 12V through PSU cables to motherboard and then if you really need anything else 12V, DC to DC conversions is more efficient on board. But you can see that Linus bench doesn't seem to have some typical 5V or 3.3V components connected to it. No SATA SSDs, no fans apart from AIO, no DVD drive, no add in PCIe cards just bare vital components. Now, ATX12VO isn't entirely perfect and at tiny loads like 10 watts efficiency still tanks to 70%, but at 20 watts it's already past 80%, meanwhile conventional PSU reaches only 54% at 10 watts and 59% at 20 watts and that's on 500 watt units. That's all I could extract from LTT and OC3D.

I know that this is a bit off topic, but my daily computer spends a lot of time at idle or low loads. With all power efficiency settings on in BIOS and balanced Windows power plan, it consumes 42-46 watts at idle. It uses 60-70 watts at mild loads and in games it uses around 180-210 watts. And yet I use a rather modest wattage Chieftec A90 550 PSU, despite my components being fast enough for me, literally half of PSU wattage is never utilized and at idle, computer spends a lot of time in 8-12% usage range. So if I was willing to upgrade, ATX12VO would be awesome in terms of idle power savings, however, such low loads are only achieved by actually leaving C states, EIST, Speedshift and etc on and by using balanced or high performance power plans. If you overclock and turn those things off or just use ultimate performance power plan, then you waste a lot of watts. I tried that myself and my computer then used around 60 watts at idle and 70-90 watts at low loads. Also around 10 watts more in gaming. But I wouldn't say that there's no benefit in keeping those things off anyway, since disabling C State improves IOPS of SSD and makes stuttering in Forza Horizon 5 a bit less bad. In same Horizon 5 system starts to complain about low bandwidth when C state and other eco stuff is on. My board also doesn't allow for CPU to enter C10 or even C8 states at all or anything lower than C4 for package, forcing those things leads to freezing and system locking up. That's not great and perhaps due to my PSU nowhere stating Haswell power state compatibility (and for that matter a legit 80+ Gold rating). So perhaps I could benefit even from improved ATX 12V power supply, assuming that motherboard itself doesn't cause instabilities. And even with "promised' gains of 2x at idle, I would only save like 20-25 watts, in summer that actually matters a lot, when I can't tolerate any extra heat at all. Gaming during day is a big no no, only viable during evening or night. And only at idle or low loads i would see gains, meanwhile gains at high loads would be the most benficial.

 

[chrcoluk]

I have always felt c-state performance hit is reasonable providing you dont go below C3, and if C3 does cause stutters then C2 will almost certainly be fine as that has very low cost impact. Sadly I had to turn off C-States on my 8600k on this board because for some reason when the voltage fell below 1.00v I got some nasty coil while. Interestingly when the 9900k went in most of the coil whine went away so I could enable C-States again.

On the flip side I have found if you enable the cpu to downclock to 800mhz at idle, the responsiveness of the system is trashy, it will have a very "feel slow" feel about it. How I got round that issue was a tool called autopowerok, so when it detects kb/mouse input the system is not deemed idle and for X amount of minutes you configure it will keep you on a specific power profile, but when X amount of minutes pass it will change to another, so on the idle profile I allow idle clocks, but the active profile I dont.

 

[lordmogul]

There is one thing overlooked in the whole "it is wasted money to buy a big card and cripple it's performance. why not by a smaller card" idea, and that is "it depends"
Unless people are playing one game, and one game only, there will always be different demands.

If I take a 3080 and limit it to a more efficient point, reducing consumption by 40% while reducing performance by 10%, that doesn't mean that I wouldn't need that extra bit at some point. It only means that in this specific game, those specific efficiency improvements are gained.
In a different game the situation is different.
What if I want to play good old UT99, Civ 4 and the newest BF or CoD on the same machine that should also be able to stay in use for a couple years?
Sure, I could simply buy a 1050 for the older games, and that will service them well, and a 6700 XT for the more demanding ones, but 1-2 years down the line I need to replace them, spending more money in total. All depends obviously how fast they are (and this how quickly they need to replaced) how much they pull, and how much the initial card is.


Is it a waste of money to buy a more expensive card and limit it's performance and power draw? It depends.
On one hand I spend more for the card but have lower running cost. Depending on how much I save per hour/day/month/year it might be totally worth it or not at all. (same is with more efficient PSUs)
On the other hand, I have better efficiency now, and down the line, when demand increases, I can always remove the limit and can delay the need to buy a new card.
And in some cases the undervolted, underclocked highend card can even draw less power and perform better than a smaller one.

As long as the performance you get at the end, is "enough", going for better efficiency is only a benefit. If you stay over your 30/60/144/165/240/etc fps goal and you can reduce power draw, why not go for it.
Back in the day we overclocked for peak performance at the cost of efficiency, with today's components it's the other way around. The things already overclock themselves to the limit, and we have to go for better efficiency ourselves.
Indel didn't release a 4 GHz Pentium 4, because the 3.8 GHz model with 115W was already hard to cool, but a 12900K that peaks at 250W is okay?
A 3090 Ti delivers around 30-55% more performance than a 2080 Ti, but draws 65-70% more power and was released at 66% higher MSRP. So not only the performance/W but also the performance/$ went down.


I bought a GTX 1060 in late 2017 that is in use even now. Back then there was obviously the choice between it and the RX 580, which is slightly faster and had a slightly lower price. Why did I go with the "worse" card then? Efficiency. That small bit of extra performance came at a much higher power draw. High enough, that with my idle/load use profile by now it would've been more costly than a GTX 1080.
Yes a RX 580 being more expensive than a GTX 1080, and that without any of the mining-related price hike. And I did test how lucrative mining would be, but the result of less than 5€ per month for 24/7 use, at the cost of not only electricity but the noise and heat output and the fact that every moment I do something else on it would reduce that revenue just meant it was not worth the hassle.

 

[The red spirit]

I have always felt c-state performance hit is reasonable providing you dont go below C3, and if C3 does cause stutters then C2 will almost certainly be fine as that has very low cost impact. Sadly I had to turn off C-States on my 8600k on this board because for some reason when the voltage fell below 1.00v I got some nasty coil while. Interestingly when the 9900k went in most of the coil whine went away so I could enable C-States again.

Well, I had some random stutters in Horizon 5 anyway. Disabling C states doesn't eliminate them, but makes them slightly less annoying. But then I also run the game from hard drive, which is probably less than ideal for it anyway.

On the flip side I have found if you enable the cpu to downclock to 800mhz at idle, the responsiveness of the system is trashy, it will have a very "feel slow" feel about it. How I got round that issue was a tool called autopowerok, so when it detects kb/mouse input the system is not deemed idle and for X amount of minutes you configure it will keep you on a specific power profile, but when X amount of minutes pass it will change to another, so on the idle profile I allow idle clocks, but the active profile I dont.

It's really not as bad as you say, but there can be some performance loss if you do that. I only measured around 10-15 ns RAM latency increase with eco stuff on. Rather than outright loss, a variation. It be this much worse, but not always. With eco stuff off, you get consistently less latency and it doesn't vary as much. Variation then remains in 3 ns range, while on average RAM is 10-15 ns faster. But that's with my RAM, yours may be vary more and be more sensitive to that and thus you may see bigger performance delta.

 

[AusWolf]

I know that this is a bit off topic, but my daily computer spends a lot of time at idle or low loads. With all power efficiency settings on in BIOS and balanced Windows power plan, it consumes 42-46 watts at idle. It uses 60-70 watts at mild loads and in games it uses around 180-210 watts. And yet I use a rather modest wattage Chieftec A90 550 PSU, despite my components being fast enough for me, literally half of PSU wattage is never utilized and at idle, computer spends a lot of time in 8-12% usage range. So if I was willing to upgrade, ATX12VO would be awesome in terms of idle power savings, however, such low loads are only achieved by actually leaving C states, EIST, Speedshift and etc on and by using balanced or high performance power plans. If you overclock and turn those things off or just use ultimate performance power plan, then you waste a lot of watts. I tried that myself and my computer then used around 60 watts at idle and 70-90 watts at low loads. Also around 10 watts more in gaming. But I wouldn't say that there's no benefit in keeping those things off anyway, since disabling C State improves IOPS of SSD and makes stuttering in Forza Horizon 5 a bit less bad. In same Horizon 5 system starts to complain about low bandwidth when C state and other eco stuff is on. My board also doesn't allow for CPU to enter C10 or even C8 states at all or anything lower than C4 for package, forcing those things leads to freezing and system locking up. That's not great and perhaps due to my PSU nowhere stating Haswell power state compatibility (and for that matter a legit 80+ Gold rating). So perhaps I could benefit even from improved ATX 12V power supply, assuming that motherboard itself doesn't cause instabilities. And even with "promised' gains of 2x at idle, I would only save like 20-25 watts, in summer that actually matters a lot, when I can't tolerate any extra heat at all. Gaming during day is a big no no, only viable during evening or night. And only at idle or low loads i would see gains, meanwhile gains at high loads would be the most benficial.

I'm not an expert, but I guess you're right: you could benefit from a more modern power supply that supports modern C-states and all.

Personally, I've never had any issue with power saving features on any system. I hear a lot of people disable C-states and complain about system responsiveness when they're on, but I've never understood why. My PC spend most of its time in idle, so I prefer to leave everything on, and minimise power consumption and heat.

 

[The red spirit]

I'm not an expert, but I guess you're right: you could benefit from a more modern power supply that supports modern C-states and all.

But that's in theory, since it won't make sense for me to replace a perfectly working unit for 10 watt savings. It's just that it would be technically better.

Personally, I've never had any issue with power saving features on any system. I hear a lot of people disable C-states and complain about system responsiveness when they're on, but I've never understood why. My PC spend most of its time in idle, so I prefer to leave everything on, and minimise power consumption and heat.

Neither I exactly do, but C states on make me lose nearly half of SSD 4k write and read performance, same with IOPS and this seems to be common for people since Nehalem. Meanwhile my AMD FM2+ and AM3+ system with C states on felt really sluggish, but you had to keep them one to have turbo working. I just gave up on turbo instead, seemed like much bigger gain than from those few hundred megahertz. But those systems also have much higher L cache latencies too, so I guess C state impact on sluggish cache CPUs is higher.

BTW it's not just C states that contribute to such effects, but things like EIST too.

 

[AusWolf]

But that's in theory, since it won't make sense for me to replace a perfectly working unit for 10 watt savings. It's just that it would be technically better.

You would be replacing it to avoid system hangups and freezes. Personally, I think a good quality unit is worth the investment, but of course, it's not my place to decide.

Neither I exactly do, but C states on make me lose nearly half of SSD 4k write and read performance, same with IOPS and this seems to be common for people since Nehalem. Meanwhile my AMD FM2+ and AM3+ system with C states on felt really sluggish, but you had to keep them one to have turbo working. I just gave up on turbo instead, seemed like much bigger gain than from those few hundred megahertz. But those systems also have much higher L cache latencies too, so I guess C state impact on sluggish cache CPUs is higher.

BTW it's not just C states that contribute to such effects, but things like EIST too.

I've never had any of those issues, either, neither with EIST. Is the SSD performance something you notice in real life apps? If not, I wouldn't worry about it.

 

[The red spirit]

You would be replacing it to avoid system hangups and freezes. Personally, I think a good quality unit is worth the investment, but of course, it's not my place to decide.

My system doesn't freeze if I set them to auto or don't force them to work, so it's not like I have actual problem, just some less than ideal functionality. It's not even technically a loss of functionality, since ATX v2 standards and substandards are intended to work and be compatible. Motherboard detect PSU as lower standard one and reacts accordingly to ensure smooth operation. My board also has options for power supplies that don't have any support for ErP at all and while computer shut down uses around 2 watts instead of 0.2-0.3 watts, it still should function perfectly fine. All in all I don't really care too much about that and certainly wouldn't replace PSU for some tiny thing like that.

I've never had any of those issues, either, neither with EIST. Is the SSD performance something you notice in real life apps? If not, I wouldn't worry about it.

Well, not exactly all that well, but it seems to help some things like games to load a bit faster, perhaps sometimes system to respond faster. It's borderline placebo with Comet Lake system, but on AM3+ or FM2+ system even with SSD, C states really ruin responsiveness and yes then you start feeling it. It seems that APM (application profile manager) is the biggest performance tanking culprit, which is needed for turbo and C states working correctly. Cool and Quiet doesn't seem to hard performance/responsiveness much if at all. C6 state with APM does.

Anyway, on Intel systems there's weirdness with EIST and Speedshift. Speedshift was supposed to be hardware level EIST equivalent with way more speed and granularity. It sounds like it should have replaced less granular, slow and purely software based EIST, but for some reason boards have both of them optional and you can enable them both at the same time. I have no idea why and even weirder it seems like both are enable by default.

 

[chrcoluk]

Well, I had some random stutters in Horizon 5 anyway. Disabling C states doesn't eliminate them, but makes them slightly less annoying. But then I also run the game from hard drive, which is probably less than ideal for it anyway.


It's really not as bad as you say, but there can be some performance loss if you do that. I only measured around 10-15 ns RAM latency increase with eco stuff on. Rather than outright loss, a variation. It be this much worse, but not always. With eco stuff off, you get consistently less latency and it doesn't vary as much. Variation then remains in 3 ns range, while on average RAM is 10-15 ns faster. But that's with my RAM, yours may be vary more and be more sensitive to that and thus you may see bigger performance delta.

I didnt bench it, I just used the system, slow to load explorer windows and so on.

 

[agent_x007]

Furmark used *less* since it doesnt surge, it's a constant sustained load

If one would undervolt GPU to NOT be power and temperature limited under Furmark test, could that mean "sudden power spikes" are impossible ?
(my thinking : Power virus is worst case, so if that's "fine" - everything else must be as well)

 

[wolf]

undervolting is the new overclocking. I'd undervolt just about any video card these days, but with the aim to retain as much performance as I can rather than going ultra drastic.

 

[nguyen]

Just like commercial aircraft, we don't use the full rated power of the engines unless required to do so , undervolting would greatly extend the lifespan of the card.

Manufacturers engineer their products to last only as long as the warranty period to cut cost, but undervolting would make your GPU last at least 2x as long. So yeah I would undervolt everything since I like my hardware to last (even if I sell them I would still like them to last)

 

[Mussels]

I have always felt c-state performance hit is reasonable providing you dont go below C3, and if C3 does cause stutters then C2 will almost certainly be fine as that has very low cost impact. Sadly I had to turn off C-States on my 8600k on this board because for some reason when the voltage fell below 1.00v I got some nasty coil while. Interestingly when the 9900k went in most of the coil whine went away so I could enable C-States again.

On the flip side I have found if you enable the cpu to downclock to 800mhz at idle, the responsiveness of the system is trashy, it will have a very "feel slow" feel about it. How I got round that issue was a tool called autopowerok, so when it detects kb/mouse input the system is not deemed idle and for X amount of minutes you configure it will keep you on a specific power profile, but when X amount of minutes pass it will change to another, so on the idle profile I allow idle clocks, but the active profile I dont.

I've never disabled C-states on anything except one garbage gigabyte X58 board long, long ago.

This slowdown you're seeing isn't normal, somethings really really wrong for that to occur.

If one would undervolt GPU to NOT be power and temperature limited under Furmark test, could that mean "sudden power spikes" are impossible ?
(my thinking : Power virus is worst case, so if that's "fine" - everything else must be as well)

the spikes are the GPU boosting, especially if the load is uneven.
Underclocking the GPU is definitely a method to fix it and really popular with ampere owners


Every horizontal dot on this graph is another step it could boost up, you remove the higher voltage ones like i did and it's impossible for it to spike at all.
The negative is i may get FPS dips in heavy action, but i'm really not worried about dropping from 140 to 120.

Lets say 25% of the GPU Cores are in use, so they max out to 2.1GHz
Suddenly, an explosion. 100% of the cores are in use, but it's taken a few nanoseconds to downclock to 1.8GHz

In that brief moment, power draw suddenly spiked up

Add in stuff like the GDDR6X being so power hungry and you can see how balancing a 375W limit when 60W goes to the VRAM and add that in to the swings as well, and the swings get bigger and bigger

 

[Outback Bronze]

undervolting is the new overclocking

Have to agree : )

Either that or I know I aint getting too many more frames @ 4K to worry about the extra power and heat.

 

[chrcoluk]

I've never disabled C-states on anything except one garbage gigabyte X58 board long, long ago.

This slowdown you're seeing isn't normal, somethings really really wrong for that to occur.

Well its an odd comment from my perspective, its nothing complicated, you could simply leave a bios on defaults, install windows, touch nothing, and play with the min cpu speed setting, and its a very visible thing to see. I have literally onserved this on probably dozens of machines, and AMD even deliberately have their min cpu speed much higher than Intel as they themselves noted the same problem, they even said in a video somewhere I watched a few years back the reason they have higher idle clocks than Intel, is they would be slowing down the machine with no practical gain.

The reason for this is quite simple, there is lag between when the load increases and when there is a reaction to that load to increase the clock speed.

So the only thing I can conclude here is if you are not noticing this (on intel cpus), then maybe you just not sensitive to it, to many people the only performance they care about is peak performance (basically benchmarks, encoding performance, frame rate) which isnt really affected by this, its transitional performance thats affected, so the speed of things whilst the cpu is exiting an idle state.

But I dont know if your comment was aimed at the 800mhz behaviour or the c-states behaviour. I think its fine for you to say "I dont notice this myself" but to say "there must be something really really wrong" means you kind of suggesting what I am saying is impossible, which I think is wrong.

Speedshift was also developed to mitigate this problem that you said must mean "something is very wrong". It accelerates the transition between p-states reducing the lag.

 

[AleXXX666]

I never realised the 1080Ti was that good power wise. i would really like a 3070/80 but still a bit too much ££ for me. I don't mind the 1080Ti though, even at 1440 it's not too bad.

1080ti=2070super. i've had latter. it's baby smooth in 1440p. lol. but where to find 1080ti or 2070s now in working condition...

 

[Mussels]

Well its an odd comment from my perspective, its nothing complicated, you could simply leave a bios on defaults, install windows, touch nothing, and play with the min cpu speed setting, and its a very visible thing to see. I have literally onserved this on probably dozens of machines, and AMD even deliberately have their min cpu speed much higher than Intel as they themselves noted the same problem, they even said in a video somewhere I watched a few years back the reason they have higher idle clocks than Intel, is they would be slowing down the machine with no practical gain.

The reason for this is quite simple, there is lag between when the load increases and when there is a reaction to that load to increase the clock speed.

So the only thing I can conclude here is if you are not noticing this (on intel cpus), then maybe you just not sensitive to it, to many people the only performance they care about is peak performance (basically benchmarks, encoding performance, frame rate) which isnt really affected by this, its transitional performance thats affected, so the speed of things whilst the cpu is exiting an idle state.

But I dont know if your comment was aimed at the 800mhz behaviour or the c-states behaviour. I think its fine for you to say "I dont notice this myself" but to say "there must be something really really wrong" means you kind of suggesting what I am saying is impossible, which I think is wrong.

Speedshift was also developed to mitigate this problem that you said must mean "something is very wrong". It accelerates the transition between p-states reducing the lag.

C-states shouldnt cause performance loss unless you're running a way too old OS, or you've ran windows on the power saver plan which slows those transitions down

I do recall something about HPET in the BIOS being an issue with some hardware, forcing the windows timer to slow down that could be related - you should try out timerbench on one of your affected PC's

edit: yep it was a known issue for skylake, which matches the era of hardware of you've mentioned

 

[chrcoluk]

C-states shouldnt cause performance loss unless you're running a way too old OS, or you've ran windows on the power saver plan which slows those transitions down

I do recall something about HPET in the BIOS being an issue with some hardware, forcing the windows timer to slow down that could be related - you should try out timerbench on one of your affected PC's

edit: yep it was a known issue for skylake, which matches the era of hardware of you've mentioned

C-states cause a performance loss during transition, if you want I can link to the manufacturer documents, the question is if the performance loss is visible, some people only care about benches and frame rates in which case the only noticable impact from c-states might be stutters depending on the game. I did state its far less noticeable to me than allowing clocks to fall to 800mhz.