[TomsHW] PCIe 4.0 - 32GB/sec over x16, 300-500W slot power

[ShintaiDK]

Considering how long it took to see PCIe 3.0 boards you're probably looking at 2018.

And 2020+ for consumer GPUs.

PCIe 4.0 cards are not backwards compatible to 3.0.

 

[Lepton87]

What? I don't even...
No, you are the one who is wrong.
Flattening a wire does not increase its current carrying capacity.

You can both be right. What you said doesn't invalidate what he said at all. Flattening a wire doesn't change its surface area and he said that's the relevant metric when it comes to current carrying capacity. Flattening a wire changes volume not surface area. Flattening decreases volume which he said is irrelevant for current.
Traces on the motherboard can almost be as good as cables because they have excellent surface area to volume ratio, because they are very thin. Wires waste a lot of volume.

 

[Senpuu]

How am I wrong? It's a specification and simple multiplication. If the speicification calls for 300w-500w per slot, and you have three or four slots, that's hundreds of amps.

Okay. Let's look at that specification to which you refer and do that simple math, shall we?

Solomon couldn't recall the exact ceiling because member companies have proposed several options. Solomon stated that the minimum would be 300W, but the ceiling “may be 400 or 500W."

Even with the minimum 300 watts, there is more than enough power provided by the edge connector to run a GeForce GTX 1080 (reference design 180 watts) and Titan X (250 watts) without an external power source.

These figures are from the article, but it seems they're good estimates of usage at max load. But, no. You want to cherry pick the very most ludicrous, worst-case scenario of...

You don't think quad Fury GPU's draw hundreds of amps? What's the point of 1200w+ power supplies then?

4x Fury X's. The best source I can find for power usage under load in that scenario shows 1000W measured at the wall. Generously, let's give our PSU 90% efficiency. 1000W*0.9 = 900W used by the entire rig. The vast majority of that power is being consumed at 12V. If you assumed that entire load went on the gfx cards, then you're looking at 900W/12V = 75A as a quick check. There are other factors at play and different voltages being used in different components, but the vast majority of power delivered through the mobo in this scenario is done so at 12V.

But let's go a step further. Let's look at some scenario of power usage that doesn't exist. Let's take a 1600W EVGA SuperNOVA PSU and look at its max output figures. Not that you can actually reach them, but the very most power you could pull off this PSU to system components is 24A @ 3.3V, 27A @ 5V, 133.3A @ 12V, and 0.5A @ 12V. The total amps that could be running through your mobo if it was completely taxed is 184.8 (a little less actually, but I'm being generously conservative here). Again, this is for an impossible imaginary scenario with a 1600W PSU that is 100% loaded somehow. And even then we're not getting to your stated hundreds.

In sane use case scenarios that fit 99% of all users in all situations including gaming you don't even get near a singular hundred amps. Not even close.

Perhaps you would have been better served to have read the article and getting the facts before telling me I'm wrong.

So perhaps you should take your own advice.

*munch*

 

[Ajay]

It just cant be done /period 300W will melt the ATX plug, 2.0 and 3.0 also supported more than 75W it was never used because it cant be applied. /period

I said dual 24 pin connectors - not dual 24 pin ATX connectors. As mention by VirtualLarry, it would have to be a different 24 pin connector. My basic point was to have just two 24pin power connectors for all the needs of the motherboard (the second 24 pin connector would just need to be keyed differently).

 

[Senpuu]

You can both be right. What you said doesn't invalidate what he said at all. Flattening a wire doesn't change its surface area and he said that's the relevant metric when it comes to current carrying capacity. Flattening a wire changes volume not surface area. Flattening decreases volume which he said is irrelevant for current.
Traces on the motherboard can almost be as good as cables because they have excellent surface area to volume ratio, because they are very thin. Wires waste a lot of volume.

Actually, he is wrong. First of all, we're talking about DC current here where the skin effect has no relevance whatsoever. And second of all, even if we were talking about AC current, at the frequencies and wire sizes being discussed, it would again have no relevance. The higher the frequency, the smaller the skin depth. The larger the wire, the more it matters.

 

[Lepton87]

Actually, he is wrong. First of all, we're talking about DC current here where the skin effect has no relevance whatsoever. And second of all, even if we were talking about AC current, at the frequencies and wire sizes being discussed, it would again have no relevance. The higher the frequency, the smaller the skin depth. The larger the wire, the more it matters.

I was just going by what was being said.

 

[Khato]

You can both be right. What you said doesn't invalidate what he said at all. Flattening a wire doesn't change its surface area and he said that's the relevant metric when it comes to current carrying capacity. Flattening a wire changes volume not surface area. Flattening decreases volume which he said is irrelevant for current.
Traces on the motherboard can almost be as good as cables because they have excellent surface area to volume ratio, because they are very thin. Wires waste a lot of volume.

It's actually quite a bit more complicated than that. Skin effect only applies to AC current... and you want loads to appear as close to DC as possible, because a constant 1 amp draw represents less wire loss than a 2 amp draw at 50% duty cycle despite same power delivery. Hence for power transmission from the power supply to the local voltage regulator it's primarily the cross sectional area of the conductor that matters. Now once you get closer to the load that changes of course, but that's all PCB anyway.

Edit: And looks like I was beat to it, oh well.

 

[Phynaz]

Snips wall of text

500w x 4 slots = 2,000w

2,000w @ 12v = 167A

>100A = "hundreds".

But whatever, go ahead and pump that that through a ATX connector. Continue to argue if you would like, I won't see your posts anymore.

 

[Carfax83]

PCIe 4.0 cards are not backwards compatible to 3.0.

What, are you serious?!

Why isn't it backwards compatible with 3.0? If true, this will be the first time the new PCI-E isn't backwards compatible with the older versions..

 

[XavierMace]

>100A = "hundreds".

That's not how plurals work. It's less than 200, therefore it's not hundreds.

 

[Khato]

Why isn't it backwards compatible with 3.0? If true, this will be the first time the new PCI-E isn't backwards compatible with the older versions..

Sounds like it's just a matter of connector compatibility, which makes sense given the intent to support more power through the connector. (In order to support more power through the connector, they need to devote more pins to such as voltage and current per pin are fixed.)

 

[ShintaiDK]

What, are you serious?!

Why isn't it backwards compatible with 3.0? If true, this will be the first time the new PCI-E isn't backwards compatible with the older versions..

Unfortunately that's what everything points to in the spec.

 

[hrga225]

Wrong beyond belief.
Current travels on surface of conductors.That is why you have traces on boards,to increase area,so that higher current can go through.One can make flat wire(trace) 100 microns thick that can carry 1 kA and tens of kW of power if it wide enough.
So please stop talking rubish.

As other posters correctly pointed I was wrong,for some reason I forgot we are talking about DC.

My apologies to Shivansps.

 

[suklee]

That's not how plurals work. It's less than 200, therefore it's not hundreds.

^ This.

 

[nathanddrews]

That's not how plurals work. It's less than 200, therefore it's not hundreds.

^
Double this.

 

[96Firebird]

I'm not seeing a great benefit of running that much power through the PCI-E slot rather than external power connectors.

 

[Sushisamurai]

Yeah, that's close to 1000W through my MB if it was all delivered through the slot instead of cables if it was my current setup. Imagine if one power delivery part failed, I'd need to rebuy a MB and rebuild the system. Huge pain if someone was on custom liquid.

 

[Senpuu]

500w x 4 slots = 2,000w

2,000w @ 12v = 167A

>100A = "hundreds".

But whatever, go ahead and pump that that through a ATX connector. Continue to argue if you would like, I won't see your posts anymore.

Not only were you wrong before, but that little back-of-the-napkin analysis you just made is horribly mistaken as well. You cannot exceed the power of your PSU -- a point I tried to make with the worst case scenario as described in my earlier 'wall of text' that you clearly either didn't read or understand. So, while it may be true that PCIe 4.0 will be capable of delivering as much as 500W through the connection, you would need a PSU that is capable of delivering those 2000W plus whatever power is required by your other systems, as well as enough GPU horsepower to utilize those four PCIe 4.0 slots at 100%. And just so we're all aware of reality, that hardware doesn't exist and will likely never exist within the coming ecosystem surrounding this new standard. Especially as the trend is for lower power requirements per card for a litany of reasons. The most power hungry card you could name only draws an average of 360W under full load and less on a per card basis as you move to multi-card configurations. It's not just <number of cards> x <single card load power>. So to be clear, you will never see 500W through a PCIe slot.

The bullet point, since you balked at two paragraph's worth of reading previously, is that you've created a fairy tale scenario to begin with; the punchline is that it still doesn't meet your claim. But I guess when you already just "know" the answer before you apply any critical thought you can always muster up the requisite mental gymnastics to be right in every scenario. It's neat how that works.

 

[DeathReborn]

I'd have liked PCIe 4.0 to mandate 150W (200W Max) through the slot rather than 300W+.

 

[n0x1ous]

Toms updated the story. PCI SIG contacted them and said it was a mistake. Still gonna be 75w slot and the rest from external connectors

Update, 8/24/16, 2:06pm PCI-SIG reached out to tell us that the power increase for PCI Express 4.0 will come from secondary connectors and not from the slot directly. They confirmed that we were initially told incorrect information. We have redacted a short passage from our original article that stated what were originally told, which is that the slot would provide at least 300W, and added clarification:

• PCIe 3.0 max power capabilities: 75W from CEM + 225W from supplemental power connectors = 300W total
• PCIe 4.0 max power capabilities: TBD

New value “P” = 75W from CEM + (P-75)W from supplemental power connectors.

 

[VirtualLarry]

Toms updated the story. PCI SIG contacted them and said it was a mistake. Still gonna be 75w slot and the rest from external connectors

Good. Sanity.

 

[Ajay]

Toms updated the story. PCI SIG contacted them and said it was a mistake. Still gonna be 75w slot and the rest from external connectors

Huh, really surprised that it didn't get bumped up to 100. Ah well, all the noise developed in this thread over a publishing error!

 

[redzo]

Toms updated the story. PCI SIG contacted them and said it was a mistake. Still gonna be 75w slot and the rest from external connectors

Jesus

It made zero sense anyways, unless it was a political decision: PCI SIG, the new gestapo.

Quite fascinating how many forum members went with the flow on this.

I'm sorry boozer. Can't help myself, mate.

modular psus cost a hell of alot more than standard psus. it is super premium. that totally destroys the cheap chinese psus argument people were having in this thread.

your turn.

cheers!

 

[LTC8K6]

Thanks for the update. I was wondering how I was going to keep a 500W video card cool...or two of them...