Folding on different PCIe slots

[WhatComesAround]

Will folding on a x1 PCIe slot or x4 PCIe slot make a difference in the performance of a card?

[GDRRiley]

not really.

[FloRolf]

no, it doesn't have to Transfer massive data, it just calculates and sends it back.

[Imakuni]

9 hours ago, GDRRiley said:

not really.

6 hours ago, FloRolf said:

no, it doesn't have to Transfer massive data, it just calculates and sends it back.

Oh yeah?

 

Then please, I would love to see recent test results to back up those claims. Please, if you guys actually have a link for such tests, I'm interested in getting them, for I just can't find that no matter how much I search.

 

On the other hand, though, the most recent test I found (which was during core 15/17 era) showed the exact opposite. The guy grabbed a Gtx 970, the latest card at the time, and showed evidence that there was a little loss in performance at 4x and a MASSIVE loss at 1x.

 

@WhatComesAround, this has been a topic for debate ever since the dawn of times. According to @whaler_99, people did testing any years ago and found no difference. That's probably the source of the 2 responses you've seen above me. But both GPUs and software have changed a lot since then, and (as I've said) there have been reports of performance losses. The fact that a free CPU core is advised to feed GPUs only helps that claim

 

So who's right here? I honestly have no idea. Again, it's back to that lack of testing I mentioned.

 

For peace of mind, though, I would avoid folding on 4x if your GPU is a powerful one. And never fold on 1x, unless it's a really potato one.

[WhatComesAround]

1 hour ago, Imakuni said:

Oh yeah?

 

Then please, I would love to see recent test results to back up those claims. Please, if you guys actually have a link for such tests, I'm interested in getting them, for I just can't find that no matter how much I search.

 

On the other hand, though, the most recent test I found (which was during core 15/17 era) showed the exact opposite. The guy grabbed a Gtx 970, the latest card at the time, and showed evidence that there was a little loss in performance at 4x and a MASSIVE loss at 1x.

 

@WhatComesAround, this has been a topic for debate ever since the dawn of times. According to @whaler_99, people did testing any years ago and found no difference. That's probably the source of the 2 responses you've seen above me. But both GPUs and software have changed a lot since then, and (as I've said) there have been reports of performance losses. The fact that a free CPU core is advised to feed GPUs only helps that claim

 

So who's right here? I honestly have no idea. Again, it's back to that lack of testing I mentioned.

 

For peace of mind, though, I would avoid folding on 4x if your GPU is a powerful one. And never fold on 1x, unless it's a really potato one.

Thank you very much for answering so many of my recent questions, you are really helping me out greatly.

[FloRolf]

1 hour ago, Imakuni said:

Oh yeah?

 

Then please, I would love to see recent test results to back up those claims. Please, if you guys actually have a link for such tests, I'm interested in getting them, for I just can't find that no matter how much I search.

 

On the other hand, though, the most recent test I found (which was during core 15/17 era) showed the exact opposite. The guy grabbed a Gtx 970, the latest card at the time, and showed evidence that there was a little loss in performance at 4x and a MASSIVE loss at 1x.

 

@WhatComesAround, this has been a topic for debate ever since the dawn of times. According to @whaler_99, people did testing any years ago and found no difference. That's probably the source of the 2 responses you've seen above me. But both GPUs and software have changed a lot since then, and (as I've said) there have been reports of performance losses. The fact that a free CPU core is advised to feed GPUs only helps that claim

 

So who's right here? I honestly have no idea. Again, it's back to that lack of testing I mentioned.

 

For peace of mind, though, I would avoid folding on 4x if your GPU is a powerful one. And never fold on 1x, unless it's a really potato one.

Well people also do mining on 1x risers. I think that maybe in the tests you are talking about they didn't have a molex power included in the riser. Because obviously 1x doesn't deliver all the power a 16x could, which is a massive 75watts.

Additionally the argument with one free cpu core is ONLY for Nvidia gpus. Amd does NOT need a free core per gpu. 

 

I haven't done much research but I will so be can have peace. Sadly I can't really test this right now. 

[Teddy07]

It will reduce your PPD around 5-8% if you use a pcie 2,0 isntead of pcie 3.0

[Imakuni]

18 minutes ago, FloRolf said:

Well people also do mining on 1x risers.

Pointless. Mining is mining, folding is folding. They 2 are completely different things. Just as an example I'm familiar with, geneferocl (for prime number testing) can go from next to no GPU usage with full CPU to screen-lag GPU load with high vRAM abuse and next to no CPU just by changing the type of number being tested, not even changing the processing algorithm.

18 minutes ago, FloRolf said:

I think that maybe in the tests you are talking about they didn't have a molex power included in the riser. Because obviously 1x doesn't deliver all the power a 16x could, which is a massive 75watts.

Quick explanation about how PCIe works:

The first part of the slot is ONLY for power, while the second is ONLY for data.

 

Now, you notice how regardless of the size (1x or 16x), the first part is exactly the same? Well, that's because it really is! Assuming the slot is following the spec, it WILL be able to provide 75w, no matter if it's 1x or 16x. In fact, it can provide even more than that, as seeing by the RX 480 no blowing up mobos with it's over 75w PCIe requirement.

 

Speaking of which, the 970 pulls less than 75w (when OCed) anyway:

 

The reason why risers tend to have molex connectors attached to them is not that the slot can't provide enough power, no. It's that motherboards aren't designed to power 6 or 7 graphics cards all at once, which is where a 1x riser would be likely to be used; to prevent the mobo from burning out, a Molex connector is added.

18 minutes ago, FloRolf said:

Additionally the argument with one free cpu core is ONLY for Nvidia gpus. Amd does NOT need a free core per gpu.

I guess I forgot to explicitly mention that, given that I had the 970 test in mind when I was writing my post. Oops.

 

But I'll question that statement for recent usage anyways, without evidence to back it up, though.

[RP]

I've tried running a lowly 750TI via a powered PCIE 1x riser, and it never passed 60k PPD. Directly in the 16x slot, it regularly hit 100k. So, there - unscientific realworld data, completely unconfirmed

 

Running Heaven benchmark actually gave a smaller performance hit - go figure, I would think that having to display textures and such would be worse. Can't remember details, but it did pull a predictable amount of power from that riser (has a kill-o-watt on it) so don't think the performance decreased was power related. It also boosted to the same levels.