Lian Li SP850 First SFX PSU with ATX 3.0 + PCIe Gen 5, Including ATX12HPWR

[oxidized]

It's mainly about power spikes, also often called transient power draw, which are spikes of very high power draw (some times up to 2x the rated power of the GPU) that last for very short periods of time, typically in the microsecond range. It's a consequence lf the massive size (in transistors/cores) of current GPUs, their high clocks, and their aggressive boost algorithms which try to clock as high as possible at all times. When that is the case and the workload changes from something light to something heavy (which happens often in games), there's a risk of a massive power spike before the boost regulation system has the time to lower clocks in response to the heavier workload.

This isn't really a massive concern, and only really applies to high end GPUs (as they are very large and have huge power budgets), but it can cause system crashes/shutdowns if the PSU isn't able to handle the spikes, either through OCP/OPP triggering or through voltage dropping too low for the system to stay on. And it varies a lot between PSUs - I've seen 850W units shut down when powering a 3080, but I've also seen 600W units powering 3090s with no issues. Hence why the ATX 3.0 spec is trying to standardize this somewhat.

How has this not caused a giant controversy? I mean, it sounds to me like it's something which can happen more often than you'd expect and could cause many problems, how has this become acceptable at this point in time?

 

[Nanochip]

How has this not caused a giant controversy? I mean, it sounds to me like it's something which can happen more often than you'd expect and could cause many problems, how has this become acceptable at this point in time?

Because atx3.0 is new and no one has any experience with the PSUs

 

[oxidized]

Because atx3.0 is new and no one has any experience with the PSUs

I meant the fact that such GPUs can have these power draws randomly.

 

[Valantar]

How has this not caused a giant controversy? I mean, it sounds to me like it's something which can happen more often than you'd expect and could cause many problems, how has this become acceptable at this point in time?

It has, to some extent. The problem is, unless you have an expensive oscilloscope and the knowledge required to use it, testing for this is impossible. And to complicate things further, transient spikes don't just depend on the GPU die in question, don't just depend on the specific AIC board design that it sits in, but even depends on the motherboard it's paired with. Ther are tons of variables in play, making it extremely difficult to pin-point. This GN video is the best look into this I've seen though:

 

[jonnyGURU]

I meant the fact that such GPUs can have these power draws randomly.

Most people seriously over provision their PSUs. Using 850W PSUs when a good 650W PSUs is adequate.

Also, a PSU tends to have enough capacitance in the secondary caps to "absorb" these spikes without the OCP of the supervisor IC even "seeing" that power has "spiked" beyond the rating of the PSU.

 

[R-T-B]

2x is normal. 2.5x is Ampere. They say the new Lovelace is 3x.

View attachment 261404

Wait... didn't you just get done saying 2x was "excessive" for the standard, now you are saying it's expected lovelace will have 3x excursions? How do you rationalize that, or am I misunderstanding?

Just to be clear, I know you are the expert in the room, just trying to understand.

 

[jonnyGURU]

Wait... didn't you just get done saying 2x was "excessive" for the standard, now you are saying it's expected lovelace will have 3x excursions? How do you rationalize that, or am I misunderstanding?

Just to be clear, I know you are the expert in the room, just trying to understand.

You're misunderstanding.

The 3x excursion is for just the GPU. So, if you have a 450W Lovelace card, they rumor that there will be 1350W power excursions for a very short period of time.

ATX 3.0's test procedure says that the PSU should be able to handle 2x it's total capacity. In other words, an 850W PSU should be able to handle a 1700W power excursion.

Keep in mind: ATX 3.0 and PCIe 5.0 are two totally separate things.

 

[R-T-B]

You're misunderstanding.

The 3x excursion is for just the GPU. So, if you have a 450W Lovelace card, they rumor that there will be 1350W power excursions for a very short period of time.

ATX 3.0's test procedure says that the PSU should be able to handle 2x it's total capacity. In other words, an 850W PSU should be able to handle a 1700W power excursion.

Keep in mind: ATX 3.0 and PCIe 5.0 are two totally separate things.

Ah, I follow now, thanks!

 

[Blaylock]

You're misunderstanding.

The 3x excursion is for just the GPU. So, if you have a 450W Lovelace card, they rumor that there will be 1350W power excursions for a very short period of time.

ATX 3.0's test procedure says that the PSU should be able to handle 2x it's total capacity. In other words, an 850W PSU should be able to handle a 1700W power excursion.

Keep in mind: ATX 3.0 and PCIe 5.0 are two totally separate things.

So, putting 2 and 2 together, there are 12HPWR cables designed for 600W. Assuming a GPU is ever designed to utilize 600W (or even close) then the new ATX 3.0 standard could potentially pull three times that or 1800W for a short time. With the other PC components connected to the PSU this could or would be enough to trip a standard 15 Amp circuit breaker (for those of us on 120v grids). This seems more than just excessive, it seems absurd.

Are circuit breakers designed to allow short burst power excursions or do they trip once the peak draw reaches the set limit?

 

[Valantar]

So, putting 2 and 2 together, there are 12HPWR cables designed for 600W. Assuming a GPU is ever designed to utilize 600W (or even close) then the new ATX 3.0 standard could potentially pull three times that or 1800W for a short time. With the other PC components connected to the PSU this could or would be enough to trip a standard 15 Amp circuit breaker (for those of us on 120v grids). This seems more than just excessive, it seems absurd.

Are circuit breakers designed to allow short burst power excursions or do they trip once the peak draw reaches the set limit?

This isn't a worry - the spike won't be long enough to trip a breaker even if it actually reaches the AC side in a noticeable way (rather than being spread out as a lower, slower load by the depletion and subsequent recharging of capacitors in the PSU). Remember, even the 450W PSU TPU recently reviewed had an inrush current of 82A! That's on the high side, but still perfectly fine simply because of its short duration. Domestic breakers just don't trip that quickly.

 

[jonnyGURU]

So, putting 2 and 2 together, there are 12HPWR cables designed for 600W. Assuming a GPU is ever designed to utilize 600W (or even close) then the new ATX 3.0 standard could potentially pull three times that or 1800W for a short time. With the other PC components connected to the PSU this could or would be enough to trip a standard 15 Amp circuit breaker (for those of us on 120v grids). This seems more than just excessive, it seems absurd.

Are circuit breakers designed to allow short burst power excursions or do they trip once the peak draw reaches the set limit?

Exactly what Valantar said above. The spike won't even make it through the main transformer of the PSU.

 

[R-T-B]

With the other PC components connected to the PSU this could or would be enough to trip a standard 15 Amp circuit breaker (for those of us on 120v grids).

15amp breakers are not that fast acting.

And many badly made thermal based ones, like my old Federal Pacific box, trip randomly or not at all depending on ambient (really need to replace that POS)

 

[Stoff3X]

There is zero mention of ATX 3.0 compliance at Lian LI:

SP850

<ul> <li aria-level="1">80 PLUS GOLD Efficiency</li> <li aria-level="1">Fully modular braided cables for a clean look</li> <li aria-level="1">High durability Japanese electrolytic capacitors</li> <li aria-level="1">ZERO RPM mode under 40% load for silent operation</li> <li...

lian-li.com

Also: "A PCIe 5.0 12-pin (12VHPWR connector cable without sideband contact) is included", which afaik is only relevant to rtx 30 series, and not future gen 5 GPUs, as ATX 3.0 requires at least 2 of the 4 sideband pins: https://edc.intel.com/content/www/u...-auxiliary-power-connectors-sideband-signals/

 

[jonnyGURU]

There is zero mention of ATX 3.0 compliance at Lian LI:

SP850

<ul> <li aria-level="1">80 PLUS GOLD Efficiency</li> <li aria-level="1">Fully modular braided cables for a clean look</li> <li aria-level="1">High durability Japanese electrolytic capacitors</li> <li aria-level="1">ZERO RPM mode under 40% load for silent operation</li> <li...

lian-li.com

Also: "A PCIe 5.0 12-pin (12VHPWR connector cable without sideband contact) is included", which afaik is only relevant to rtx 30 series, and not future gen 5 GPUs, as ATX 3.0 requires at least 2 of the 4 sideband pins: https://edc.intel.com/content/www/u...-auxiliary-power-connectors-sideband-signals/

Nice catch!

Actually.... they can still use that +12VHPWR connector, but WITHOUT sidebands, it's only rated for 150W.

 

[Stoff3X]

Nice catch!

Actually.... they can still use that +12VHPWR connector, but WITHOUT sidebands, it's only rated for 150W.
View attachment 261897

I reckon tpu should edit the article...
I returned the sp850 as I see no benefit to change PSU until we can get a proper SFX ATX 3.0 unit.

 

[Valantar]

I reckon tpu should edit the article...
I returned the sp850 as I see no benefit to change PSU until we can get a proper SFX ATX 3.0 unit.

Seems like Cooler Master answered your prayers.