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

CPU choice for budget desktop

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

Like Tree7Likes
  • 2 Post By ships26
  • 1 Post By flotus1
  • 1 Post By Simbelmynė
  • 2 Post By flotus1
  • 1 Post By Simbelmynė

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   July 8, 2020, 19:37
Default CPU choice for budget desktop
  #1
New Member
 
Francisco
Join Date: Sep 2018
Location: Portugal
Posts: 14
Rep Power: 3
ships26 is on a distinguished road
Hello there!

I need some help choosing a cpu for a desktop on which I will be developing CFD code and running validation simulations which should be < 2*10^6 cells with up to ~1000€. Anything larger than that will most likely be ran on the uni's workstations.

The cpus I'm currently considering are the Threadripper 1900X, i5-10600K and the Ryzen 5 3600.

The 1900X is the optimal choice for CFD simulations because of the quad channel memory. However, I'd be paying ~1000€ for a desktop with a 2.5 year old cpu, which makes me a little sceptical about upgradability and how well it'll age. I was also worried about memory compatibility if I try to go for 4x8 GB RAM at 3000-3200MHz.

Both the i5-10600K(F) and the the Ryzen 5 would be pretty standard desktop builds for ~750-900€ with 16GB RAM (2x8). They're on my mind above most other cpus because of the core scalability with dual channel memory, so I was trying to get the best hexa core performance possible, if that makes sense.
That being said, the 3600 is cheaper, and supports faster (3200MHz) memory out of the box. In Intel's favor, when looking at this, I get the feeling that the i5's performance for quad or hexa core simulations would be significantly better, so it could be worth the extra investment.

Do I have better options with this budget (i7, R7 3700 or 3800x), or is my reasoning faulty somewhere? The main point is that I'm looking to optimize the build for CFD while preventing it from becoming obsolete for everyday tasks too fast, if this even a legit worry...


Thanks in advance for your help!
aRazs and robbertflord134 like this.
ships26 is offline   Reply With Quote

Old   July 10, 2020, 03:56
Default
  #2
Senior Member
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 2,500
Rep Power: 35
flotus1 will become famous soon enoughflotus1 will become famous soon enough
Considering that you would need a pretty expensive motherboard paired with a TR 1900X, I think you have a better option at a similar price: Intel I7-9800x.
It sure costs more (~340€ in Europe) than a TR 1900x, but you can get a 200$ motherboard to go along with it. ASRock X299 Extreme 4.

With much higher single-core performance and no need to worry about NUMA, it is hands down the better choice for a development workstation, compared to first gen Threadripper.
Habib-CFD likes this.
__________________
Please do not send me CFD-related questions via PM
flotus1 is offline   Reply With Quote

Old   July 10, 2020, 15:55
Default
  #3
New Member
 
Francisco
Join Date: Sep 2018
Location: Portugal
Posts: 14
Rep Power: 3
ships26 is on a distinguished road
Quote:
Originally Posted by flotus1 View Post
Considering that you would need a pretty expensive motherboard paired with a TR 1900X, I think you have a better option at a similar price: Intel I7-9800x.
It sure costs more (~340€ in Europe) than a TR 1900x, but you can get a 200$ motherboard to go along with it. ASRock X299 Extreme 4.
Thank you very much for you feedback! That is quite an interesting idea! I didn't know there was a 200€ a motherboard to go with a Skylake X. It gets a bit more expensive, as I'm only finding the 9800X for ~390ish€, but it certainly makes more sense than the 1900X.

Meanwhile, I thought I should run the OF benchmark case of the stickied thread on the machine I have access to. Just to make sure there is no overlap before taking a decision.

May I ask you how high you think the 9800X would be in that benchmark? Maybe slightly above the 1920X?

ships26 is offline   Reply With Quote

Old   July 10, 2020, 16:21
Default
  #4
Senior Member
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 2,500
Rep Power: 35
flotus1 will become famous soon enoughflotus1 will become famous soon enough
Right on the first page of that thread, we have results for a 7820x and a 7940x.
Since Intel is selling the same CPUs under different names for years now, you can take these values as a very accurate estimate.
Give or take some margin for different software, memory and settings. As you can see, there is a 1950x in that list, slower than a 1920x. So add some healthy error bars to the results in that thread.
Listings starting at 340€: https://geizhals.eu/intel-core-i7-98...-a1910522.html
__________________
Please do not send me CFD-related questions via PM
flotus1 is offline   Reply With Quote

Old   July 11, 2020, 06:46
Default
  #5
Senior Member
 
Simbelmynė's Avatar
 
Join Date: May 2012
Posts: 427
Rep Power: 12
Simbelmynė is on a distinguished road
2 x 2690v2 is still a very good choice, imho, if you are on a budget. There are reports in the thread where such systems manage 1 iteration per second (not listed in the figure though). You would need good 1866 MT/s memory for that I guess, but it should be fairly easy to obtain. It may be a bit too expensive though, at approx 1700 Euros. Anyhow, you will not get close to those values with either 7800X, 9800X or 10900X series or any Zen 1, Zen 1+ Threadripper. Also, the budget will be difficult (impossible) if you try to build a HEDT with new parts, unless you pick 1900X or possibly 1920X. The cheapest 7820X are around 350 Euros on eBay, excluding shipping, so up to 400 Euros (strange considering 9800X is about the same new in some stores!). A 1900X is 150 Euros new and 1920X is 210 Euros. The X399 platform is about 100 Euros more expensive than the X299. So if you are on said budget, then a 1900X or 1920X may be the best option. The difficult part will be to find dual rank memory @ 3200 MT/S that works with the CPU and motherboard in that case. If you pick a HEDT then I think Intel is probably the safer choice since you have less problems with memory support and you also have an easier time overclocking (if you would be interested in that).



GL; HF!
ships26 likes this.
Simbelmynė is offline   Reply With Quote

Old   July 11, 2020, 08:12
Default
  #6
Senior Member
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 2,500
Rep Power: 35
flotus1 will become famous soon enoughflotus1 will become famous soon enough
The new meta for used dual-socket builds is the Xeon E5-2673 v3. They can be had for around 100€ each on aliexpress. Or a bit more expensive if you want to buy within Europe. Prices for other v3 Xeons have come down as well.
Paired with this motherboard: https://www.ebay.de/p/218396160?iid=383440639934

But since OP has access to different machines for doing the heavy lifting, I thought he might get more out of the higher single-core performance of a 9800X.

You already mentioned some of the issues with AMDs X399: boards are more expensive than Intel X299, and getting compatible high frequency memory for first gen TR can be difficult, and add even more to the cost. Combined with lower single-core performance and two NUMA nodes per CPU, I don't see the appeal for a development workstation.
Simbelmynė and ships26 like this.
__________________
Please do not send me CFD-related questions via PM
flotus1 is offline   Reply With Quote

Old   July 11, 2020, 08:37
Default
  #7
Senior Member
 
Simbelmynė's Avatar
 
Join Date: May 2012
Posts: 427
Rep Power: 12
Simbelmynė is on a distinguished road
@flotus1


How much faster would you say this system is compared to a 2690v2?


Can we guesstimate around 10-15 % if 2133 memory is used?
Simbelmynė is offline   Reply With Quote

Old   July 11, 2020, 09:17
Default
  #8
Senior Member
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 2,500
Rep Power: 35
flotus1 will become famous soon enoughflotus1 will become famous soon enough
Hard to tell exactly, but probably somewhere in the 10-20% range. And I would guess at a slightly lower power consumption, with the overall more modern platform.
Since DDR3-1866 reg ECC is still being sold at rather high prices, using the lowest tier DDR4 memory should not add to the cost too much.
__________________
Please do not send me CFD-related questions via PM
flotus1 is offline   Reply With Quote

Old   July 11, 2020, 11:43
Default
  #9
New Member
 
Francisco
Join Date: Sep 2018
Location: Portugal
Posts: 14
Rep Power: 3
ships26 is on a distinguished road
@flotus:
Quote:
Originally Posted by flotus1 View Post
Thank you very much! It is way less expensive what I was coming up with!

@Simbelmynė: Thank you for your feedback as well! I believe the 2 x 2690v2 would be a bit too expensive for me (not that I wouldn't like to say I have a 20 core monster-machine in my bedroom, of course! ). It would be a heck of a build! But, like flotus said, I was leaning more towards a HEDT, due to the price and me having access to other machines.

I was trying to gauge the price delta vs performance between a HEDT with quad channel memory, a more standard dual channel build and the machine I'm currently using. That being said, I ran the benchmark case a bit ago:
Code:
# cores   Wall time (s):
------------------------
1 1546.8
2 859.93
4 414.82
8 309.71
10 297.45
  12 295.6
So, from what other people have reported in that thread, any of the HEDT choices would provide a decent boost in performance (maybe up to twice as much). The i5 and the 3600 would provide an improvement as well, if they are close to the results reported for the Ryzen 7s.

The price difference I'm coming up with between the 1900x and the 9800x builds is ~100€, if the same memory was to be used (the memory in there is just a placeholder):
9800x- https://pt.pcpartpicker.com/list/TYR3TC
1900x- https://pt.pcpartpicker.com/list/sKzvJb


Regarding memory: do you know whether the 9800x scale so well with RAM frequency as the 1900x for numerical simulation?
Also, as I'm sorta new to memory channels: would it be effective/easier to find a pack of 2 dual rank sticks (like 2x16GB) instead of a 4pack of single rank (4x8GB) memory sticks?


Thank you again!
ships26 is offline   Reply With Quote

Old   July 11, 2020, 14:01
Default
  #10
Senior Member
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 2,500
Rep Power: 35
flotus1 will become famous soon enoughflotus1 will become famous soon enough
Quote:
Regarding memory: do you know whether the 9800x scale so well with RAM frequency as the 1900x for numerical simulation?
Different CPUs benefiting more than others from faster memory is a myth. Given the same workload without other bottlenecks, both AMD and Intel will benefit from faster memory in a similar fashion.

Quote:
Also, as I'm sorta new to memory channels: would it be effective/easier to find a pack of 2 dual rank sticks (like 2x16GB) instead of a 4pack of single rank (4x8GB) memory sticks?
Memory ranks and channels are not interchangeable. If you have 4 memory channels, you need 4 DIMMs to populate them all.

I recommend against the Noctua NH-U12s CPU cooler. Compared to dual-tower coolers from other brands in the 50$ price range, it is pretty weak.
__________________
Please do not send me CFD-related questions via PM
flotus1 is offline   Reply With Quote

Old   July 11, 2020, 15:21
Default
  #11
Senior Member
 
Simbelmynė's Avatar
 
Join Date: May 2012
Posts: 427
Rep Power: 12
Simbelmynė is on a distinguished road
Quote:
Originally Posted by flotus1 View Post
Different CPUs benefiting more than others from faster memory is a myth. Given the same workload without other bottlenecks, both AMD and Intel will benefit from faster memory in a similar fashion.

Agreed-ish, the difficulty (as I see it) is finding reliable information on how to set the timings in Intel systems. Primary timings are easy, but secondary and tertiary are very difficult to find. If Intel systems behave in a similar manner as AMD then secondary and tertiary timings are very important. Finally, Zen improve with increasing infinityband frequency, and as such, 3600 MHz (up to 3800 or more with XT series) is the sweet-spot for AMD. Intel should benefit more when going higher (but we can discuss the benefits of decreasing timings instead). Ring-bus seems better (so far) but my guess is that Zen 3 will end that benefit.



On Ryzen 3000 systems the timings can improve the simulation results by more than 30 % on a given memory frequency. That is no small potato.
Simbelmynė is offline   Reply With Quote

Old   July 11, 2020, 15:33
Default
  #12
Senior Member
 
Simbelmynė's Avatar
 
Join Date: May 2012
Posts: 427
Rep Power: 12
Simbelmynė is on a distinguished road
Quote:
Originally Posted by ships26 View Post
@flotus:

@Simbelmynė: Thank you for your feedback as well! I believe the 2 x 2690v2 would be a bit too expensive for me (not that I wouldn't like to say I have a 20 core monster-machine in my bedroom, of course! ).

This is off-topic but a friend of mine at the university had a Scanning Electron Microscope (SEM) installed in his bedroom =)
ships26 likes this.
Simbelmynė is offline   Reply With Quote

Old   July 11, 2020, 16:39
Default
  #13
New Member
 
Francisco
Join Date: Sep 2018
Location: Portugal
Posts: 14
Rep Power: 3
ships26 is on a distinguished road
Fair enough! Thank you for clearing that up! I'll have to read up on memory and find some memory tests with both processors. The 9800x looks like a better choice, though.

Quote:
Originally Posted by flotus1 View Post
I recommend against the Noctua NH-U12s CPU cooler. Compared to dual-tower coolers from other brands in the 50$ price range, it is pretty weak.
May I ask why? I have found the reviews to be good, but I do reckon that if I ever resort to overclocking it is not enough.
ships26 is offline   Reply With Quote

Old   July 11, 2020, 17:05
Default
  #14
New Member
 
Francisco
Join Date: Sep 2018
Location: Portugal
Posts: 14
Rep Power: 3
ships26 is on a distinguished road
Quote:
Originally Posted by Simbelmynė View Post
Agreed-ish, the difficulty (as I see it) is finding reliable information on how to set the timings in Intel systems. Primary timings are easy, but secondary and tertiary are very difficult to find. If Intel systems behave in a similar manner as AMD then secondary and tertiary timings are very important. Finally, Zen improve with increasing infinityband frequency, and as such, 3600 MHz (up to 3800 or more with XT series) is the sweet-spot for AMD. Intel should benefit more when going higher (but we can discuss the benefits of decreasing timings instead). Ring-bus seems better (so far) but my guess is that Zen 3 will end that benefit.

Man, where do you guys learn all of this? I have a ton to learn about...
Anyway, I want to make the most out of the system, but wouldn't the price of memory above 3600MHz go up very quickly?


Quote:
Originally Posted by Simbelmynė View Post
On Ryzen 3000 systems the timings can improve the simulation results by more than 30 % on a given memory frequency. That is no small potato.

I think I remember seeing that on the benchmark thread, it is a huge difference! I had the "lower" end alternative builds for the i5 and the Ryzen 5 sort of closed, with the latter one using 3200MHz RAM due to the official support for that. I suppose I should rethink that if I'm to consider than as an option.



Quote:
Originally Posted by Simbelmynė View Post
This is off-topic but a friend of mine at the university had a Scanning Electron Microscope (SEM) installed in his bedroom =)

Oh wow! Imagine bringing someone over and having one of those just casually lying around
ships26 is offline   Reply With Quote

Old   July 12, 2020, 01:56
Default
  #15
Senior Member
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 2,500
Rep Power: 35
flotus1 will become famous soon enoughflotus1 will become famous soon enough
Quote:
Originally Posted by ships26 View Post
May I ask why? I have found the reviews to be good, but I do reckon that if I ever resort to overclocking it is not enough.
It is not a bad cooler. Noctua generally makes pretty high quality stuff, and this is no exception.
But when it comes to CPU air coolers, size matters. And the NH-U12s is just way smaller than its competitors in the same price range, leading to significantly lower performance, compared to dual-tower designs.
That means the fan has to spin faster, creating more noise. It doesn't help that the included fan has these weird stator blades to increase static pressure, leading to a rather unpleasant noise. Noctua has been experimenting for years with active noise reduction, on a similar fan design. Because this one creates a rather narrow frequency noise that is easier to eliminate with ANR.
Trust me on this one, I have this CPU cooler. But it got relegated into one of the low-power systems.

Edit: found the video
https://www.youtube.com/watch?v=-atLQYr14zI
__________________
Please do not send me CFD-related questions via PM
flotus1 is offline   Reply With Quote

Old   July 12, 2020, 08:53
Default
  #16
New Member
 
Francisco
Join Date: Sep 2018
Location: Portugal
Posts: 14
Rep Power: 3
ships26 is on a distinguished road
Quote:
Originally Posted by flotus1 View Post
Trust me on this one, I have this CPU cooler. But it got relegated into one of the low-power systems.
You have convinced me. Thanks for taking the time to explain it!
I was checking some alternatives in the mean time, and, according to Noctua's website, the 14S, should be enough even for heavy overclocking, so it would be more silent. Unluckily, it won't fit inside the case (which has already been gifted to me).
From Noctua I could only find single fan + dual heat sink setups that both fit the case and are at the 14S's level or better: the NH-D15S or the NH-D14, both at a little above the 12S's price. The NH-D15S doesn't require any extra mounting kits for the LGA2066, so I could go for that.

I'll also be checking Cooler Master and Corsair, to see if I can find any better deals, but the NH-D15S seems like a good idea if I can't find any good 140mm fans that fit the case.


Quote:
Originally Posted by flotus1 View Post
Oh, and this is really cool! I can see it being an interesting option for servers as well.
ships26 is offline   Reply With Quote

Old   July 12, 2020, 14:10
Default
  #17
Senior Member
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 2,500
Rep Power: 35
flotus1 will become famous soon enoughflotus1 will become famous soon enough
Scythe Fuma 2 is an excellent choice. High performance, quiet, relatively compact and good value overall.

Offtopic: I don't think anyone would be willing to spend extra for lower noise in servers. Quiet operation must be very low on the priority list. If it is on a list at all.
__________________
Please do not send me CFD-related questions via PM
flotus1 is offline   Reply With Quote

Old   July 14, 2020, 14:46
Default
  #18
New Member
 
Francisco
Join Date: Sep 2018
Location: Portugal
Posts: 14
Rep Power: 3
ships26 is on a distinguished road
I think I'll follow your suggestion once more. It's looks like pretty good cooler for a great price, even with the ports!

Thanks for the suggestion, flotus!



Quote:
Originally Posted by flotus1 View Post

Offtopic: I don't think anyone would be willing to spend extra for lower noise in servers. Quiet operation must be very low on the priority list. If it is on a list at all.

Regarding this... Yes, you're right. It really isn't, unless there's a noisy one in my basement. But that's not really going to happen anytime soon, if it ever does, and I'm sure there's other ways to soundproof it.
ships26 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
CPU choice zero_custom Hardware 3 January 12, 2016 07:23
Multi 5820k CPU in a desktop or workstation? sharonyue Hardware 5 December 23, 2015 14:35
Superlinear speedup in OpenFOAM 13 msrinath80 OpenFOAM Running, Solving & CFD 18 March 3, 2015 05:36
Star cd es-ice solver error ernarasimman STAR-CD 2 September 12, 2014 00:01
OpenFOAM 13 Intel quadcore parallel results msrinath80 OpenFOAM Running, Solving & CFD 13 February 5, 2008 05:26


All times are GMT -4. The time now is 09:43.