OTOY Forums

Hi Guys!

Has any of you with 1.x extenders (cheap USB ones and/or those expensive Amefltec) experienced slowdowns/issues on Octane 3.03xx ?
I've tested a lot last days and as soon as i plug in my 1x USB riser card my GPU on riser card is 30-35% slower than same GPU directly on Motherbard. When i plug 3 GPUs on MB + 4th GPU on PCI-E 1x extender all GPUs slowdown (less OB score/points) ? Octane 3.03xx just uses less GPU load than it usually does with 3 GPUs or even 2 GPUs ?

My GTX 1080 at Motherboard gives OB score 125-128 and same GPU at PCI-E extender gives 81-85 points ???

Thanks

ertaro wrote:I maked these!

wow... 1283! ... amazing... isn't that great to see the OctaneBench graphic change and change... no standing still in this community and technology.
Ertaro, you are the Neil Armstrong of this moment

rappet wrote:

ertaro wrote:I maked these!

wow... 1283! ... amazing... isn't that great to see the OctaneBench graphic change and change... no standing still in this community and technology.
Ertaro, you are the Neil Armstrong of this moment

lol, briefly, unfortunately)

Lewis wrote:Hi Guys!

Has any of you with 1.x extenders (cheap USB ones and/or those expensive Amefltec) experienced slowdowns/issues on Octane 3.03xx ?
I've tested a lot last days and as soon as i plug in my 1x USB riser card my GPU on riser card is 30-35% slower than same GPU directly on Motherbard. When i plug 3 GPUs on MB + 4th GPU on PCI-E 1x extender all GPUs slowdown (less OB score/points) ? Octane 3.03xx just uses less GPU load than it usually does with 3 GPUs or even 2 GPUs ?

My GTX 1080 at Motherboard gives OB score 125-128 and same GPU at PCI-E extender gives 81-85 points ???

Thanks

Hi Lewis,

This is not your imagination. I don't know the mechanics why, but at least with USB risers and splitters, there is a diminishing return per GPU as you add more, despite the overall linear rise as you add more GPU.
This appears to the case with the USB risers and splitters, I am not sure how it works with straight motherboard connections.

Here below I have a more scientific analysis of the relationship between USB riser 1x vs motherboard at 16x.

Some background, my rig has 1 Titan X in case on motherboard.
Then I have 12 Titan Z (6 cards) connected via Amfeltec Clusters, powered via 2 external PSUs.
Then I also have 2 Titan X cards connected via risers, powered via a 3rd external PSU.

I can never run all 15 GPU at same time as my BIOS currently can only give me 13GPU cores at once. So I either run in one of 2 'modes':
(A) 1 Titan X mobo and 12 Titan Z Amfeltec, and
(B) 1 Titan X mobo and 2 Titan X USB riser


So, with the Titan X's I have the following 3 cards:
(1) Titan X on mobo @16X
(2) Titan X on USB 3.0 Riser @1x
(3) Titan X Pascal on USB Riser 3.0 @1x
So, I positioned a test scene where I got a 100 s/px score with the Titan X on the motherboard 16x, to benchmark, with no overclockling. Next here is the 'regular' Titan X on the USB Riser 1x (no overclock). Gave ~75-80 s/px Then here is the 'regular' Titan X on the USB Riser 1x, overclocked +150 MHz. Gave ~83 s/px Last, the Titan Titan X Pascal on the USB Riser 1x (no overclock). Gave ~100 s/px Looking at this, I see that there was a 20-25% drop for the same Titan X card on the riser. Overclocking the card +150 MHz bought me another ~5-8%.
What is interesting then is the Pascal, which is a higher powered version of the 'regular' Titan X, matched almost exactly the 'regular' card on the motherboard. So, I am thinking if I saw a 20-25% drop for the riser Titan X, likely the Pascal would be 20-25% greater if connected 16x on the motherboard (I am imagining a score of ~133).

The moral of the story is, yes, the cards will drop 25% on a USB riser, but you can either overclock it to compensate, or if you get a Pascal version of the same card, you can pretend it's a 'regular' card on the motherboard.

Thanks for confirmation, i reaylly appreciate you tested it 'coz i posted my result son separate topic (viewtopic.php?f=23&t=56062) but i got no much response form OTOY

As for overlcock i can overlcok (and i always do) but drop is still same 'coz i'm overlcocking ones on motherboard equally as on extender(s) .

A very, very wise man once said...

Re: External Graphics Cards PC
Post by Notiusweb » Tue Nov 24, 2015 9:10 am
Heading for some colder weather now, we'll probably see some higher Bench scores

Just an update, have the Pascal Titan X with the standard fan, and also an old school genuine beachwood-aged classic Titan X, both no-enclosure open-air on USB risers, and the Pascal not only runs faster (see my earlier comparing post), but also cooler.

For example, full-render with MSI Afterburner OC, I am hitting ~74 on Old Titan X and ~64 on Pascal. I do not know why this is, but I can say that the Titan Z always would hit ~83-84 at full-render OC, and even would crash/disconnect at long renders... So the Pascal is both faster and more heat efficient, which is pretty cool....
GET IT??? "PRETTY COOL"...GET IT!!!???

Notiusweb wrote:A very, very wise man once said...

Re: External Graphics Cards PC
Post by Notiusweb » Tue Nov 24, 2015 9:10 am
Heading for some colder weather now, we'll probably see some higher Bench scores

In Australia and New Zealand they are heading for summer
But I catch your drift

rappet wrote:

Notiusweb wrote:A very, very wise man once said...

Re: External Graphics Cards PC
Post by Notiusweb » Tue Nov 24, 2015 9:10 am
Heading for some colder weather now, we'll probably see some higher Bench scores

In Australia and New Zealand they are heading for summer
But I catch your drift

How hot in the summer can it get in the Land Down Under or in New Zealand?

Tutor wrote:

ertaro wrote:Hi guys.
Help me, please.
What motherboard I need for 8 external GPU and 2 internal? I use amfeltec clusters.

Thanks.

Sorry for delay. Here's what Amfeltec told me:

"The motherboard limitation is for all general purpose motherboards. Some vendors like ASUS supports maximum 7 GPUs, some can support 8. All GPUs requesting IO space in the limited low 640K RAM. The motherboard BIOS allocated IO space first for the on motherboard peripheral and then the space that left can be allocated for GPUs. To be able support 7-8 GPUs on the general purpose motherboard sometimes requested disable extra peripherals to free up more IO space for GPUs. The server type motherboards like Super Micro (for example X9DRX+-F) can support 12-13 GPUs in dual CPU configuration. It is possible because Super Micro use on motherboard peripheral that doesn’t request IO space."

I have three Supermicro X9DRX+-Fs [ https://www.supermicro.com/products/mot ... drx_-f.cfm & http://www.superbiiz.com/detail.php?name=MB-X9DRXFB ] based systems. The second to last sentence in the above quote (that contains "12-13 GPUs") applies to an X9DRX+-F system running Windows. Linus OSes, like Linux Mint, support even more GPUs with the X9DRX (or X10DRX). Whether the GPUs are internally or externally placed doesn't really significantly affect the IO space limitation, except that it is possible to place more GPUs internally in a system with a LIan LI D8000 case with an X9DRX motherboard.

@ Tutor: I would appreciate it very much if you could provide some information regarding the gpus that your are using. I am thinking about ordering that mobo but frightened of incompatibilities. Do you think I will be able using my ordinary AMD gaming gpus for it?
Sorry for low level question, new to server hardware.

Hi all!

I tried to read and get as mouch info as possible in this treadh.

A few qusestions, is it worth the hassle witth external Gpu cluster, i have 4 1080 ti hybrids in the computer.
And when i upgraded i got 4 1070 hybrid cards over.
So im trying to find out what to do, get like a amfeltec and run some cards that way.
Or maybe something like Tom glimps old build with a special custom case and run the cards on risers.
But then i have the issue with power, have a evga 1600w supernova(Gues that want be enough)

So any thoughts on that?