z640 (2014) upgrade

[Maxine_22]

I have a trusty z640 workstation, bought in around 2014.

I'd like to show it some love with an upgrade this year and I'm hoping someone could offer some practical suggestions. It's used for 3D modeling & rendering, photoshop & video production.

Current specs are as follows;

HP z640

2 x Xeon E5-2660 v3 @ 2.6Ghz

32GB RAM

GPU: Nvidia M4000 8Gb (in Slot 2 PCIe3)

Boot Drive: HP zTurbo 512GB (in Slot 3 PCIe2)

Data Drive: WD6000HLHX 600Gb VelociRaptor HDD in internal drive bay 1

First thoughts are a GPU upgrade to Quadro RTX, and replacement of the HDD to a zTurbo (or equivalent).

Any ideas? I'd like to then re-use the new components in a complete z workstation replacement in a couple of years.

Thank you,

Max

[DGroves]

you may also want to consider  a used intel 3600 series SSD.  Currently on ebay a SUN/Oracle 3605 1.6 TB PCI-e card ssd is around 170.00 the sun card actually has better specs than the retail intel 3600 card and will give the overall system a major speed boost

currently i would not buy any 3x series video card as the price is well over the suggested retail price, however if you can get a quadro 3x card for retail then it's well worth upgrading

consider also upgrading ram to 64GB or more as video work will consume lots of it

depending on exactly what you are doing video wise, changing the cpu to a faster and/or more core count might be worthwhile however the cost of a cpu/video upgrade may not be worthwhile depending on your needs and if the system is used professionally or personally

[Maxine_22]

Thank you very much for your response - I've been distracted by replacing my laptop in the meantime, but will now certainly follow some of your suggestions for my z640. The SSD is a good starting point, and I'm also looking at a Quadro RTX  card.

I probably won't upgrade the CPUs - will rather do that in a couple of years when I replace the entire machine.

Thanks again,

[BambiBoomZ]

Maxine_22,

The applications listed, in total, require strong performance in every subsystem: 3D modeling needs a strong single-thread performance, which means a high clock speed on one or more cores, CPU rendering and video production can typically use as many cores, and rendering and video need a lot of system memory, VRAM, and fast disks. 

GPU:  If the rendering is ray-tracing, then yes, an RTX is recommended.  Unless the application requires a certified driver and/or the 3D-modeling is Solidworks or Catia which have been traditionally carefully oriented towards particular Quadro viewport support, consider an RTX 3070 8GB, preferably having a single fan blower-style cooling. the Quadro RTX 4000 8GB has excellent performance but with CUDA cores and with a Passmark 3D= 15293 with 2,304 CUDA cores in the US for $900 as compared to the average RTX 3070 3D= 21628 with 5,886 CUDA cores and someday when the scalpers are through, it will cost $500+. as compared to the RTX 4000, the RTX 3070 has +40% 3D performance  with 2.5X the CUDA cores for .5X the cost. An RTX 3070 will be a good component for the future system.

Disk:  The 600GB VelociRaptor was of course a fast mech'l HD in 2014, but that is far from adequate storage for a system including video editing today.  In the main office system here, a z620, without video editing, there is an HP ZTurbo 256GB OS.programs drive, a 500GB NVMe M.2 for active projects, an internal 4TB, and an external backup drive of 6TB. Retain the Z Turbo for the OS/Programs, and consider a 500+ NVMe active projects drive, and an internal storage HD of drive at least 4TB. a large external drive, containing a full backup and a full system recovery  image and that is run only when backing up is also a good protection.  If a virus or ransomware strikes, the system can be fully recovered top to bottom in a hour or so.  

CPU: The specification becomes even more complicated when considering the difference between real-time working and set it going and leave tasks. In my view, if there is 3D modeling, which is real-time labor- intensive and requires a high clock speed CPU and GPU, it seems preferable to have the processor performance prioritize that by having a high clock speed and then assuming that the fewer cores could make rendering a bit longer. When I work on a 3D model,  I would rather have the modeling be fluid without navigation and regen time and then if the rendering requires an extra fifteen minutes, that is time that I would would doing something else anyway. As well. sometimes a faster 8-core can accomplish more clock cycles per unit time (IPC) than two of a slower CPU. For example, in one HP z640,  one Xeon E5-2690 v4 14-core made a Passmark CPU score = 21687 whereas the best performing 2X E5-2660 v3 10-cores =  21566.  A z640 / 2X E5-2667 v3 8-core /Quadro K4200 /128GB RAM / Samsung XP941 256GB = 24576.  Importantly the Passmark single thread rating for the E5-2660 v3 (10C @ 2.6/3.3GHz )= 1,770 , whereas the rating for the E5-2667 v3 (8C @ 3.2 / 3.6)= 2,053.  When rendering, the E5-2667 v3 will have a higher all-core clock speed. Consider changing the CPU's to 2X E5-2667 v3.  The  2667 v3 CPU's of course will not transfer to a new platform, but some cost may be recovered when selling the current CPU's and the  E5-2667 v3's should retain a reasonable resale value.

BambiBoomZ

HP z620_2 (2017) (R7) > Xeon E5-1680 v2 (8C@ 4.3GHz) / z420 Liquid Cooling / 64GB (HP/Samsung 8X 8GB DDR3-1866 ECC registered) / Quadro P2000 5GB _ GTX 1070 Ti 8GB / HP Z Turbo Drive M.2 256GB AHCI + Samsung 970 EVO M.2 NVMe 500GB + HGST 7K6000 4TB + HP/HGST Enterprise 6TB / Focusrite Scarlett 2i4 sound interface + 2X Mackie MR824 / 825W PSU / Windows 7 Prof.’l 64-bit (HP OEM) > 2X Dell Ultrasharp U2715H (2560 X 1440)

[ Passmark Rating = 6280 / CPU rating = 17178 / 2D = 819 / 3D= 12629 / Mem = 3002 / Disk = 13751 / Single Thread Mark = 2368 [10.23.18]

HP z420_3: (2015) (R11) Xeon E5-1650 v2 (6C@ 4.3GHz) / z420 Liquid cooling / 64GB (HP/Samsung 8X 8GB DDR3-1866 ECC registered) / NVIDIA GeForce GTX 1060 6GB/ Samsung 860 EVO 500GB + HGST 4TB / ASUS Essence STX + Logitech z2300 2.1 / 600W PSU > Windows 7 Professional 64-bit (HP OEM ) > Samsung 40" 4K

[Passmark System Rating: = 5644 / CPU = 15293 / 2D = 847 / 3D = 10953 / Mem = 2997 Disk = 4858 /Single Thread Mark = 2384 [6.27.19]

HP ZBook 17 G2: (2015 ) i7-4940MX Extreme (4C@3.1/ 4.0GHz) / 32GB / Quadro K3100M 4GB / Kingston 480GB SATA SSD > 17.3" LCD 1920 X1080 panel > HP docking station> video externally to HP 2711x 27" LCD + Dell 17" (2007) / Logitech 533 _2.1 speaker system

[Passmark System Rating: = 3980 / CPU = 10140 / 2D = 618 / 3D = 2779 / Mem = 2559 Disk = 4662 / Single Thread Mark = 2387 [1.3.20]

Dell Precision 390 (2007) (R9_T2): Xeon X3230 (4C@ 2.67GHz) / 8 GB DDR2-667 ECC / Quadro K2200 4GB / Samsung 850 EVO 250GB _ HGST 7K4000 1TB / Creative Audigy 2ZS S/C > Windows 7 Professional 64-bit
[ Passmark PT9: Rating = 1964 / CPU = 3402 / 2D= 398 / 3D=3300 / Mem= 845 / Disk= 2691] [STM= 1021] 12.3.20

Network: Netgear GS108-400NAS

[Maxine_22]

Thank you so much for your thorough reply - I really appreciate the time people take in this community to respond to queries.

I've just invested in a high end zBook to see me through the next while (not yet delivered), so for my z640 I'll only upgrade the VelociRaptor to the SSD mentioned by DGroves previously.

However, I have some urgent project work coming up that requires some heavy duty GPU work using Twinmotion/ LumenRT. I hope to purchase a NVIDIA RTX 6000 this week, to use in my z640 - and transplant it into a new workstation in the next year or two. I've not been able to find a definitive answer, and hope that you could help. Is the NVIDIA RTX 6000 compatible with my HP z640? The supplier I've identified has a 20% handling fee on returns, so a mismatch would be a disaster for me.

The RTX 6000 occupies two PCIe slots (although physically connected to just one), so I won't be able to get it into Slot 2 because Slot 3 currently hosts my zTurbo C drive.

Will it go into Slot 5 (PCIe3 16x)?

You've suggested the RTX 4000, so I really hope the RTX 6000 will work but need to be 100% sure. Please let me know what you think.

Thank you!

[BambiBoomZ]

Maxine_22,

Twinmotion will benefit from a GPU having a large amount of RAM, but I question, as to whether there is any benefit in using a Quadro.  Also, verify that Twinmotion is efficiently ray-tracing based. Quadro's have the same graphics processors as their RTX counterparts, but have certified drivers required by certain applications for business use- Autodesk and Solidworks / Catia for example , for business technical support,and are slightly detuned with lower clock speeds.  The  Q RTX 6000 is also in the US, almost $4,000. It's possible at the moment to buy a complete 4.6+GHz  8-core, RTX 3070 system for less than $2,000.

Compare the RTX 3090 to the Quadro RTX 6000:

RTX 3090 as compared to Quadro RTX 6000 has :

60MHz faster GPU clock speed  (1,395MHz vs1,335MHz)

47.2 GPixel/s higher pixel rate (189.8 GPixel/svs142.6 GPixel/s)

21.89 TFLOPS higher floating-point performance (35.58 TFLOPS vs13.686 TFLOPS)

128.3 GTexels/s higher texture rate (556 GTexels/s vs427.7 GTexels/s)

7500MHz higher effective memory clock speed (19,500MHz vs12,000MHz)

360GB/s more memory bandwidth (936GB/s vs 576GB/s)

5,887 more shading units (10496 vs 4608)

9,700million more transistors (28,300 million vs18,600 million)

Moreover,  both of the above will be strongly CPU limited by the E5-2660 v3 3.3GHz turbo speed. There are 34 Quadro RTX 6000's tested on Passmark. On an i7-8086K system with a Turbo clock of 5.2GHz, the 3D score = 21366, but on a dual Xeon Gold 6150 2.7 / 3.4GHz system, the 3D= 14414, a loss of 48%.  For comparison, an AMD Ryzen 9 3900X system, as it has a 4.6GHz Turbo, with a RTX 6000 scores 20535 , so a future system having a high clock speed would have more benefit.  Unless the project has a strict deadline and every hour delay is costly,  consider an RTX 3080:

GPU: GA102

Cores: 8,704  (as compared to RTX 6000 4,608)

TMUs: 272

ROPs: 96

Memory Size: 10 GB

Memory Type GDDR6X < Faster than the RTX 6000 GDDR6

Bus Width: 320-bit

When supply meets demand, and the cryptocurrency mining boom is over after the crash at the end of this year, an RTX 3080 will be about $750.  The RTX 3070  (8GB / GA104 / 5,887 cores)will likewise, when near the MSRP  of $500+ be a good value.

There doesn't appear to be references to "LumenRT" .  Perhaps that is referring to Lumion.

A workstation system is best when the components are balanced in terms of components: a top end GPU matched with a slow, obsolete architecture processor is wasted.

As for the placement of the GPU, the second PCIe x16 slot is fine. After installation, check in Setup how it is listed. In the z620 here, because of the double height GTX 1070 Ti and the Z Turbo Drive, the Samsung 970 EVO NVMe M.2 500GB drive is using the second GPU slot, where it actually exceeds it's rated performance. 

BambiBoomZ

[Maxine_22]

Again, thank you so much for your time and advice.

Have to say, I was on the verge of just jumping in and buying the Q RTX 6000 tomorrow out of instinct and experience that my 6 year old Quadro M4000 has served me incredibly well. I've dabbled in the Quadro/ GForce debate over the years and am left none the wiser each time. On paper, those RTX 3090 specs look highly competitive (and price is no comparison) - but why do I keep feeling I'm going to run into stability issues unless it's Quadro?

Just to fill you in further - LumenRT is a Bentley GPU rendering product, something I use intensively and has recently incorporated RTX. The devs have been pushing Quadro cards (particularly large RAM) as the way to go. Twinmotion will surely have RTX support at some point in the future.

My dilemma is this upcoming project - I can't do an overall workstation upgrade now, but need something to handle large files (2Gb+) and crank out frames per second rather than frames per ten seconds. Perhaps I should get away from my "future-proof" ideals (which has served well for 6 years), go for a cheaper option just to get me through this hitch.

Any further insights from you would be incredibly useful - I hope to make some kind of decision tomorrow.

Thanks again!

[DGroves]

if a application can make use of a quadeo/RTX driver, some of the graphic cards calations are speed up massively

because of this a rather low end quadro card can actually be faster than a high end gaming card also the precision of a professional card is a know quantity IE- a circle is actually round instead of somewhat round which is quite important when building things like bridges high rise buildings and so on

so if your needs (time/accuracy) require these then a gaming card is not recommended and the application makers recommendations should be followed

i don't think miners/scalpers are going after the quadro cards due to the initial high price these professional cards start out at

[BambiBoomZ]

Maxine_22,

Yes, the description of the project scale and time criticality does change the situation as regarding specification. The mention of FPS suggesting animation / fly-throughs and the 2GB project size does also suggest the criticality of stability.

I had a look at the Bentley site and the minimal system requirements list is quite amazing:

"This is the machine on which you will author scenes and animations inside the LumenRT LiveCube.

• Windows Windows 7, 8.1, or 10 : Must be 64-bit based OS platforms
• When authoring LumenRT LiveCubes, we recommend a graphics a graphics card containing at least 4 GB of VRAM with a 3D Mark FireStrike score of 10,000 or better (http://www.futuremark.com/hardware/gpu)"
  • > The RTX 3080 scored 10667 in that benchmark. there is no Firestrike benchmark for the Quadro RTX 6000.
• 2GHz Intel processor or faster (multi-core CPU recommended)
• 4GB of free RAM (16GB recommended)
• 12GB of free Hard Disk space.

Notice that the GPU specification is related to the Firestrike benchmark. 

In the subject project, it seems worthwhile to consider an Adobe Premiere optimized system as a model.  the Puget Systems (pugetsystems.com) articles on optimizing for various common applications are well-reasoned, based on extensive testing and development. As well, the recommended system optimized for 4K Premier editing  has a 16-core Ryzen with a Turbo speed of 4.9GHz, 128GB of RAM, an RTX 3080, and the OS SSD is M.2 PCIe4.0. configuration. The interesting aspect of that system is that it's not terribly more costly than a Quadro RTX6000 on it's own. The Quadro RTX6000 option in that system adds +$3,400 over the RTX 3080 version.  From reviews and comments, it seems less likely that Puget would offer the RTX 3080 if it was not reliable in professional applications.

This is mentioned to reiterate the need- or rather the desirability of  fast, high thread-count processors (s), a lot of RAM, high performance GPU, drives , and so on- high performance everything. The system is only as strong as the weakest link as mentioned in comparing the same GPU with various clock speed processors- a slow CPU can cut the GPU performance nearly in half.

Upgrading with one eye on the future is a difficult equation. If it was necessary to upgrade the CPU's, GPU, and RAM on the subject z640 to complete the work in the allotted time, and only the GPU could transfer to the next system, that could represent a poor cost/benefit situation.

BambiBoomZ