AX1600i Configurable OCP through Digital Monitoring?

If the OCP could set itself higher in the event of a higher amount of current delivery, then it would defeat the purpose of the OCP being there in the first place.

The OCP needs to be a hard cut. You can put in a delay that, let's say, cuts the power if > 40A for 20ms vs. > 60A for 10ms. But anything more than that would serve no purpose.

There is no issue with using the 2080 Ti or Vega 64 with OCP "on" if the user uses two separate PCIe cables.

There is absolutely no issue using a highend GPU with two separate PCI-E cable, but the issue is with one cable drawing 40A peaks. I'm thinking the solution would be to fine tune OCP based on the behaviour of the card.

 

But by your saying, I assume OCP has no relation to precise monitoring, as you imply the amount of time and current is easily configured through the supervisor IC. Does this mean no digital monitoring is required?

There is no supervisor IC. It's an MCU in the AX1600i.

OCP is a safety that prevents the PSU from delivering too much current in the event of a short. e.g. a short creates a low resistance load that draws too much power from the PSU. By allowing the PSU to adjust OCP based on normal loads, like that from the graphics card, you negate the usefulness of the OCP in the event that there's actually a short.

I have not seen adjustable OCP on any PSU. They are designed for the model in question and they are unchangeable.

 

Corsair Link can set fan speeds and change the PSU from multi rail to single rail. That is all you will be able to do.

AXi allows you to set OCP level.

Now if only Corsair could provide some better 650W models, 80 plus platinum instead of the lower quality RM650x which is only 80 plus gold

That's because once you get to the higher efficiency level products, there is no money saved by making lower wattages. You could "cost down" an HX750 to make an HX650, but the cost savings is less than $1. So what do you do? Sell the HX650 for the same price as the HX750?

Gold efficiency PSUs cost a bit less to make the cost delta to make a 650W vs. 750W is much lower. So you do it. But you run into the same problem trying to me a 550W. Cost delta is very close between 550W and 650W.

But the cost delta is greater when only trying to make a Bronze 550W vs. a Bronze 650W PSU. So you do both. And even a Bronze 450W because that costs even less.

See?

AFAIK Vega 64 do not consume more than 30A in normal conditions, it is just the peaks that is tripping the OCP even using one rail. And those peaks don't last as long as what you'd call a short. You said that you can in fact adjust both the time and the amount of current in the MCU (that acts as a supervisor IC).

 

I assume you're against my "solution", why is that precisely?

Vega 64 does not trip OCP when you use multiple PCIe cables.

Where have you seen that it does?

I didn't say it would trip using two cables, what I said was it won't trip under normal loads even when using one cable using multi rail OCP... However if an MCU+digital/precise monitoring has the ability to detect the difference between a short and power peaks, one cable, with OCP on that cable set at 35A shouldn't trip.

It sounds like we're trying to solve a problem that doesn't exist. :D

BTW: The new Intel guidelines explains using two OCP trip points, as I described earlier, where one trips when the load is higher for a shorter period of time, vs. lower for a longer period of time. But that's talking about when trying to adhere to the 20A OCP specification.

We've explored doing this for 40A OCP, but abandoned it because it served no purpose.

Hmm... Basically what I was referring to is this picture:

 

 

Here it is shown how Vega 64 consumes power. It hits above 30A in a couple of seconds, but it does so in relatively short bursts/period of time. It's way shorter than what a short would be. This is what I know to be causing tripping/shut down issues when using one cable w/ multi-rail OCP on a high-end GPU. The thought is that if you have a digital MCU, you could differentiate between the two and use random cables without tripping.

 

Yes. I am aware of this. Every time a new graphics card comes out, we test all of our PSUs with them. The duration of the spike in power is too fast to trip OCP.

That's very interesting, but I can't find the details of that. Does Intel account for the power peaks I mentioned above? In this document, there is no mention of the standard you are talking about, however.

 

https://www.intel.com/content/dam/www/public/us/en/documents/guides/power-supply-design-guide-june.pdf

Hmm.. Maybe it's a draft and not public yet. That version is pretty old.

But, it's like I said.. It essentially says that if the current exceeds 20A for a shorter period, the PSU shouldn't latch off. The 20A number is a UL thing.

https://www.hardwaresecrets.com/everything-you-need-to-know-about-power-supplies/12/

Why is that exactly? Does an OCP that accounts short GPU peaks means that it has to at the same time can't differentiate between that and a normal short?

Like I said, because the current has to exceed the OCP trip point for longer than 15 to 30ms.

OK, so that clears things up. I still have some extra questions. If it isn't the short power bursts that trips the OCP when using one cable, what actually trips the OCP? Is it because cards like Vega 64 / Radeon 7 like to generally consume more than 30A (35-40A to be precise, as multi-rail OCP is usually set at this range) from the PCI-E 12V cable? Or is there something else that comes into play?

The MCU or supervisor IC is looking for one of two conditions, or both.

Either the power spike is too high regardless of how fast it is, or it's "just over" but so fast that it knows it's not a short.

So that's why you split the connectors across two cables. It completely eliminates the possibility that the spike will exceed the OCP for TWO connectors with their own OCP trip points.

Another factor is voltage drop. This is actually a more common problem then OCP tripping. Obviously, putting more of a load on a single connector (the connector on the PSU) is going to produce more resistance than twice as many conductors using two cables. So splitting up to two cables also prevents UVP from tripping.

I have not seen adjustable OCP on any PSU. They are designed for the model in question and they are unchangeable.

Guess what? AXi has configurable OCP.