Need advice/recommendation - dual X5690 -> Ryzen 1700x?

Would you upgrade from a dual X5690 to a Ryzen 7 1700x?

  • Yes, go for, you will not regret it!

    Votes: 2 40.0%
  • Not worth upgrading...

    Votes: 3 60.0%

  • Total voters
    5

pjmssn

Member
Aug 17, 2017
89
11
71
Hello all,

I just joined the forum as a member and look forward to contributing.
I currently have a Dell T5500 workstation with dual X5690 (3.46GHz) and 72GB of RAM (DDR3) (~1575 cb) that I use for the following tasks:
- scientific computing (parallel c++, python with MPI)
- finite element simulations (usually several long simulations in parallel)
- video editing
The processors are pretty much loaded 24/7 these days. Everything works great but I know the cpus are many generations behind current cpus and the power consumption (~300W 24/7) makes me think about possibly upgrading.
I definitely need the processing power and multiple threads.
I am considering getting a Ryzen 7 1700x system with 64GB of RAM to replace my dual-X5690 machine, the new system would give me pretty much the same processing power with less power consumption, a more recent CPU architecture and more future proof system.

Do you think it would be a good upgrade or I owuld barely see the difference and should hold on to y current workstation?
Any additional drawbacks/advantages in going with a Ryzen system?

Thanks for your help and advice.
Phil.
 

aigomorla

CPU, Cases&Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
20,839
3,174
126
no ecc ram on the 1700 kills the upgade...

wait for the EPYC with ECC support...

your building workstation machine which is responsible for accurate and preise calculations.
that means ECC RAM....

This aint a gaming rig with RGB LED's... :p
 
Feb 25, 2011
16,776
1,466
126
I agree with aigomorla re: ECC RAM and stability w/ pro workloads.

Performance-wise, Ryzen 1700X bench is 1541 cb. http://www.anandtech.com/bench/product/1852 You're going from 12c/24t to 8c/16t, which is a big deal, even if the individual cores on the Ryzen are faster and more modern.

I depends on your workload, though - particularly for the scientific stuff, where you're presumably writing some of your own code - if you can make use of the newer features of Ryzen, you'll get an outsized performance boost.

If you're actually using that 72GB of RAM, there are some caveats - 64GB is the maximum a single Ryzen 7 CPU/motherboard supports. And it would cost around $1k to fill those slots. Which means paying a lot of money for no real improvement, if you're memory bottlenecked.

But is also means that the opportunity cost of going from a Ryzen 7 to, say, a Threadripper 1920X, (+$500 or so) is actually a relatively modest % of the overall system budget. And that would not only get you additional RAM support (you can always add more later) but it would also get you a substantial CPU performance boost.

From a simple price/performance standpoint, it would probably be a better move to get another off-lease 5500 on eBay with another pair of X5690s and add it to your pool of compute resources, using one machine for one task while the other does automated of batch jobs. Or rewrite some of your code to leverage GPGPU APIs and drop in a Titan. But I think Threadripper is in your future, unless you wait long enough to get an off-lease EPYC. You can go really nuts with used server hardware.
 
Last edited:

pjmssn

Member
Aug 17, 2017
89
11
71
Thank you so much for the great advice. I did not think about the RAM limitation and the non-ECC support.
You made great suggestions and I have decided to pass on an upgrade to the Ryzen 7 1700x. I very often need/wish I had more than 64GB of RAM to run large models and the ability to add more in the future is very important (my current T5500 is maxed out at 72GB).
I will wait for the EPYC or consider a threadripper with ECC memory, it gives me more time to get the funds for a new workstation. For now, I will likely get a second Xeon workstation with slightly newer processor(s) than my X5690s to help with the workload.
 

XavierMace

Diamond Member
Apr 20, 2013
4,307
450
126
Have you tested if the T5500 will work with Quad Rank DIMM's? I know they're not "officially" supported, but I've used 16Gb DIMM's on a variety of systems running the Intel 5520 chipset and 5600 series Xeon's. They aren't cheap, but it would get you more memory on your current box. The same memory was used up to the Sandy Bridge era Xeon's (E5 V2's) so if you're looking to pick up one of those, the memory wouldn't go to waste.
 

pjmssn

Member
Aug 17, 2017
89
11
71
This is a great suggestion, thanks! I did not know about Quad Rank DIMMs. I searched on ebay and found 128GB (8*16GB) DDR3 ECC for $350. I will need to get an additional stick since the T5500 takes 9 RAM sticks. I am also thinking about getting an additional workstation with a E5-1650 V2 with 128GB RAM for the jobs not requiring many cores, but more memory.
 

BambiBoom

Junior Member
Dec 29, 2016
5
3
36
Hello all,

I just joined the forum as a member and look forward to contributing.
I currently have a Dell T5500 workstation with dual X5690 (3.46GHz) and 72GB of RAM (DDR3) (~1575 cb) that I use for the following tasks:
- scientific computing (parallel c++, python with MPI)
- finite element simulations (usually several long simulations in parallel)
- video editing
The processors are pretty much loaded 24/7 these days. Everything works great but I know the cpus are many generations behind current cpus and the power consumption (~300W 24/7) makes me think about possibly upgrading.
I definitely need the processing power and multiple threads.
I am considering getting a Ryzen 7 1700x system with 64GB of RAM to replace my dual-X5690 machine, the new system would give me pretty much the same processing power with less power consumption, a more recent CPU architecture and more future proof system.

Do you think it would be a good upgrade or I owuld barely see the difference and should hold on to y current workstation?
Any additional drawbacks/advantages in going with a Ryzen system?

Thanks for your help and advice.
Phil.


pjmssm,

If the applications are heavily threaded / parallel, the better solution is going to be to use a dual Xeon system with both a high core/thread count and if possible a higher clock speed to run the single-threaded components. There needs to be plenty of RAM and a fast disk system. In simulation and analytics, there is the need for high double precision, so, as others have commented, ECC RAM and a workstation GPU are importnt. Because video editing is GPU based and not CPU based- as is single image rendering- the application demand add up to having a system that has strong performance in every subsystem. Note the limitation of Ryzen in the number of PCIe lanes.

My earlier solution to this kind of system was to upgrade a Dell Precision T5500:

Dell Precision T5500 (2011) (Revised) > 2X Xeon X5680 (6-core @ 3.33 / 3.6GHz), 48GB DDR3 1333 ECC Reg./ Quadro K2200 (4GB ) / PERC H310 / Samsung 840 250GB + WD RE4 Enterprise 1TB / M-Audio 192 sound card + Logitech z313 speakers /> 875W PSU > Windows 7 Professional 64 > HP 2711x (27", 1920 X 1080) /
[ Passmark system rating = 3844 > CPU = 15047 / 2D= 662 / 3D= 3550 / Mem= 1785 / Disk= 2649] (12.30.15)

Note that this system uses a PERC H310 RAID controller that converts the disk system to 6GB/s

> And that may resemble your current system somewhat.

However, for Matlab, Wolfram Mathematica, and various mech'l /thermal simulations, I needed more cores, higher clock speed, greater memory bandwidth, faster disk, and etc.

I purchased for $270:

HP z620_1 (Original) Xeon E5-1620 (4-core @ 3.6 /3.8GHz) / 8GB (1X 8GB DDR3-1333) / AMD Firepro V5900 (2GB) / Seagate Barracuda 750GB + Samsung 500GB + WD 500GB
[ Passmark System Rating= 2408 / CPU= 8361 / 2D= 846 / 3D = 1613 / Mem =1584 / Disk = 574 ] 7.13.16

> and upgraded this to:

HP z620_1 (2012) (Rev 4) 2X Xeon E5-2690 (8-core @ 2.9 / 3.8GHz) / 64GB DDR3-1600 ECC reg) / Quadro K2200 (4GB) + Tesla M2090 (6GB) / HP Z Turbo Drive M.2 (256GB) + Samsung 850 Evo 250GB + Seagate Constellation ES.3 (1TB) / Creative Sound Blaster X-Fi Titanium PCIe sound card + Logitech z2300 2.1 Sound / 800W / Windows 7 Professional 64-bit > 2X Dell Ultrasharp U2715H (2560 X 1440)
[ Passmark System Rating= 5675 / CPU= 22625 / 2D= 815 / 3D = 3580 / Mem = 2522 / Disk = 12640 ] 9.25.16 Single Thread Mark = 1903
[ Cinebench R15: CPU = 2209 cb / Single core 130 cb / OpenGL= 119.23 fps / MP Ratio 16.84x] 10.31.16

However, today for the GPU choice, I would recommend using one of the new Pascal Quadros which are as fast as NVIDIA GTX. For you use, a Quadro P4000 8GB.

I needed to consolidate the 3D modeling and simulation systems into a single system, with an emphasis on improving the single-thread performance for 3D modeling. A couple of months ago I built a new HP z620 from parts using a P2000 5GB and the results were very good:

HP z620_2 (2017) (Rev 1) > Xeon E5-1680 v2 (8-core@ 4.3GHz) / z420 Liquid Cooling / 64GB DDR3-1866 ECC Reg / Quadro P2000 5GB / HP Z Turbo Drive M.2 256GB + Intel 730 480GB + Seagate Constellation ES.3 1TB / ASUS Essence STX PCIe sound card + Logitech z2300 2.1 Sound / 825W PSU /> Windows 7 Prof.’l 64-bit > 2X Dell Ultrasharp U2715H (2560 X 1440) /
[ Passmark Rating = 6322 / CPU rating = 17178 / 2D = 852 / 3D= 9012 / Mem = 3032 / Disk = 14227 / Single Thread Mark = 2339 [7.3.17]
[ Cinebench R15 = cb1214 (CPU) / 153 (Single Threaded) / 150.77 (OpenGL) MP Ratio 7.92x / Accuracy 99.6% ] 7.21.17

Note that the Xeon E5-1680 v2 is one of the rare Xeon E5's that may be overclocked and with liquid cooling, it's possible to run it at 4.3GHz on all cores. A fellow in the UK is running one at 4.7GHz!

It is also possible to use a Quadro and GTX in the same system and using a Displayport switch, change to the GTX for video rendering.

In summary, I'd say that Ryzen will not be adequate for your needs- not enough cores, not enough PCIe lanes. Depending on the budget, consider upgrading a used dual processor Dell Precision or HP z620 or z820. If the budget is higher, look into the AMD Threadripper 1920X 12-core @ 3.5 /4.0Ghz (64 PCIe lanes), 64GB with room for an eventual 128GB of RAM, a Quadro P4000 or P2000 with a GTX 1080 Ti on a switch, and a Samsung 960 Evo 512GB M.2 boot drive- that kind of thing. From experience, I believe results will be better if the system can have a single processor. There is only one 1920X on Passmark, having a compute rating of 23575.7. Compare that to the 22625 of the 16 cores of the HP z620_1's dual Xeon E5-2690's. the difference is though, the Z620 case, chassia, motherboard/ power supply /OS cost $270- less than the Threadrippers motherboard alone would cost and the E5-2690 cost $310 for the pair while a Threadripper 1920X is $800.

What is your budget?

BambiBoom
 

richaron

Golden Member
Mar 27, 2012
1,357
329
136
I don't get the ECC arguments so far in this thread. There are multiple AM4 motherboards with ECC support, surely you know this? Or am I missing something? AFAIK one of the biggest advantages of Ryzen and Thread Ripper is the ability to run ECC RAM and have a cheaper workstation.

I'm also looking for a new rig to run sciency stuff. The problem is than Ryzen/TR likes higher memory speeds so atm even the fastest 32GB 2400/2666 ECC costs almost the same as 64GB of much faster non-ECC. Also consider that many of these tasks can be GPU accelerated, so it's worth considering if there's a point of ECC system RAM without also an ECC GPU. Lastly I've seen a couple of articles about whether ECC is even necessary for scientific computing (error rates and other problems vs P values).

Personally for large data sets on a budget I'm looking at non-ECC system with a 3d Xpoint (or similar higher reliability) SSD instead of lots of RAM. With more budget you can then start thinking about Pro GPUs, lots of RAM, ECC RAM, and SSD arrays.
 
Last edited:

Torn Mind

Lifer
Nov 25, 2012
11,567
2,626
136
I don't get the ECC arguments so far in this thread. There are multiple AM4 motherboards with ECC support, surely you know this? Or am I missing something? AFAIK one of the biggest advantages of Ryzen and Thread Ripper is the ability to run ECC RAM and have a cheaper workstation.

I'm also looking for a new rig to run sciency stuff. The problem is than Ryzen/TR likes higher memory speeds so atm even the fastest 32GB 2400/2666 ECC costs almost the same as 64GB of much faster non-ECC. Also consider that many of these tasks can be GPU accelerated, so it's worth considering if there's a point of ECC system RAM without also an ECC GPU. Lastly I've seen a couple of articles about whether ECC is even necessary for scientific computing (error rates and other problems vs P values).

Personally for large data sets on a budget I'm looking at non-ECC system with a 3d Xpoint (or similar higher reliability) SSD instead of lots of RAM. With more budget you can then start thinking about Pro GPUs, lots of RAM, ECC RAM, and SSD arrays.
Those boards are hard to find then. There are TR4 motherboards with ECC do exist and are listed on pcpartpicker. None are listed there for AM4. Not that pcpartpicker is perfectly comprehensive, but they have most boards on Newegg and elsewhere.
I think he made it beyond clear that his tasks do gobble up memory and leave him wanting more. He would not be buying more sticks and considering another box with a full 128GB of. RAM
 

richaron

Golden Member
Mar 27, 2012
1,357
329
136
Those boards are hard to find then. There are TR4 motherboards with ECC do exist and are listed on pcpartpicker. None are listed there for AM4. Not that pcpartpicker is perfectly comprehensive, but they have most boards on Newegg and elsewhere.
I think he made it beyond clear that his tasks do gobble up memory and leave him wanting more. He would not be buying more sticks and considering another box with a full 128GB of. RAM

Oh, I never found ECC boards hard to find. I guess because I know high end ASUS boards have supported ECC on AMD for a while :p

Just go the the manufacturer page and check the specs.

Also on the point of more RAM, that's why I brought up the idea of 3d Xpoint SSDs. At some point RAM becomes prohibitively expensive, then these SSDs (with higher R/W resilience than conventional NAND) will allow for larger data sets. Obviously slower than system RAM, but also faster/lower latency than conventional options.

Lastly, Thread Ripper with lots of ECC RAM slots should also be an obvious option. And I'm sure I've read a few reviews with it spanking the competition in sciency benchmarks.
 
Last edited:

Torn Mind

Lifer
Nov 25, 2012
11,567
2,626
136
Oh, I never found ECC boards hard to find. I guess because I know high end ASUS boards have supported ECC on AMD for a while :p

Just go the the manufacturer page and check the specs.

Also on the point of more RAM, that's why I brought up the idea of 3d Xpoint SSDs. At some point RAM becomes prohibitively expensive, then these SSDs (with higher R/W resilience than conventional NAND) will allow for larger data sets. Obviously slower than system RAM, but also faster/lower latency than conventional options.

Lastly, Thread Ripper with lots of ECC RAM slots should also be an obvious option. And I'm sure I've read a few reviews with it spanking the competition in sciency benchmarks.
Don't know who is more at fault, Newegg or ASUS, but they could have put it in the spec sheet on Newegg's page so one doesn't have to go to ASUS's website to find out which board has the ECC support and which don't.

Also, some A320 boards also support ECC, so it's inclusion is also on more affordable boards.
 

pjmssn

Member
Aug 17, 2017
89
11
71
At this point, I will not go for the Ryzen 1700x. I like the idea of getting a previous generation server with lots of cores. Whether or not ECC memory is required for my workflow, I don't know, I have read many different opinions and from my experience, I never had any issue with memory failing during computation (I had bad memory or mother boards though).
Some of my applications are very sensitive to memory bandwidth, much more so than CPU clock. Video encoding is a very small fraction of my activities, so I would like to focus on raw computing power with the option to get a good gpu at a later time.
I would consider purchasing a new workstation/ server with at least 64GB of RAM (Quad channel) with possibility to increase to 128Gb or more in the future and dual Xeon with at least 12 cores total (more would be welcome). My current budget is around $1500, but I could stretch it a bit if needed.
I would welcome any suggestions for a system, I am not very familiar with the Xeon models and generations. I will be digging through ebay listings.
 

BambiBoom

Junior Member
Dec 29, 2016
5
3
36
At this point, I will not go for the Ryzen 1700x. I like the idea of getting a previous generation server with lots of cores. Whether or not ECC memory is required for my workflow, I don't know, I have read many different opinions and from my experience, I never had any issue with memory failing during computation (I had bad memory or mother boards though).
Some of my applications are very sensitive to memory bandwidth, much more so than CPU clock. Video encoding is a very small fraction of my activities, so I would like to focus on raw computing power with the option to get a good gpu at a later time.
I would consider purchasing a new workstation/ server with at least 64GB of RAM (Quad channel) with possibility to increase to 128Gb or more in the future and dual Xeon with at least 12 cores total (more would be welcome). My current budget is around $1500, but I could stretch it a bit if needed.
I would welcome any suggestions for a system, I am not very familiar with the Xeon models and generations. I will be digging through ebay listings.

pjmssn,

With regard to your search for a high-parallelization compute load workstation, and within reasonable proximity to your budget, consider of this kind of system, which until a couple of weeks ago was my simulation, analysis, and CPU rendering - the high compute load applications- system:

HP z620_1 (2012)(Rev 5) 2X Xeon E5-2690 (8-core @ 2.9 /3.8GHz) / 64GB (4X 8GB +4X 2GB DDR3-1600) / Quadro K2200 (4GB) / Samsung 850 Evo 250GB + Seagate Constellation ES.3 1TB / 800W > Windows 7 Professional 64-bit >

That is running two Xeon E5-2690 CPU's which means 16- cores at 2.9GHz, a Turbo rate of 3.8GHz- one of the highest every for dual processor Xeons (4.0GHz was the highest Xeon E5-2687w v2)- and an all-core compute clock speed of 3.4GHz. Those processors cost $2,050 each when new. The Passmark CPU Mark of 22625 which reflect the number of total compute cycles per unit time is very good. For comparison, I have a Dell Precision T5500 with two Xeon X5680s 6C@ 3.33 /3.6Ghz and the Passmark score is 15047. Besides the higher memory bandwidth of Xeon E5s: 51.2 GB/s as compared to the 32GB/s of the X5680, the RAM capacity of the z620 is 192GB of DDR3-1600 ECC registered instead of DDR3-1333. One important advantages of the Z620 system over the T5500 is the SATAIII 6GB/s disk system instead of SATAII 3GB/s. The USB 3.0 instead of USB 2.0 is also welcome. The single thread rate is quite good and with a strong GPU, the z620 are perfectly practical for 3D CAD. These are beautifully made =very heavy and is also the quietest workstation I've had. I liked the z620 so much I built a new one around a Xeon E5-2680 v2 and Quadro P2000 5GB. The z620 can use AHCI M.2 drives- I used an HP Z Turbo M.2 256GB (Samsung SM951) in my z620 with the files stored on a Samsung 850 Evo scratch disk and it's possible to use NVMe M.2 with the Samsung 950 Pro, running on the legacy BIOS which installs an NVMe module. For those more technically braver than I, a NVMe module can be added to the system BIOS. There are starting to be an increasing number of easier to use NVMe modules fixes.

Buying a system using depreciated CPUs is remarkably less expensive for comparable results. The nearest current equivalent of the E5-2690 is the Xeon E5-2667 v4 ( 8-core @ 3.2 /3.6Ghz). On Passmark, the average CPU Mark for a pair of Xeon E5-2667 v4's is 24952. That performance represents +10.3% increase in compute rate over the pair of E5-2690s but a pair of E5-2667 v4 cost $4,102. An appropriate complete system using the E5-2667 v4 I would estimate at about $7,000-8,000 if self-assembled and configured, but that would include a used Tesla coprocessor. On that subject, consider in future adding a Tesla or other GPU compute unit- you can use a GTX with the correct configuration.

With some careful shopping an E5-2690 system like the above can be completed for about $1,700. A low specification z620 is about $500, the two CPU's about $350, CPU riser $140, 64GB RAM about $170, Quadro K2200 $325, new Samsung 850 Evo : $100, Seagate ES.5 $100. If there is not to be a substantial 3D requirement, something could be saved by using a lower level Quadro. although I would suggest at least 2GB of video memory- 4GB is preferable as this relieves some compute tasks of the CPU's. If you go this route, be very careful as to the condition of the base system. There are a lot of off-lease, high-mileage systems that have been kicked around on pallets. Buy, if possible from a listing that shows the exact system you're buying inside and out.

Here's a similar system, but 48GB RAM- sold on Ebahhh for $1,700 on 4 Aug:

http://www.ebay.com/itm/HP-Z620-SolidWorks-CAD-Workstation-2x-E5-2690-16-Core-SSD-Quadro-K4000-Win10-/142444788863?hash=item212a5efc7f:g:1-oAAOSwsXVZZ-ke&nma=true&si=05sgvrAvFwdx6BfgaEI8HrWi2VA%3D&orig_cvip=true&rt=nc&_trksid=p2047675.l2557

> and that one has a looks reasonably good- although the "mint" comment is a bit optimistic. I like to see a lot of photos of the actual system. The seller appears knowledgeable and updated BIOS, drivers and etc. and the cosmetic quality is reasonable. The Quadro K4000 3GB is an older series, but quite good as well. Today I would buy a Quadro P1000 4GB ($330)- a Pascal Quadro with very good peformance for the cost and supports 4 monitors. This is the kind of listing to look for.

What kind of projects are you doing? Software?

Cheers,

BambiBoom
 

pjmssn

Member
Aug 17, 2017
89
11
71
Thank you so much for the suggestion, I placed an order for a z620 with dual E5-2690 and 64GB of RAM. I think it will fit my workflow very nicely and I am eager to start playing with it. I should have it within 7-10 days.
Regarding the use of the computer, I do quite a bit of projects as a consultant in the area of numerical multi-physics modeling, design and optimization, with lots of FEA simulations (some with discontinuous galerkin, heavily multi-threaded) some very large 3D, Monte-Carlo analysis requiring thousands of data points/simulations and some in-house code development, mostly C, C++ and python using MPI.
I have access to a computer cluster at my regular work (University), but I am not allowed to use it for consulting work.
Cheers,
Pjmssn