[s002wjh]

so haven't purchase psu for long time, and need a new one for new build. i remember seasonic is good. what other psu is good these days. 750 to 1kw range

 

[shilka]

There are a number of options such as the Super Flower Leadex III the Corsair RMx and HX and the Seasonic Focus and Prime series

The EVGA G6 is a Seasonic focus rebrand and so is the Asus Strix the Phanteks AMP and the Antec High Current Gamer Gold

The Antec Signature Titanium is a Seasonic Prime rebrand and so is the newest Corsair AX series as well as the Asus Thor
There is also the Fractal Design Ion series and the Be Quiet Straight Power 11 and Dark Power 12

 

[s002wjh]

so bascially seasonic 😁 seem like everything is rebrand of seasonic.

 

[shilka]

Super Flower makes their own units the Corsair HX and RMx are made by CWT and the Fractal Design Ion series are made by High Power

Most of the Be Quiet units are made by FSP
So no not everything is made by Seasonic

 

[Always Counterclockwise]

I just did a build with the Fractal Ion+ 2 660 W and the soft flex of the cables made them nice to route, the silent mode and 140 mm fan are handy, and I couldn't find any disadvantage compared to other similar units. Fractal tops out at 860 W, though, and the 860s can be hard to find.

For another build I'm currently considering Seasonic Prime PX (500–1300 W) and be quiet! Straight Power 11 (550–1200 W). The Primes cost a bit more but are quieter. It's more expensive for me to get Super Flower or FSP directly than relabeled, but that varies with location.

 

[Dogzilla07]

Fractal Ion cables are indeed awesome, best bang for buck cables. Gotta look at Corsair SF itx or the Asus Thor 2 for finer cables.

 

[Kaltenbrunner]

Has much changed in the last 5 or so years, in terms of top brands, and series ??

 

[Dogzilla07]

Has much changed in the last 5 or so years, in terms of top brands, and series ??

For the first time in 20 years, everything changes for PSUs, later this year, early next.

The new ATX 3.0 PCI-E 5.0 spec with the new 16-pin (12VHWRP) cable requiring a pseudo-seperate 12V rail is incoming. And Aris mentioned the new 160% and 180% PE (Power Excursion) requirements will be a breaking point for PSUs, as soon as 4070/4080/4090 possibly.

This time it's not a question of if anymore, just a question of when (as everyone in the industry is working with the new standard)

The first Gen 5 PSU has already failed the 160%, 180% Transient PE testing as required by PCI-SIG + Intel (in collaboration with Nvidia, and others in the industry).

Gigabyte UD1000GM PG5 1000 W Review - The First PCIe 5.0 PSU

The Gigabyte UD1000GM PG5 is the first available PSU with a 16-pin PCI-Express 5.0 connector. The only graphics card requiring such a connector is the NVIDIA GeForce RTX 3090 Ti, which also comes with an adapter, but the upcoming 4000 series will need it as well, so it is good to be ready.

www.techpowerup.com

MSI MEG Ai PCIE 5.0 series high-end power supply unit compatible with PCIe5.0 / ATX3.0

MSI launched the MEG Ai PCIE 5.0 series, the company's first PCI-Express 5.0 / ATX 3.0 compliant power supply unit....

www.guru3d.com

Currently it looks like the GPUs will just automatically throttle themselves to 70%, 50%, 30% of their capability and reduces their power usage (akin to 1x8pin cable failing and the GPU running drastically below it's TDP), if a certain PSU is not capable of handling them. Though it's quite possible for shutdowns to occur, on some older PSUs. Just have to wait and see when 4070/4080/4090 launch.

 

[shilka]

Has much changed in the last 5 or so years, in terms of top brands, and series ??

Yes other than some of the other 2 and 3 series and the new G6 series EVGA is a total joke and should be avoided
They are more interested in flooding the market with 10 mediocre series instead of sticking with 2-4 good ones

 

[Always Counterclockwise]

Bumping this thread since it's been quiet and the build I mentioned last month is shaping up. Looks a 3070 Ti (ASUS TUF O8G Gaming, probably) and, while it seems like most of the problems with Ampere triggering overcurrent protection are with the 3080 and 3090, it'd suck to get parts for an all new machine and then have the 3070 Ti crowbar it.

So, do folks have supply suggestions here? This is for AI GPGPU so I'm expecting the card to be running at full load but I'm not having much luck searching, either for finding people saying a supply's good or that a 3070 Ti is tripping it.

I know the minimum recommended for the 3070 Ti is 750 W but the system's got enough other draw it'll be 850 or 1000 W. Trying to stay with a 150–170 mm supply as the selected case is a little tight, so looking towards Fractal Ion+ 2, Seasonic Prime PX, and Super Flower Leadex VI Platinum PRO. I know Seasonic had lots of issues around Ampere triggering their overcurrent but it seems like those have been fixed.

While I mentioned Straight Power up thread, be quiet!'s split rails seem maybe higher risk for Ampere triggering overcurrent (a month ago the thinking was Navi).

 

[Dogzilla07]

PSU Tier List rev. 16.0b - Cultists Network is an incredible tier list

Cybenetics Labs – PSU Efficiency & Noise Level Certifications - Power Supplies almost every PSU is here (and great to see the noise the fan makes on various loads for each psu).

geizhals.eu is the best site if you want to sort, filter and compare

Best reviews are on techpowerup+tomshardware (Aris who made cybenetics above), and GamersNexus is a close second

saying a supply's good or that a 3070 Ti is tripping it.

The biggest offenders for tripping PSUs, both on transient spikes (old Seasonic Focus and some others [that are mentioned in the fine print on the tier list above]), and 12v sense wire noise issue (old Seasonic Prime =>850w)

I know the minimum recommended for the 3070 Ti is 750 W

Official minimum for PSUs from NVidia/AMD etc is not that accurate, the best way is to cross-reference the awesome O!technology roundups:

And double-check power draw from good 3rd party testing like GamersNexus, HardwareUnboxed, Techpowerup

ASUS GeForce RTX 3070 TUF Gaming OC Review

In our ASUS GeForce RTX 3070 TUF Gaming review we find that the card comes with an amazing cooler that achieves whisper-quiet noise levels and incredibly low temperatures at the same time. A dual-BIOS feature is included, too, as well as a second HDMI output.

www.techpowerup.com

For your GPU specifically it will never go over ~250W stock, or ~280W overclocked.

 

[Dogzilla07]

@Always Counterclockwise For what you're looking specifically, Adata XPG Core reactor is an option that is currently the most perfect on the market in every metric except how soft the cables are. Fractal Design Ion+/Ion+2 is the second most perfect option as it has soft cables (and is very close to the XPG Core reactor). EVGA G6 is also a very good shared second spot. Super Flower Leadex VI Platinum PRO is also a great option, it also has soft cables (slightly less than Fractal but more than other PSUs). It's a bit lower quality/performance overall, but for a 3070 Ti still more than enough.

So yeah, the Ion + Leadex VI you mentioned are a great options, and I'd just add the XPG Core Reactor and EVGA G6 to the mix,

1. Ion+ Leadex (soft cables and 135/140mm fan in a 150mm deep chassis).

2. EVGA G6 (harder cables and 135mm fan in a 140mm deep chassis)

3. XPG Core Reactor (harder cables and 120mm in a 140mm deep chassis)

Now the only other thing to keep in mind is PSUs with thicker cables mean you can use pig-tails (plug in 2x8pin into the GPU, from one cable from the PSU), while that is not recommended for PSUIs with thinner cables. Now for your 3070 TI, even 18AWG cables should be fine. from the 4 mentioned above at =>850W they all have 16AWG thick except the Super Flower.

There's no confirmation (only 1 review so far), but based on previous PSUs they've done superflower puts thin cables (as they have different plug design on the PSU.

So the SuperFlower might be risky and might have even 20AWG up to the first PCI-E 8 pin, which would make it risky to use 1 cable for the plugs even for your GPU. And it would be recommendable to go 1x cable to 1 x plug on the PSU for the Super Flower.

the Leadex VI will still work fine, just you would have to use more cables.

 

[Always Counterclockwise]

I've unfortunately found review coverage is generally spotty. For any given supply there's very likely a decent review of some broadly similar but somewhat older version from the manufacturer somewhere. But that doesn't help much with checking whether mistakes were made in an update or in quantifying version to version differences like noise reductions. Maybe you get lucky and the version you're interested in has a review though.

So, yeah, I'd prefer if there was more information on the Leadex VIs but 1:1 cable to plug is the plan for the 3070 Ti. Impedances in parallel divide, after all. It'd be nice if manufacturers speced their AWGs but I guess the six extra characters would be just too much typing. Be even better to know the RLCG parameters―OCP triggering droop in response to a GPU spike is probably mainly controlled by inductance―but I can't see that happening anytime soon, quite possibly never.

So far as I know the EVGA G6 and P6 are pretty much Seasonic Focus GX-1000 and PX-1000 supplies. I'm not expecting the cable routing to be so tight their being 140 mm makes a big difference. Mostly what I was trying to say is a 180 or 200 mm supply, like some of the Leadex versions or Corsairs (Chicony?), probably isn't a great idea. I haven't been able to find any clarity on whether EVGA offers functional improvements from the Seasonics or whether the changes are just cosmetic. It's been similarly difficult to get clarity on differences between current Prime and Focus models though most of what I've encountered seems to hint toward the Primes.

I'm more concerned with maximum on application noise rather than Cybenetics' average noise across the fan's RPM range in an anechoic chamber. Aris also uses average noise in his reviews and, from what I can tell, it ends up being an indication of a supply's fan stop load percentage more than anything else. So whilst it's nice XPG―unlike EVGA, Fractal, Seasonic, and Super Flower―holds current Cybenetics certifications I've found it difficult to translate the information into some sort of decision. Not that the compliance testing for 80+ is a bad thing, I'm just not sensitive to the efficiency details around platinum level.

All this leaves me feeling like it's a guessing game of picking a supply to test and hoping you don't dislike it enough to send it back and try something else. That's probably a bit pessimistic but at least with the Vento Pro on XPG's Cybercore 1000W you know the fan itself isn't likely to be a noise problem. Aris does express a desire for it to be a 140 but I suspect a power supply has a good chance of presenting enough flow resistance a 120 can provide somewhat more noise-normalized airflow than a 140.

For your GPU specifically it will never go over ~250W stock, or ~280W overclocked.

TechPowerUp seems to have the best setup for this and they've measured 330 W sustained and 425 W 20 ms spike from a 3070 Ti. That's consistent with the Ti usually running about 50 W more than the non-Ti 3070 (for example, in the TUF 3070 non-Ti review you linked TechPowerUp measured 280 W sustained).

Since waiting for ATX 3.0 supplies and cards isn't an option for this build, how to rate the GPU is main thing which makes me hesitate over an 850 W supply. Calling the 3070 Ti 330 W and using the old 1.3x rule suggests 850 W as a minimum supply. With current GPUs 1.5–1.6x seems more common, which indicates a 1000 W supply. Alternatively, using 1.3x on 425 W also indicates for 1000 W.

 

[Dogzilla07]

I've unfortunately found review coverage is generally spotty.

Depends on the brands, the spreadsheet on the cultists Tier list has incredible vast and up to date reviews:

PSU tier list sources & info [cultists.network]

About <a href="https://cultists.network/140/psu-tier-list/">PSU Tier List at our site : Cultists.Network</a> Contact : Discord:,<a href="https://discord.gg/cQrnrtyd8b">Our server</a> LukeSavenije#9722 (English / Dutch) Elisis#9169 (English) Juular#6129 (English / Russian) Meowrice#0001 (German/...

docs.google.com

But that doesn't help much with checking whether mistakes were made in an update or in quantifying version to version differences like noise reductions. Maybe you get lucky and the version you're interested in has a review though.

Only source for this would be Aris youtube channel Hardware Busters International and GamersNexus channel and PSU cultists Discord. Aris for instance did go in-depth of early sample vs quickly fixed-re-done samples of Gospower made Cooler Master V SFX and MWE V2 White/Bronze which had issues.

https://www.youtube.com/c/HardwareBustersInternational/videos

clarity on whether EVGA offers functional improvements from the Seasonics or whether the changes are just cosmetic. It's been similarly difficult to get clarity on differences between current Prime and Focus models though most of what I've encountered seems to hint toward the Primes.

EVGA did do some meaningful changes, they got 16AWG cables to the first 8 pin PCI-E pig-tail (but only on 850W and above xD). and they got a true 50c full range of the PSU. As for Prime vs Focus, yeah that was weird, with all primes 750W and below actually being based on the focus platform. Best place to get info like this is linustechtips power supply forum section (jonnyguru and a couple of other really knowledgable/experienced people post regularly), and the discord for PSU cultists.

This is the fan in the EVGA G6/P6 units:

Seasonic PRIME GX-650 | 6. Sistema di raffreddamento | Recensione

 

[Dogzilla07]

I'm more concerned with maximum on application noise rather than Cybenetics' average noise across the fan's RPM range in an anechoic chamber. Aris also uses average noise in his reviews and, from what I can tell, it ends up being an indication of a supply's fan stop load percentage more than anything else.

Yup, if you're really sensitive to noise, and/or have a SFF build close on the table, the standard/A/A+ tag is not useful, you do end up needing to check the actual Cybenetics PDF. And then of course low frequency noise coil whine isn't covered xD

TechPowerUp seems to have the best setup for this and they've measured 330 W sustained and 425 W 20 ms spike from a 3070 Ti.

Yeah that's the Extreme version xD, they nicely mentioned ~50w more on top of the MSI 3070 Ti you have

Calling the 3070 Ti 330 W and using the old 1.3x rule suggests 850 W as a minimum supply

That's not necessary, at least for the 3070 Ti, the spikes are covered within OCP triggering points, without over-speccing the PSU. For 4000 series, well it's gonna be interesting.

Aris does express a desire for it to be a 140 but I suspect a power supply has a good chance of presenting enough flow resistance a 120 can provide somewhat more noise-normalized airflow than a 140.

It's time-consuming yes. You need to completely redesign the inlet, and have to use grill shroud to make effective use of 135/140mm fans. For instance Be Quiet! Dark Power Pro 12 has a novel never before seen inlet and shroud (a meshfilter like on Lian Li/Phanteks cases). And it does wonders. But it added dozens if not hundreds of hours of work, and who know how much it cost xD

Another thing in be quiet's favour is that their bearing is based on sleeve design and not ball-bearing, so it has noise advantages. (it also doesn't mind operating horizontally as it's probably matsushita rifled or whatever they've done to make it immortal like Noctua's xD

 

[Always Counterclockwise]

you do end up needing to check the actual Cybenetics PDF

None of the .pdfs I've looked at have dB(A) as a function of RPM or load, only the average. And, since most supplies of interest here lack certs, .pdfs typically don't exist to check. Kitguru, TechPowerUp, and Aris' Tom's versions can give enough dB(A) points to have a decent idea. Depends on the review. At least it's possible to get some sense of what's typical and pick up on some of the fails. Like the 56 dB(A) Seasonic SnowSilent 1050W.

I'm not particularly noise sensitive but, given the tendency towards fairly substantial increases in noise above 750 W, it is looking like the power supply is likely the loudest noise source in the build. So some before purchase knowledge of the fan curve would be nice for trying to get the supply not to run louder than the GPU fans in actual use. I guess where I am with this is the PX-1000 probably has tighter load regulation and quite possibly something like 15 dB(A) more fan noise than the Cybercore 1000W. How much coil whine might result from the Cybercore's looser regulation is hard to tell but my guess is most likely it wouldn't be worse than the fan.

For the moment I find myself leaning towards XPG. Wouldn't say they're great at communicating what they're building but they're at least somewhat less bad at it than the obvious competitors.

That's not necessary, at least for the 3070 Ti

A 20 ms spike is basically DC as far as the control loop and several protection mechanisms are concerned. Whilst forum viewpoints tend towards willingness to push operating margins, I think mostly due to lack of familiarity with power supply design and operation, it's common engineering design practice to rate more conservatively. The underlying question here is how brief a transient needs to be before it stops meriting consideration from as essentially steady state, which isn't a hard cutoff and has an application dependent transition. Not being an ATX designer I can't speak to this directly but I can say, for the supplies I've been involved with, 1) the transition from DC-like to transient consideration mostly happens in the 10 ms to 3 ms range and 2) I'm not seeing anything in ATX motivating substantially different considerations.

For example, Gamers Nexus pointed out yesterday they're seeing 2–3x 100 μs spikes from Ampere. Those are well into the transient zone. So, yeah, I wouldn't be rating a 3070 Ti at 660 or 990 W for power supply specification. But rating at 425 W for a longer duration 1.3x spike doesn't look unreasonable to me. A good portion of the design I have done has had to handle 1.4–1.5x at similar bandwidths.

 

[Dogzilla07]

I think mostly due to lack of familiarity with power supply design and operation, it's common engineering design practice to rate more conservatively. The underlying question here is how brief a transient needs to be before it stops meriting consideration from as essentially steady state, which isn't a hard cutoff and has an application dependent transition. Not being an ATX designer

What I'm saying is coming directly from posts i observed from Jonnyguru the chief designer at Corsair, and the previous owner of the most prestigious and venerated PSU review website, and couple of other peeps that know their stuff (as well as contributor to the tier list). not "users" only. People are running 3070 Ti on Corsair SF 750W platinum np and 12700k stock and it's only 25-30dB at ~570w range).

None of the .pdfs I've looked at have dB(A) as a function of RPM or load, only the average.

Corsair TX-M Series 750 W Review

If you need a PSU with high build quality and don't care about the semi-modular cable design, or top-performance in all areas, the TX750M fits the bill. It uses better parts than the RM750 and is sold at about the same price. However, the RM750 provides notably higher overall performance, and...

www.techpowerup.com

This is as good as it gets , plus the same map for rails+consumption noise for a huge number of ones on cybenetics

But rating at 425 W for a longer duration 1.3x spike doesn't look unreasonable to me

And it's gonna do nothing to quite a big chunk of Tier A PSUs on that tier list.

1) the transition from DC-like to transient consideration mostly happens in the 10 ms to 3 ms range and

Yup the 10ms to 1ms range is the most crucial, and is gonna be a big make or brake moment for 4000 series and PSUs. Aris talked about, how everyone focuses on 100us 200% spike/excursion value, but how that is nothing, and than the 10ms 160% spike is the biggest concern:

For the moment I find myself leaning towards XPG.

Yeah they really knocked it out of the park with their first PSU, same factory like Corsair, very well designed, in a smaller package than Corsair xD

 

[Dogzilla07]

So, yeah, I wouldn't be rating a 3070 Ti at 660 or 990 W for power supply specification. But rating at 425 W for a longer duration 1.3x spike doesn't look unreasonable to me. A good portion of the design I have done has had to handle 1.4–1.5x at similar bandwidths.

Lets say you're running that Zotac (it's spikey, but not so bad, and it's the only really spikey part in your system)

so 330w and 425w spike.

lets say 70-75w rest of the system (but probably less for most) and lets say ~170W CPU (12700k stock) that's ~570w normal and ~670W when GPU spikes and lets say it's the XPG Core Reactor (OPP for the 750w model is set at 885.6 W). So that's 35dB (at 1m) when not spiking. Which is not that great noise wise. 850w isn't much better fan is still at ~1600RPM.

The corsair RMx is much better in this regard

Fractal Ion+ even better (Ion+2 is very similar) and that's for the 760w unit:

Evga g6 750w not really good, a bit noisy:

But you get the gist, you can almost max a PSU at ~750W and get as good noise as if you would overspec some brand by many hundreds of watts. There's a decent number of outlier 600-900 PSUs which punch above their weight noise-wise even when they're maxed.

 

[Always Counterclockwise]

But you get the gist, you can almost max a PSU at ~750W and get as good noise as if you would overspec some brand by many hundreds of watts.

Yep, decent PSU fan curves all track similarly for noise as a function of load since they're generally using comparable fans and dealing with about the amount of heat because there's not big efficiency spreads among, say, 80+ gold and platinum supplies. In my case the system wattage is a bit higher than what you picked, the best proxy I can find for the PX-1000 suggests an aggressive fan curve, and the Cybercore 1000 runs a relaxed curve with an atypically quiet fan (for a PSU). That pushes the acoustic noise estimate towards a large difference between the two.

Particularly if the PX-1000 fan stop switch lowers the upper part of the curve the estimate may be off. Not finding info about that so I should probably email Seasonic.

 

[Always Counterclockwise]

Not finding info about that so I should probably email Seasonic.

I'd missed the S3FC stub page; at least as diagrammed, enabling fan stop (what Seasonic calls fanless mode) doesn't affect the upper part of the curve (cooling mode).