Functional 4-Way SLi PASCAL Titan X

4.a if you plan on doing 4 way sli, you CAN use the standard v2 bridge, however, if you observe the 4 way sli bridges you'll notice that, the main card only gets 1 connection, the pattern is 1,2,1,2. In order to get the bandwidth on 4-way, you need a 2,2,2,2 config

Agree,if i not make a mistake,it is the why you have build your bridge and obtain more communication bandwidth betwen SLI bridge.

 

4.b it's not enough to make the ribbon bridges with v2 connectors, there are some resistors that must connect to certain pins to enable the driver to get rid of the high bandwidth warning, and get the last bit of scaling.

Ok, we will see the details.

 

5. OS, bios -tweaks. You must have a real understanding of the OS, windows in this case. There are registry settings that can make a world of difference. There confit and setup requirements that can also give you a substantial increase in results and stability. I'll answer this again. windows 10 is the slowest os for me and provides the least amount of benefits and control. My personal choice in order are win server 2012r2, windows 7 64, ubuntu 17.04, ubuntu 16, windows 8.1 64 and windows xp 64/ seven 2003 64. I've tried repeatedly server 16 and find it highly unstable and unreliable and slower than 2012r2, 7-64, 8.1- 64, and so forth.

I was thinking to use windows 8.1 64. Ok about Regedit and Windows Registry.
Yes, i am very limited by Win10. Win10 change some thing all time and one day we find decrease performance.

Thanks

 

Hi,
Cuda benefit of your SLI. It was the first benefit i wanted to know.
Thanks very much

now, the system permit me private mesage...

 

No no, Z10PE-D8 WS not work at the same:
ASUS-Z10PE-D8-WS Quad SLI multi-graphics card setups only support x8/x8/x8/x8 PCIe configuration
http://news.softpedia.com/news/The-...PE-D8-WS-Dual-Socket-Motherboard-459168.shtml
Expansion slots : 4 x PCI Express® 3.0 x16 slots (dual at x16/x16; quad at x8/x8/x8/x8)
SPECIFICATIONS[1]
[1] Specifications, content and product availability are all subject to change without notice and may differ from country to country. Actual performance may vary depending on applications, usage, environment and other factors.
https://www.asus.com/News/ChCMUvu3bWfJTIfq/
quad at x8/x8/x8/x8 = 32 Pcie lanes only
ASUS-Z10PE-D8-WS less performance than mono CPU with X99-E WS... you are as like 2 video SLI cards performance maximum. forget ASUS-Z10PE-D8-WS. For SLI performance, it is better to keep Z9PE-D8-WS

Difference betwen GTX for Render, i use Adobe for to Know the best GTX running with full Render function. after First Titan, (and Titan Black) you lose some Cuda render compatibility function, Double precision... for me there are 1 way, water cooling for first Titan for not lose few day of job. Official GTX List compatibility:
https://helpx.adobe.com/after-effects/system-requirements.html
For me GTX Titan are the best Cuda compatibility with lot of render. But, there are lot of render application with GPU only in CC 2017. I was hope to use CC2014, but there are not so much render by GPU. The minimum are CC 2015… CC2017 are better but with lot of other performance problem
https://blogs.adobe.com/creativecloud/after-effects-cc-2015-3-in-depth-gpu-accelerated-effects/
https://en.wikipedia.org/wiki/Adobe_After_Effects
First Titan are semi-professional cards, run better than professional, you have lot of benefit by this cards, see Double precision :
https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units

the new GTX video cards have high performance but lose compatibility with some Cuda and video render for separate professional and video game cards.

Quadro and other with SLI, run with some mother board and only with full professionnal hardware and software $$$$$

Out of first Titan, check the compatibility with yours render software function VS GTX, and 2 or 4 limited SLI first. Good luck

You want new 4 SLI...SuperMicro X10DRG-Q
CPU?
http://ppbm8.com/CPU/CPU.html
you will obtain in render +40% minimum, possible +90% than your dual 2687w V1... but with incredible more expensive.

For only video render, you need to chose betwen 4 true way SLI with SuperMicro X10DRG-Q, or 2 SLI with only old 4 core CPU , do you like i7 6700K overclocked ? or I7 6950x?:
http://ppbm7.com/index.php/tweakers-page/95-single-or-dual-cpu/109-single-or-dual-cpu
https://www.pugetsystems.com/labs/articles/Adobe-After-Effects-CC-2015-3-CPU-Comparison-847/
https://www.pugetsystems.com/labs/articles/Adobe-After-Effects-CC-2015-3-Multi-Core-Performance-843/
https://www.pugetsystems.com/labs/articles/Adobe-After-Effects-CC-2015-Multi-Core-Performance-714/
https://www.pugetsystems.com/labs/a...o-CC-2015-Multi-Core-Performance-Update1-806/
https://www.pugetsystems.com/labs/articles/Should-you-use-a-Dual-Xeon-for-Premiere-Pro-CC-2017-932/

I use Adobe for help to chose. some performance change with other software... it is high frequency first... number of core it is after, and need to chose 4 SLI from Dual Xeon or 2 SLI from mono CPU 4 or 6 core but overclocked... Xeon not accept overclocking as like i7.

Best regards

 

I sent a few PMs out, please view


as for the difference in mobo, yes, the x10drg-q does end up being a more expensive build fore that

I did start off with the z10ped8-ws, then proceeded to the x10drg-q

 

Venturi, what you've achieved here is mightily impressive, and I have to ask if you're willing to share screenshots of your Inspector profiles with me.

It appears I'm going to need 15 posts before being able to PM anyone, but I'm guessing you can PM me, right? If so, can you send screenshots of your settings for every game you've got 4way-SLI working on? I realise this is asking a lot, but I want to be sure that I have all the relevant information before I take this very expensive plunge.

I'm planning to ape your exact setup, at least as far as the mobo, cpus and gpus are concerned, so let me know if it's possible to send the screenshots via PM or other private channels. I have secondary email addresses I'm willing to share publicly, if that would be simpler than PM.

 

Venturi, A ton of thanks for making this happen. I now have a functional 3 WAY SLI System based on 3 x Geforce 1080 ti's. Its not fully complete yet but I am seeing good scaling. Cant post images or links yet as this is my first post. :-(

I have so far had success with

GAMES:
Rise of the Tomb Raider (DX 12 Multi GPU Patch)
Crysis 3
Mass Effect Andromeda (Turn off AA)
Witcher 3 Wild Hunt (Turn of AA)
Metro Last Night Redux (Medium Gains)
Far Cry 4
Far Cry Primal
Deus X Mankind Divided
Borderlands 2 (Medium gains)

Benchmarks:
Firestrike Ultra
Furmark
Unigine Heavens
Unigine Superposition (Custom SLI Profile)

Some games failed to scale well or did not scale at all, had issues
DOOM 2016

As I write, I am awaiting the MSI V2 3 WAY Bridge

The above scores are from flex ribbon bridges I currently run.

Will update soon on further progress.

 

I have sent out the relevant PMs

thank you always

v

 

Hi email me at addiecool at gmail for Witcher 3. Cant post PM's yet

 

Hello Venturi,

I'm having a lot of Trouble with my 3 Way Sli Titan x Pascal Setup. I just don't get all Cards to work to 100%, even in 3d Mark. This is my Rig:

Intel 7900x
Gigabyte X299 gaming 9 mobo
32 GB 3800Mhz DDR 4
3 Way SLI Titan x Pascal
Win 10

Because of the Slot space of 60 + 40 mm i only found an Asrock Sli Bridge with plastic connectors.

Although I Setup all Settings in Nvidia Inspector for the Benchmark run I found in the Forum, the max. usage across all cards is 70-80 %. In Timespy I managed only 23000 Point of Graphics score.

I would be very very Happy if you could give me some tips.

Please contact me by PM, I haven't got enough Posts.

Greetings from Germany,

Danny

 

JaxMacFL has rock star paralysis and can not comment properly at the moment!!!

Nice rig dbb!!!!!


OBTW Ven, you are untouchable!!!!

 

Thank you sir


I'll be posting a few how too docs on the setup

 

How do you get GPU usage to 99%?. I'm running 3x 1080tis and only get 70% on wild lands. Thx

 

PM sent

 

Thanks Jay. Just sent you email. Cant send PM yet .

 

This is just great think ima give this shizzle a blast with my xeon 20 core 40 thread rig. Time to have fun god i love modding and you lot at 3dguru , always feels like a tweaking a ferrari

 

Hello Venturi, SIr.

It seems like you have become quite a legend since I last viewed this thread, congratulations on being (arguably) the central individual responsible for making extreme SLI (3 to 4 cards) seem like such a viable option for many consumers around the world in the present and future!

Sadly due to current obligations I am stuck with the same "rig" (if you can call it that) that I've had for around 5 years, still waiting for the appropriate time to upgrade (in my current situation the justification for investing in a new configuration would be sketchy at best, with increasingly powerful GPUs and undisclosed MOBO configurations looming on the horizon. Plus, I do not play PC games anymore ).

Reading through this forum has left me with peak levels of curiosity and many questions which I hope that you have the time to answer, they are as follows:



1. What are the benefits of the dual Xeon (44 core) Quad SLI setup over a Core-i9 (12 core) Quad SLI setup? Are there any advantages to fewer cores besides speed?

2. With what you've seen, how does the speed of individual cores affect the end result in average frames per second and overall stability?

3. On average how would you say a system with fewer cores and limited resources per GPU scales against a system with more cores and sufficient resource allocation between multiple GPUs? Do more cores and available resources allow each GPU to function more consistently when it comes to efficient scaling? What percentages would represent the scaling curve between a well configured single and dual core system.

For example ( in GPU utilization) if A=dual socket MOBO with 44 medium-speed cores, and B=single socket MOBO with 12 high speed cores. Is A more like 96%, 95%, 94%, 93%, a consistently smaller performance decrease overall each GPU in the system? While B is more like 95%, 90%, 85%, 80%, a consistently larger decrease in efficiency over all GPUs in the system?

4. What are the 4-way SLI connectors you would recommend? Would any that have metal connectors be sufficient?

5. Are you still completing your How to guide or is it finished? And could I know where to find it when it is (or if it is already) done?



Sorry if some of my questions seem a bit redundant, if you could answer them I'd greatly appreciate it.

But thank you for everything you have already answered! Without you there would be no true high frame rate 4k or 8k gaming videos out there at all!

 

Hi.

The ASUS-Z10PE-D8-WS has the ability to use 80 PCI-E express lanes and fully supports quad SLI. So it can run x16/x16/x16/x16 (64 PCI-E lanes), dividing the resources of each Xeon, with 36 lanes remaining.

Was your post about quad SLI support and PCI-E lanes, or did I mess up and misunderstand what you meant?

 

The amount of lanes is important and the asus z10pd-d8 can do a decent job (provided its version 1.05 or above and has the v3-v4 fix, more on this if you are using the asus board)

the allocation of which lanes go to each socket is important as well as in the bios how that is set up (those bios shots are also in this thread)

with the supermicro board I'm using the pch lanes are separate from the count, so pose an advantage, in addition on the super micro board, the primary first slot for video is the LOWEST on a vertically mounted board so the primary card is on the outside and not closest to the cpus. This make a huge thermal advantage

lastly, the super micro board has double spaced pci slots, so one can have quad sli and still be able to use the other slots. On the asus, the slots between the card pci are covered by the dual space cards so the slots are no longer accessible.


the asus has an advantage that it has an nvme slot on the board, however, the super micro having the other free slots available can host an 8x name adapter easily.



to answer the prior question, more cores (putting aside the turbo function which on a v4 2699 goes to 3.6ghz on the first 4 cores, then 3.5, then 3.4, etc) across all lanes can feed the GPUs what they need, also it provides the necessary components for the grid. The new gx1 design from nvidia that hit the market, uses in some degree the same methodology.

So there is a positive difference in quad sli with 44 cores at 2.3ghz/3.6ghz, vs a set of 4.2ghz chips while one is fully utilizing 90%+ per card in quad sli.


truth is more complicated:

IMPORTANT

Cache on Die, Early snoop, numa, etc have to also be implemented correctly in bios.
you MUST have MMIOHBase set up correctly as well as MMIO High Size also set up in bios, this will make or break a successful build and some of those settings are not functional in the asus build





on the bridge:
minimum
v2 metalclips minimum toget 4 way working (6 connections)

better
make your own with resistors and clips to get 8 connections working
the primary card has 2 connections that way, emulating HB/ 8x/ 4 way