HP Z420 and new Nvidia Quadro graphic cards

[GregaJ]

Hey BambiBoomZ,

Thanks for extended answer. I will use the information if I will decide for a new graphics card. About your question which graphics card I'm using, it was mentioned a few times and can also be seen from my hardware picture, that this is Nvidia Quadro K4000.

Thanks also for heads up about RTX 4000. I wasn't even considering it but now I can see that it is a little bit more expensive than P4000:

RTX 4000: https://www.computeruniverse.net/de/pny-quadro-rtx4000-8-gb 

P4000: https://www.computeruniverse.net/de/pny-quadro-p4000-8-gb 

P2200: https://www.computeruniverse.net/de/pny-quadro-p2200-5-gb 

As I can read from your comment, you think that RTX 4000 will last longer than P4000? 

The thing is I don't know what to do exactly. As I mentioned I will try different card drivers and then I will try better graphic card. We have a policy in web shoping here that you can return a product without further notice in two weeks after you bought it. So I will probably try this and see if better and still supported card is the solution. I can still return it with full money refund if it is not good. That is what I can't do with buying a used card.

Further comments are still welcome...

[DGroves]

using a registry cleaner on windows 10 is not recommended, all of the more common cleaners when tested were found to actually cause issues rather than fixing them

the reason it's not recommended is that the auto clean feature of the programs while good, is not foolproof and is unable to determine 100% of the time what is safe to delete or modify and using the advanced modes requires you the user to have a very good knowledge of how the OS works

windows 10 manages the registry quite well on it's own for 99.9% of the users of windows so there is really no need for 3rd party registry cleaners, with that said, if you do use one i strongly recommend you fully understand how the registry works and not blindly take the program makers  auto/default setting of "safe to delete entry" without going over each entry

note that i am only talking about the registry cleaning option here not any of the other utils that may or may not come bundled with the application.

[BambiBoomZ]

GregaJ,

Choosing a GPU is always difficult as the performance for a particular card will be very good in some application and possibly terrible in another.  There are very few GPU's that can do everything well- but I think that's the intended job of some of the NVIDIA Titan designs.  The The Titan Xp is a good gaming card, but also scores well on viewport performance in the way a Quadro or Firepro does also. 

The important decision making is to know the demands of each application in priority  Solidworks fortunately makes the choice relatively, simple; it must be a Quadro.  However, the features and performance level has to evaluated such that the card will not be sufficient nor be a waste of money for performance or features that's not needed. 

In this decision, my way of choosing would be:

If relatively strong performance,  a GPU capable for medium-scale SW projects, plus cost / performance, in my view,  the GPU should be a newer design  that will be useful a long while, plus be power efficient, my choice would be the P2200.

K4000  ___ 2879

P2200  ___ 9344

If the assemblies are large and/or complex, or there is need for more involved compute / analysis /simulation, the Quadro P4000 is a good choice.  However, in my view, as it was released 2/2017, it is one previous generation in the past.  My guess is that it may possibly be replaced within a year, perhaps in the Spring.  P4000's are still valuable when used, but if they are to be obsolete in the near future, the value will drop. If the P4000 is choice, consider buying one that is used and therefore already partly depreciated.

That leaves the other suggestion of the RTX 4000.  For me, this is the forward-looking choice.  Yes,  expensive,  but like the very expensive Titans, it does everything well- probably gaming included.  Importantly,  the trend is already obvious that high end workstation applications such as SW and Catia will increasingly make use of the ray-tracing for rendering,and the AI and machine learning features for compute and simulation functions.  At first it seemed odd to me that gamers were buying RTX 2080 Ti's as though they're obsessed, while only three or four common games could use the central ray-tracing capability.  It doesn't make sense, but they may be wisely planning ahead; the game designers are beginning to add ray-tracing enhancements so that $1,300 investment can eventually by demonstrated to be a good decision .

Used GPUs:  I've mentioned used Quadros a number of times and I know that in DE, used computer components are not considered as favorably as in the US,  but I think Quadros are in a different category than consumer / gaming hardware.  They use less power, so are running cooler, the clock speeds are reduced (= less stress) and the drivers are very highly developed for both precision and reliability.  The drivers development costs in fact are one the important reasons Quadros are more expensive. 

I have a 2004 Quadro FX 580 (512MB) that I bought used in 2010. That has many thousands of hours of use and I still use it to set up a new system as it needs no power connection.  One of the office systems is a 2007 Dell Precision 390 and the used, $61 Quadro K2200 , which our friend SDH mentioned, works really well.  The 390 with a 12-year old, 2.6GHz non-hyperthreading Xeon X3230 WS and still made a reasonable Passmark 3D score of 3286. Compare that to the K4000 average of 2879.  In the office HP z420_1 (2013) ( Xeon E5-1620 3.6/3.8GHz) , another K2200 3D was 3481.  This is not to advocate buying a K2200- I'd suggest a K4200 in that series (average = 4473) but to suggest that older Quadros appear to provide completely useful performance results and work forever!

BambiBoomZ

[GregaJ]

BambiBoomZ,

Can you please post a link to these graphic cards scores you are mentioning? And what do they mean? I found some charts myself, but when it comes to particular application as SW in my case, things are more specific, as you know...

Not much of these posts are related to one of my questions about graphic card drivers and which one to use, when there is no more support from software side. Trial and error is probably the way, but some experiences would be good.

[DGroves]

please keep in mind that passmark scores  will differ depending on which version of the program generated the score and that anyone moderately skilled can edit and then upload a score

while passmark does try to remove the obviously fake scores, they are unable to catch the ones that have just been moderately tweaked also the listed score does not state which version of the program generated the score which makes a apples to apples comparison impossible

this is not to say passmark is a useless program (far from it) only that like any other benchmarking program you have to keep it's limitations in mind. it's a great program for checking if a change on your system was/is worthwhile as you will be using the same version and know the posted score is accurate

in your case the cinebench 15 or 20 program might be useful as can  separately benchmark the CPU or OPENGL (v15)

while the v20 program increases the load on the system to compensate for newer systems having more cpu cores

[GregaJ]

Ok. I checked some of the data. Curentlly I came to conclusion that P2200 or RTX 4000 is the way to go, because both are quite recent cards. P4000 will probably be outdated soon and nobody knows how it will be supported from SW side. And If I'm prepared to pay for P4000, RTX 4000 is not so far away.

The only thing now is money side. There is double difference in money between RTX 4000 and P2200. I will see if there will be any good black friday deals. The end of year is close and I have to keep my eyes on good deals.

If anybody has more opinions, they are of course welcome...

[BambiBoomZ]

GregaJ,

If it's in the budget, I think the Quadro RTX 4000 is the choice for a GPU that can do anything well of almost any scale and perhaps for the next 5-years.  If the projects are of medium complexity and there is not high compute load for animation, dynamic stress, thermal, gas flow analysis, the P2200 is extremely good value.

Passmark publishes and constantly updates it's searchable benchmark charts for CPU's, GPU's, and hard drives. 

It's possible to access all three from the one chosen:

As we are discussing GPU's:

https://www.videocardbenchmark.net/gpu_list.php

To find the relative benchmark order for Quadros, click on the column heading = "Rank (lower is better) so that the number 1 appears- currently TITAN V CEO Edition.   Then in the "Find Videocard", enter the word Quadro and click on "Find Videocard".  this will maintain the ranking order and highlight all the Quadros on that order.  I like this display configuration as it's possible to compare the relative performance of Quadros to AMD Firepros and NVIDIA consumer cards: GTX and RTX.  In that display, for example, it's possible to see that the RTX 4000 at 13387 has a higher average 3D performance than the RTX 5000 (13240) and is just below the $3,500 Quadro P6000.  That the difference of only one generation form Pascal to Turing. As I use a GTX 1070 Ti, It's possible to quickly see that it's 8 rankings lower than the RTX 4000 at 12325.

If a GPU's performance is not already generally known in Passmark, instead of having to scan through hundreds of listings, it's useful to reorder the chart  by  "Videocard Name".  In that display, all the Quadros will be listed together alphabetically.   Then, it's quickly possible to look over the entire range of Quadros, new and obsolete; and put them into perspective, for example,   my vintage FX 580 3D= 268,  a K2200 = 3598, K4000= 2879, and the K4200 (which I thought was a great Quadro) = 4473. 

These two search methods are very efficient in providing a good idea of  relative performance.  However, the usefulness is vastly improved if the chart is compared to the actual benchmark results of the system being upgraded.  For this, I recommend, installing, and testing your current system on Passmark Performance Test.  There is a free 30-day trial, but it is so useful- and so inexpensive, I recommends just buying it:

https://www.passmark.com/products/performancetest/index.php

Running this benchmark and comparing the results in the baselines which have a systems rating, CPU, 2D, 3D, memory, and disk benchmarks can quickly reveal problem areas and chart improvements at whatever level.  For example here is the original Precision 390 test:

Dell Precision 390 (2007) (Original): Intel Core2 Duo E6300 2-core @ 1.86Ghz / 2GB (2X 1GB) DDR2-667 ECC > Quadro FX550 / 2X WD 320GB ( RAID 1) / 375W
[Passmark Rating = 397, CPU = 586, 2D= 339, 3D=75, Mem = 585, Disk = 552 [STM= 709]

and the latest test:

Dell Precision 390 (2007) (Revision 4): Xeon X3230 quad-core @ 2.67GHz / 8 GB DDR2-ECC 667 ECC / Quadro K2200 4GB / WD Blue 500GB + WD 320GB /Creative Audigy 2ZS SC > Windows 7 Professional 64-bit 7.18.18

[ Passmark PT9: Rating = 1464 / CPU = 3408 / 2D= 431 / 3D=3286 / Mem= 863 / Disk= 514  [STM= 1051]  7.27.19

 Besides the 3D , for 3D modeling, the CPU / Single thread performance is particularly important for SW.  There is very good Passmark chart , the "CPU Megalist"  in which it's possible to put CPU's in order of single-threaded performance:

https://www.cpubenchmark.net/CPU_mega_page.html

The main systems in the office have single thread marks of 2368 (Z620_3) and 2384 (z420_3).  I believe Brian1965's z620 single threaded is about 2550 and my next system I'd like to have at least 2700. The recent AMD Ryzen 3900X 12-core has an average of 2925 and the new Ryzen 7 Pro 3700 8-core = 3044.    These baselines are extremely valuable as they list the CPU, motherboard model, GPU, drives, and amount of memory.

Benchmarks of user systems of course, are not rigorously tested, some users report tests on the same system  multiple times, and users  that are trying to diagnose existing performance problems will lower the averages significantly.  However, with a little experience, it's possible  to work out which components combinations are better in the parameters that are important to the use.  I keep a list of all the tests in order to evaluate progress. I've seen quite a few i9-9900K systems, the top gaming processor with a single threaded  = 2893  with Quadros and a number will list Solidworks in the title of the system . With SW, modeling needs a high single threaded, simulations need a lot of cores and RAM, and so on. 

Another source of good information to aid development of a SW system is Puget Systems, who describe the best approach to SW systems:

https://www.pugetsystems.com/recommended/Recommended-Systems-for-SOLIDWORKS-150

The description of how they select the components to run SW - separated by  a modeling focus and rendering & simulation is applicable  to the HP z-series

BambiBoomZ

[GregaJ]

BambiBoomZ,

Thanks for extended answer. I will check all this things and they will be very useful.

But one question, just out of curiosity. Did you see Quadro RTX 8000 on the chart? Very low, but price is off the charts... What is the deal here? If I understand correctly this should be the highest ranking Quadro for now.

[BambiBoomZ]

GregaJ,

You're welcome.  No one benchmark is sufficient, but Passmark, as it lists the major components and seperate each benchmark into further categories, is still very useful in making relative comparisons.  One can quickly see general trends in certain motherboards having better results for certain processors, plus the effect of CPU clock speed on 3D performance- that kind of thing.

As for the Quadro RTX 8000 low average performance figures, this is an example of the need to look at a benchmark in greater detail. Going to a baselines Advanced Search> Graphics Card = 'Quadro RTX 8000" , one is struck that there are only 6 listings.  The fact there only six of course reflects the price of more than ^,100 in the US and that few users need 48GB of VRAM.  As Passmark will mention occasionally, a component with very view results make those results less indicative.  For the RTX 8000 the 3D results are:

15383

12827

12517

10966

  9616

  4500

It's quickly obvious that a spread of scores from 4500 to 15383 will skew the average considerably lower. There are often baselines that seem well off the probability curve. 

For comparison, the RTX 4000  has a sample of 87 tests with a 3D high mark of  19344 (i7-9700K @5.3GHz) and two low of 0.0000, one being on a Xeon Gold 6126 12C@ 2.6/3.7GHz.  The 19344 demonstrates the CPU clock speed, which = the single-threaded performance  effect on 3D.  I don't know if 0.0000's are averaged in; 3147 is the lowest positive figure.  When one goes to a quite common GPU, the skewing by anomalous results is reduced.  For the GTX 1080 Ti, of which there are 15,088 tested, the high 3D mark is 24533 and there more than  400X 0.0000 with the first positive at 0.9.  There is another lesson to be learned with Passmark , for one thing, the 24533 is probably two GTX 1080 Ti's in SLI and I often see drive scores that have to be multiple drives in RAID 0 or 0+1 configurations.

Passmark results need to be navigated / researched carefully, but when the limitations are understood it's really useful. Another use is to see if a particular component is compatible - processor, amount of RAM, GPU, drive, etc. in your system.  If there is only one listing , it must work, but may need special parts or configuration and if  there are several listings, it works and probably does not very special installation parts or procedures.

BambiBoomZ

[GregaJ]

One more question. Is there possible to set anything under Nvidia Control panel to get the most out of Solidworks or in general out of graphics card?