cancel
Showing results for 
Search instead for 
Did you mean: 

General Discussions

ramchyld
Elite

Why only one X399 motherboard with three x16 PCIe slots?

So this has been bugging me for a while now. I've been looking for a X399 motherboard with three PCIe 3.0 slots that function at full x16 speeds. However, only the Asrock Taichi X399M has such a layout, all the others are x16/x16/x8/x8. also need one PCIe x1 2.0 slot which the Taichi X399M doesn't have by virtue of being a mATX board and thus has to skimp on slots.

So yeah. Why aren't there any boards out there that offer only three slots albeit with all three functioning at x16 speeds aside from that one Asrock offering? I think there are valid scenarios where having all three PCIe 3 slots at 16x is useful (ie first two slots used to drive GPUs in CFX mode, third slot running a separate, standalone GPU driving a 4K touch panel, or third slot running four NVMe drives in a 4x4 adapter), plus the one PCIe 2 x1 slot for a SoundBlaster.

0 Likes
12 Replies

Because of the bandwidth provided by PCIe 3.0, and the fact that high end gaming cards are not bottlenecked by x8. As you can see in the benchmarks done by TechPowerup, there is no performance difference between PCIe 3.0 x8 and x16 in gaming. As for NVMe storage, the performance benefit in home applications is nonexistant. While NVMe trumps SATA in benchmarks, unless you're running a server dealing with dozens of VMs or database handling dozens of simultaneous requests for different data, you're not any worse off using an SATA SSD, as shown by TomsHardware, because you're not getting into the ultra high queue depths and taking advantage of the much increased I/O capabilities of the drives.

https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080_PCI_Express_Scaling/

https://www.tomshardware.com/reviews/samsung-sm951-nvme-versus-ahci-sata,4137.html#p1

This is their test with the largest difference, 10 seconds over 6 minutes, 3% difference.

image021

0 Likes

I was looking for the same thing. Thanks

0 Likes

Well, the thing is, I feel that users will be better served by providing only three x16 slots. I believe that by going with only three slots on an ATX design, boards can be designed with a one-card gap between each video card allowing them to breathe better (the X1 slots can always be pulled out using a riser cable).

Here's a layout simulation I created, using a Taichi X399 as a base design:

x399_3slot.png

Furthermore, money is saved from omitting one PCIe slot, meaning the motherboard can be produced (maybe slightly) cheaper.

And well, 4K content I think hit hard enough on the NVMe loads. When 8K goes mainstream in 2020 (Japan plans to transmit the Olympics in 8K, which I assume is when 8K will go mainstream), content creators can really benefit from RAIDing multiple NVMe disks.

Also, there's the intent here to use higher-end cards for rendering or AI research, which will probably take better advantage of the lanes.

TBH I have no intention to game on a TR machine, I accept that it's lousy for gaming due to the segmented memory design and it's for content creation work like video rendering.

0 Likes

The new TR processors have 60 lanes so should be enough for 2 dual GPU cards and a RAID card

0 Likes

Well, It has enough lanes for three NVMe sticks, and three x16 slots. But yeah. For some reason most motherboard makers prefer to divide one of the x16 into two x8s. Honestly, I'd prefer three x16 slots after what happened last Tuesday. I have four RX570s on my X399 board for render farm purposes, and the board in question is currently being RMAed because apparently something overheated and burnt, I heard arcing and there's this strange smell, and then the workstation shut off, and after that any GPUs plugged into the third slot now have an erratic fan behavior in which the fan spins up at full speed for a second every few seconds. Incident got me thinking that I would be better served if there is only three slots.

0 Likes

Rendering is pretty demanding and it may be better to use 2 cards per box and use a couple of machines, this way the heat will be managble

I have done lots of video over the years and it can be brutal on a processor and GPU

0 Likes

I agree, but still, if you look at the current situation:

- Almost all NVMe-to-PCIe risers need a x16 link (capable of dividing into 4x4)

- Futureproofing is a good idea.

- 3 GPUs wouldn't generate as much heat, especially with the layout I conceived which gives each card a one-card breathing space (except maybe the last card depending on chassis design). I've actually have a three 470 gaming rig setup and that rig has absolutely no problems whatsoever - I have run GTA V in tri-CFX mode on that rig several hours straight - compared to this four 570 workstation, where something on the motherboard promptly overheated while doing a test 1080p50 render. I want to knock it down to three 570, but I'd like a motherboard that offers three x16 instead of two x16 and two x8.

0 Likes

reason I stick to dual GPU, mostly due to thermal load problems which motherboards are not well designed to tolerate

a pair of RX Vega 64 cards would render lots of work assuming the tools can handle the modern card effectively

0 Likes

Well, the thing is, Vega 64 is expensive. A Vega 64 is going to cost me more than the four RX 570s.

Also, if I use the Vega 64s, then there's still the two x8 slots unaccounted for. Again, I stand that it's better as a single x16 slot since most PCIe-to-NVMe adapters takes four drives and work best with a single x16 slot (and one that can split it's IO to four x4s at that). So my idea is that the third slot runs at x16 as well, but can split into 4x4 via a BIOS toggle (what the third slot can do at the moment), while the first two slots run exclusively at x16.

0 Likes

On the X399a with 60 lanes the board has enough to afford some U.2 ports for extreme SSD products but so far nobody is doing that. Yo....
  

0 Likes

U.2 is a dead standard, no one uses it afaik. However, I am running three NVMe sticks on the mobo atm. In RAID-0. Wanted to use RAID-5 but AMD doesn’t provide RAID-5 as an option...

That said, there are many other legitimate uses for multiple GPUs. Vega cards are expensive at here I am, A typical Vega 64 card costs over four thousand Malaysian ringgits, and I’m sure I don’t need a Vega 64 for rendering 1080p50 workload. However multiple low end GPUs are useful with video editors since it can split up rendering workload among the cards. For example, if the video is made up of several hundred smaller clips, the workload can be split up for parallelism.

On my current FX8320e Machine, a 20 minute video takes over an hour to render, which is unacceptable.

0 Likes

The Intel 750 SSD in the 15mm 2.5" format are available but you need an adapter for M.2 or a PCI Express card with U.2 sockets on it.

0 Likes