Basics of PCIe questions

[FlashRZ]

I've got a question that really pertains to all mobos, but especially the EVGA x299 Dark I just picked up- I absolutely love it, but then I saw the user manual says that only 1 pcie port runs 16x. All the rest are either 8x or even 4x. I was originally going to use it for a benching project with 4 titan xp's in sli, but now I'm going to use a ROG 370 zenith extreme from asus for said project because I want the increased bandwidth for the cards.

There's a concept that just doesnt make sense to me about all this. If the x299 dark is supposed to be all about performance, why not have two or more pcie slots x 16? I'm wondering if anyone might know a reason this might be. Come to think of it, the most I've seen on high quality mobos with 4 PCIe slots is two 16x slots, why is that? (the only one I've ever seen with more than two is a riserless mining mobo with 8 PCI x 16 slots that can run off a celeron processor.)

Also, this is a very general question, but important for me to understand- when it says pcie x8 that means that whatever card is in that slot will only be able to create half the output as an x16 slot, right?

Thank for your insights. RZ

 

[Dolk]

Intel has several different CPUs that can be utilized with the X299 boards, Skylake X i7 & i9, and Kaby Lake X. Each one of these CPUs has a total of different PCI-e lanes that can be utilized by a system. EVGA does a good job braking down what its board can do when you have a specific processor in place.

https://www.evga.com/support/manuals/files/151-SX-E299.pdf

A Kaby Lake X would be an awful idea for trying to do Titan Xp 4-way. Your best bet is to go with an i9 Skylake-X.

 

[FlashRZ]

I’m using an i9 7820 extreme, and now that I look more closely, you are right, more lanes for me Thanks!!

 

[Johan45]

Here's the breakdown on the DARK, as you can see with 4 cards one is still x16 while the others are x8. The difference between x8 and x16 isn't huge and honestly I would'nt be surprised if it c=scores better on the Intel. The AMD CPU will be slower and typically Intel will score higher even when all clocks are the same. The PCIe lanes are determined by the CPU being used, AMD has allowed for more in TR Vs the Intel i9 CPUs, in other words you have have 20 PCIe x16 slots on a mobo but it's still limited by the CPU's capability

 

[FlashRZ]

The AMD CPU will be slower and typically Intel will score higher even when all clocks are the same.

Even if the AMD cpu is a Threadripper 1950x?

 

[Johan45]

I have never used a TR CPU but I have done a lot of benching and can say that 3D benching just works better on Intel hands down. Just take a look at the graphics score in particular here the CPUs were close in speed the 6 core intel had a slight (200 MHz) advantage over the 5.0 GHz Ryzen 8 core CPU. The scores are close because of the physics test but that's irreleant. The Intel hit 31,550 for GFX score while thw Ryzen only scored 28741 the slight difference in GPU speed isn't the difference either. It's just Intel works better for 3D. Add to that you'll get a higher base CPU speed with the Intel and it's a given that you will score better.

 

[Dolk]

Yep, Intel still has a lead on pipelining the PCI-e over AMD. However, I'm not sure of the difference when comparing GPU intensive work loads outside of 3D/CAD.