CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > General Forums > Hardware

Specs for a good workstation

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   April 23, 2021, 11:22
Default Specs for a good workstation
  #1
New Member
 
Join Date: Apr 2016
Posts: 19
Rep Power: 7
gcoopermax is on a distinguished road
I am trying to finalise specs for my workstation (Dell Precision 7820), with a view to using 2 HPC packs (so 32 core licence) with FLUENT (standard 4 core licence) and need some advice. I have provided the details below, along with the points on which I need help.

Processor: Dual Xeon Gold 6242R (40C, 3.1GHz base, 4.1GHz Turbo) or Dual 6246R (32C, 3.4GHz base, 4.1GHz Turbo)
6242R has lower base clock speed but more no. of cores, so I would be able to utilise 32 cores with HPC along with 4 cores with standard licence, and still leaving 4 cores for any other tasks (e.g. pre/post processing etc). 6246R has higher clock speed (+0.3GHz) however 32 cores so I can't even utilise all the cores included in HPC and standard licence. While turbo frequency for both is same at 4.1 GHz, according to Dell, at a time turbo mode is only used for 8 min and the clock speeds go back to base level to manage thermal workload, so I am assuming that both processors will largely run at base clock speed. I am inclined to go with 6242R despite its lower clock speed, however am interested in any recommendations

Memory: 12X16GB (=192GB) or 12X32GB (=384GB) DDR4 2933 MHz ECC RAM
I will be running the simulations of mostly internal flows (single phase and multiphase/multispecies, which could at times be transient) with mesh count upto 15 million elements in FLUENT and wondering whether there is any point in going beyond 192GB RAM. Since this is a workstation, I will be using it while it is running simulations (for pre/post processing or CAD etc). however trying to understand if there is rationale in getting 192GB+ RAM. Also can someone please explain why is it important to populate all memory channels of the processors?

Boot drive: 2TB M2 NVMe SSD
This will avoid any I/O bottlenecks while running simulations so preferred over SATA SSD.

Graphics card: NVIDIA RTX 4000 8GB
Suitable for general rendering and animations etc. Not intending to use GPU acceleration as this is fairly low specced card
gcoopermax is offline   Reply With Quote

Old   April 23, 2021, 12:49
Default
  #2
Super Moderator
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 2,829
Rep Power: 40
flotus1 has a spectacular aura aboutflotus1 has a spectacular aura about
Quote:
according to Dell, at a time turbo mode is only used for 8 min and the clock speeds go back to base level to manage thermal workload, so I am assuming that both processors will largely run at base clock speed
Unless Dell is doing something particularly stupid, these chips should not run at base clock speeds with light workloads. That's just not how turbo boost works.

Before going into detail about any of this: Intel has already released their third generation of "scalable" Xeon CPUs: https://www.anandtech.com/show/16594...calable-review
If you can, get a workstation with those CPUs. They have been out in the wild for quite some time now, and I see no reason why large OEMs like Dell would not be able to sell workstations with current-gen CPUs by now.
flotus1 is offline   Reply With Quote

Old   April 23, 2021, 13:32
Default
  #3
New Member
 
Join Date: Apr 2016
Posts: 19
Rep Power: 7
gcoopermax is on a distinguished road
Quote:
Unless Dell is doing something particularly stupid, these chips should not run at base clock speeds with light workloads. That's just not how turbo boost works.
The turbo max speed for both the CPUs is same, so I was hoping that when FLUENT is solving, both will provide comparable performance per core, thus 6242R will win as it has more no. of cores. Isn't it a reasonable assumption that FLUENT will try to push the processors beyond base speed (admittedly I don't know much about the turbo boost)?


Quote:
Before going into detail about any of this: Intel has already released their third generation of "scalable" Xeon CPUs
Interesting, I have considered 2nd generation Cascade Lake processors here based on ANSYS' latest hardware recommendations. I didn't see the third generation Ice Lake processors listed on Dell's website, I could always ask them. The comparison between 2nd and 3rd generation 6246 processors below suggests that the latest counterpart, despite having larger memory speed (3200 vs 2999 MHz) and no. of memory channels (8 vs 6), is cheaper!
https://ark.intel.com/content/www/us...=199353,212457


That said, I am limited to a budget of 10k, and want to utilise 2 HPC packs fully (so 32+4 cores) hence looking at the 40 core processor. We need the machine in next couple of weeks, so if Dell can't manage the third generation chips, we have to contend with 2nd generation. Can you comment on the two processors listed and my dilemma about no. of cores and clock speed etc? Also would appreciate any thoughts on the other aspects of the configuration I have listed.
gcoopermax is offline   Reply With Quote

Old   April 24, 2021, 02:43
Default
  #4
Super Moderator
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 2,829
Rep Power: 40
flotus1 has a spectacular aura aboutflotus1 has a spectacular aura about
About Intels turbo boost and similar technologies: https://en.wikichip.org/wiki/intel/frequency_behavior

Turbo tables for individual CPUs are often hard to come by. Here is one for the Xeon Gold 6242: https://en.wikichip.org/wiki/intel/xeon_gold/6242
I can't find the same info for the refresh variant: https://en.wikichip.org/wiki/intel/xeon_gold/6242r

But in general: the higher the TDP rating, the higher the turbo boost table values. Or if you compare two CPUs with similar TDP but different core counts: For the same amount of cores active, they will usually run at similar frequencies. So don't rule out CPUs with high core counts, just because their base frequency is lower. It has to be, in order to fit the TDP envelope when all cores are active.

Quote:
Isn't it a reasonable assumption that FLUENT will try to push the processors beyond base speed
The application has little influence on CPU frequency, apart from the type of instructions it is running. Ansys Fluent will fall under the "normal" category -i.e. not heavy AVX workload- for turbo boost bins.
What matters here is the power setting in the operating system, on top of any limitation that might come from the bios. E.g. you could disable turbo boost entirely in the bios, or the vendor could do that for you. Which is what I was referring to as "something particularly stupid"

tl;dr: the 20-core 6242R is a good choice, despite its lower base clock speed. Picking a CPU with lower core count just for the higher base clock speed is not worth it.

Quote:
however trying to understand if there is rationale in getting 192GB+ RAM. Also can someone please explain why is it important to populate all memory channels of the processors?
With 15M cells maximum, there is really no point in going above 192GB of RAM. You will still have plenty of free memory to run other stuff while solving a model.
Filling all memory channels is important, because memory bandwidth matters in parallel CFD. That's the main reason why I brought up 3rd gen scalable CPUs. And you can only get the maximum memory bandwidth by populating each channel with a DIMM. General recommendations for CFD hardware [WIP]
In order to squeeze out the last few percent of memory performance, you could try asking Dell to use dual-rank memory modules for that 12x16GB configuration. But that's probably not something you have any control over.

Quote:
Boot drive: 2TB M2 NVMe SSD
This will avoid any I/O bottlenecks while running simulations so preferred over SATA SSD.
Graphics card: NVIDIA RTX 4000 8GB
Suitable for general rendering and animations etc. Not intending to use GPU acceleration as this is fairly low specced card
No objections here.
flotus1 is offline   Reply With Quote

Old   April 26, 2021, 11:08
Default
  #5
New Member
 
Join Date: Apr 2016
Posts: 19
Rep Power: 7
gcoopermax is on a distinguished road
Quote:
With 15M cells maximum, there is really no point in going above 192GB of RAM. You will still have plenty of free memory to run other stuff while solving a model
15 million cells for a typical problem, however there may be cases where I will require 40+million cells and hence I thought it maybe useful to go with higher memory capacity. Interestingly, 6242R has 40cores running at 3.1Ghz, while 6246 has 32 cores running at 3.4GHz, but 6242R is cheaper than 6246, hence the money saved can be used to bump up the memory. It is bizarre that going from 192GB to 384GB while configuring the workstation costs about 1k extra. However if I stay with 192GB and add another 192GB later then it will cost over 4k (one chip costs 347)! I don't understand the maths here.
https://www.dell.com/en-uk/work/shop...a579532/memory

Quote:
tl;dr: the 20-core 6242R is a good choice, despite its lower base clock speed. Picking a CPU with lower core count just for the higher base clock speed is not worth it.
I am inclined to go for 6242R as well. With 6246, I can only take advantage of 32 HPC core licence (2 packs), and licence for 4 standard cores is wasted. But with 6242R, I would be able to run 32HPC+4 standard cores, utilising the licences fully. 36 cores, even running at lower speed, will get done things more quickly compared to 32 cores, and still this leaves 4 cores for any other tasks.


Quote:
In order to squeeze out the last few percent of memory performance, you could try asking Dell to use dual-rank memory modules for that 12x16GB configuration. But that's probably not something you have any control over.
This is a good point, I have asked this clarification. The link for the standalone memory chip above appears to be a dual channel (2RX8) however on the page where I configure the workstation, they don't mention the type.

Quote:
That's the main reason why I brought up 3rd gen scalable CPUs. And you can only get the maximum memory bandwidth by populating each channel...
On the call, the Dell agent said that since the third generation processors were released recently, it could be 6 months before they are tested and deployed for the workstations. However he is going to confirm it.
gcoopermax is offline   Reply With Quote

Old   April 28, 2021, 10:36
Default
  #6
Super Moderator
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 2,829
Rep Power: 40
flotus1 has a spectacular aura aboutflotus1 has a spectacular aura about
Quote:
192GB to 384GB while configuring the workstation costs about 1k extra. However if I stay with 192GB and add another 192GB later then it will cost over 4k (one chip costs 347)! I don't understand the maths here.
The accountants at Dell will know best. From my experience, the markup for parts within a configuration is steeper than that. But selling 16GB of RAM for 350 just because it went through your validation process is pretty hilarious. Nothing stops you from buying compatible memory somewhere else, for a quarter of the price.
Well maybe apart from the amount of free DIMM slots on these machines, please double-check before you attempt something like that.

Quote:
6242R has 40cores running at 3.1Ghz, while 6246 has 32 cores running at 3.4GHz
Again: base frequency. A 6242R running on only 16 out of 20 cores will turbo to pretty similar frequencies as a 6246 running on all 16 cores.

Quote:
On the call, the Dell agent said that since the third generation processors were released recently, it could be 6 months before they are tested and deployed for the workstations. However he is going to confirm it.
It is entirely possible that Dell hasn't updated their workstations with the latest CPUs, and may still take a while to get them to market. If they don't skip this generation altogether.
But that explanation you got is just a bunch of nonsense.
Large OEMs like DELL, HP etc. get access to these CPUs way before they are released publicly, and they work closely together with the CPU manufacturer. With the explicit goal of having systems ready for sale as close to the official release as possible. For this generation in particular, the CPUs have been available to "special" customers many months before the official launch. The CPUs being launched a short while ago is an excuse, not an explanation.
flotus1 is offline   Reply With Quote

Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Selection of the best workstation Tensian Hardware 17 September 11, 2020 01:39
CFD workstation configuration calling for help 2 Freewill1 Hardware 6 July 8, 2020 21:17
What workstation to Buy for solidworks Flow Simulation John M123 Hardware 7 April 7, 2016 15:24
Dual cpu workstation VS 2 node cluster single cpu workstation Verdi Hardware 18 September 2, 2013 03:09
PC vs. Workstation Tim Franke Main CFD Forum 5 September 29, 1999 15:01


All times are GMT -4. The time now is 13:53.