Dell Precision 7820 Xeon Workstation, dual Quadro RTX 4000 SLI performance stability woes using AutoCAD 2018 Navisworks Revizto PractiCAD

I spent a whole day on trying to stabilize a Dell workstation and its ability to run 3 CAD modeling programs at the same time using two each Dell provided RTX 4000 video cards in SLI mode. Two exact systems did not work for like 2 months since they got them, specialty built, about $8k each. So here goes my Dec 3rd wasted yet challenged. Mind you, the entire office and staff were happy to see I could confirm their headaches and cussing and would not tolerate these kind of issues for 2 hours or more let alone 2 months.

 

So, with that said, After 7 hours straight, aside from going to get some DP-HDMI video cables from a wholesaler down the street, and countless attempts to stabilize and prevent lockups, I might have found the two each RTX 4000 in SLI don't work very well. Single RTX 4000 was acceptable performance and stable with DEP enabled. They 8/8 versus 8/16 hyper-threading wasn't liked by one of the modeling programs so I get it, they would force the bios configuration change and reboot; I guess special programmers don't like look ahead pipelines, etc. Well, for a $8k system, you are throttled to 8/8 instead of 8/16, whatever. I can tell you. I would have a conversation with the programmers on that issue. More to come on that subject.

 

So here are the configuration basics:

 

* Dell Precision 7280 Xeon Extreme, 32GB ram, 2 each SSD drives (oddly the original ship shows an m.2 2280)?, Windows 10 64 bit, 2 each NVIDIA (dell p/n) Quadro RTX 4000 video cards (see OEM support link at the bottom)

* Autocad 2018 with PractiCAD, Navisworks, Revizto

* Two programs running.. fine, all three then slow to locks up... 3 monitors... LG 32" SVGA/HDMI ports of some sort; nice. I check the LG specs and it was good.

* Two months old computers, on time about 50 hours or so... locks up and just has to go back to old computer on Windows 7... the three programs worked, slow, but worked, and just low on disk space. 3rd party support recommendation upgrade hardware and os and applications.

 

Dell support and 3rd party it support just could not figure it out after 3 different reviews and 3 different attempts of their short lived visits. What can we do they said? find a 3rd party to look at it.

 

This is where I step in per a special request from a recommended 3rd party.... and here we go... a challenge after 1-2 hours, and still unacceptable behavior.. 5 complex variables and had to isolate which one or pair was the problem(s). the 1-2 hours aka 3 hour tour became a 7 hours full pull..

 

let's get started...

 

1. 21 point @GGPCTU #GGPCTU Tune-up, all the windows updates and patches, review and clear and monitor eventviewer, task and performance viewer, logs, all monitored. nothing. just autocad not responding with 3 modeling programs all open with some pretty huge files. 4GB network model file, 186MB local autocad file in a dropbox folder, etc.

2. Drivers and program reset preferences, program repairs, sfc scans, nada...

3. video cables swap out of DP to SVGA instead to DP to HDMI

steps. 1 - 3 showed no signs of helping... mind you that was after a few hours

4. mobo bios reviewed, 5 things not enabled that I normally enable for this type of workstation.

 

* the bios/mobo it was OEM custom designed, one could tell. therefore, enable fans from auto to on 25% since the RTX 4000 were really hot and the chassis had no additional fan support for the dual RTX 4000 addition, and hyper threading enable, from 8/8, an 8 core set to just 8 threads, instead of 8/16 yet one of the modeling programs forced the CPU and MOBO back to 8/8. odd. note: raid mode selected, not ahci nor legacy/ide mode.

 

5. just thinking about all the thrashing and lock ups, disk I/O, memory I/O, GPU I/O. none of that was evident for these lock ups.

6. so pull a video card and go with one RTX 4000. why not, maybe the modeling was struggling with SLI and vendor driver support on the three modeling programs? well that result was interesting... stable... faster and don't know why.. but stable. fyi ... all the autocad performance tricks for dual cards was enabled, and all the 3d etc was enabled and mode modeling support enabled, so do go back down that path of woe.

 

Finally, after 7 hours almost straight and it made no sense, all apps configured, a matching system next to it almost with the same behaviors just not as back, just autocad 2021 instead of autocad 2018... still locks up, full freeze for 15 minutes, then release and freeze... several times and slow performance... gigabit network to 3D model source on the net.. yes, some of the files were cached in Dropbox.

 

I mean the performance monitor is moving along nothing stuck except not responding to the autocad app and nothing else.. just a full 100% lock, 3 apps, 2 fine, 3 bad, totally quiet and full lock up and freeze --

 

so are we talking memory collisions, memory sharing violation and windows 10 won't fault and BSOD nor thrown and exception, yet it should.... what if we look at DEP and just enable that.. I saw that earlier and thought, nah.. why would that matter, this is not a server...what if it was DEP like on servers that need to just be enabled? so I did, and voila..

 

I still pondered... after 7 hours.... DEP seemed like a logical step. was there memory trashing and thrashing? system stable, I mean this is a workstation, a heavy duty one in fact, yet it was behaving badly like every app needed isolation and no sharing or thrashing of resource heaps and cross overs between unused shared resources. I mean most workstations DEP should be off and only one and the steps to isolate an APP is rare, few and far between. What the heck. the AutoCAD 2018 2D was fine the 3D still a little jerky (app spread across two screens, 2D on the center screen, 3D on the right and Navisworks and Revizto on the 3rd left screen.

 

Today, we will see. We do have to get the user settings MENULOAD refigured for Autocad PractiCAD since it won't load right after resetting settings.

 

Note: after getting done for the day working with this client, I looked at my history with the M2000 and M4000 versus the RTX 4000 and yes, the difference in cards is major. a 2015 card using Maxwell tech versus a 2018 card using Turing. I have built 20 systems for civil structural survey engineering in the past few years, using M2000 M4000 etc cards and rock solid on i7-8700K and i9-9900K and no problems, using premo ssd and ddr4. whatever. the specs said the RTX4000 was better than the M4000 M6000 yet I wonder when it comes to the software programmers and their hardware optimizations between 2018 and 2021 AutoCAD. I know the issue and deal is real. The RTX 4000 had better performance yet the M4000 and M6000 had better performance overall in some of the key areas of using AutoCAD. What if the M4000 worked better in Autocad 2018 and the RTX 4000 worked better with Autocad 2021. There was a lot of discussion on this difference, whereas the programmers for AutoCAD and modeling changed alot and the hardware needed to change with it and not necessarily forward and back compatible.

 

REF:  http://www.meta-lab.com/index.shtml?/english/pdsw.shtml

 

note: none of the video software drivers from Dell for this workstation by service tag was in the OEM original build configuration. the card did have a dell part number sticker on it, yet the 3 dell drivers and direct nvidia (3 versions) did not work. had to let Windows figure it out after 5-10 minutes of waiting for the HAL to figure out what it had and Internet update failed, just had to let the INF files, I guess get re-read.

 

REF: Service Tag # JT1CH63 @ https://www.dell.com/support/home/en-us/product-support/servicetag/0-YkhPRUdxUERDSktqVkNROVFoSHZPQT090/overview

 

My M4000 vs RTX 4000 performance review, just look at the price flip flop from release to now and the tech.

 

REFs:

https://gpu.userbenchmark.com/Compare/Nvidia-Quadro-RTX-4000-vs-Nvidia-Quadro-M4000/m716215vsm37687

 

 

 

ORIGINAL OEM build configuration:

 
817-BBBC : Not selected in this configura tion
780-BBCJ : No RAID
658-BCUV : Dell Developed Recovery Enviro nment
658-BBMQ : Enable Low Power Mode
640-BBRC : Dell Precision Optimizer with AI
640-BBLW : Dell(TM) Digital Delivery Cirr us Client
631-ABML : No Out-of-Band Systems Managem ent
630-ABGO : Microsoft Office Professional 2019
620-AALW : OS-Windows Media Not Included
619-AMSU : Windows 10 Pro for Workstation s (4 Cores Plus) Multi - Engli sh, French, Spanish
580-ABIS : No Additional Keyboard
575-BBPB : CPU clip, assemble CPU with he atsink
575-BBCH : No Stand included
570-AADK : No Mouse
555-BBNI : No Driver
555-BBJO : No Additional Network Card Sel ected (Integrated NIC included )
525-BBCL : SupportAssist
520-AABF : External Speaker Not Included
490-BFHO : Dual Nvidia Quadro RTX4000, 8G B, 3DP, VirtualLink (XX20T)
470-AAKG : US Power Cord
461-AABV : No Accessories
444-BBBS : BIOS binary check enabled and verified
444-BBBG : BIOS match checked back to fac tory
430-XXYU : Resource DVD not Included
429-ABER : No Optical
429-ABEP : Slim filler panel (no opt.)
414-BBBV : Intel NVMe PCIe SSD (Front PCI e FlexBay)
412-AAKK : Heatsink for Single CPU
411-XXYB : GPT is 100% required for all o rder
401-ABJT : M.2 512GB PCIe NVMe Class 40 S olid State Drive
400-AVDR : Dell M.2 carrier
400-AKZR : No Hard Drive
389-CGJW : 7820 Tower Regulatory Label (D AO)
387-BBBE : Not ENERGY STAR Qualified
370-ADVE : 32GB 2x16GB DDR4 2666MHz RDIMM ECC Memory
370-AAIP : Performance Optimized
340-CKSZ : No AutoPilot
340-BYIW : Placemat 7820 Tower DAO
340-AGIK : SERI Guide (ENG/FR/Multi)
340-AEYP : SHIP,PWS,LNK,NO,NO,AMF
340-ADBJ : Thank You for Choosing Dell
338-BSMP : Intel Xeon Gold 5217 3.0GHz,(3 .7GHz Turbo, 8C, 10.4GT/s 2UPI , 11MB Cache, HT (115W) DDR4-2 666 1st)
329-BBJL : Trusted Platform Module (Discr ete TPM Enabled)
328-BCRU : Ship Material Tower 5820,7820
321-BENP : Precision 7820 Tower 950W PCIe FlexBay Chassis CL
321-BDWI : Intel Integrated controller (R ST-e) with 1-2 Front FlexBay N VMe PCIe Drives
210-AMDT : Precision 7820 Tower XCTO Base

 

more to come...