Core i7 vs Xeon: Workstation Build

Ali_N

Distinguished
Jun 9, 2015
63
2
18,535
Hi,

I need to decide between these two systems for 24/7 100% CPU loading:

system #1

CPU: Intel Core i7-4790K @ 4.00GHz
MB: Asus H97-PRO

system #2
CPU: Intel Xeon E3-1245 V3 @ 3.40GHz
MB: Supermicro X10SAE

common config:

RAM: ADATA XPG V3 8GB(2×4GB) DDR3 1600MHz CL9
SSD: Samsung 850 EVO 120GB
HDD: Western Digital Caviar Blue 1TB
ODD: Asus 24X DRW-24D3ST
PSU: EVGA SuperNova 650W G1 *Gold*
CPU Cooler: Cooler Master Hyper 212 EVO
Case: Cooler Master N300

I am going to work on a molecular dynamics project which will last for 1.5 years. The running time will vary between several minutes and 10 days depending on the model sizes. About 30% of the work needs such heavy computing (~10 days). So the system must withstand "100% CPU loading" for a "non-stop period of about 10 days".

I can afford 1150 socket CPUs only and, at this stage, I don't need any graphics card. Also, I don't plan on overclocking or gaming. So if you are asking why 4790k instead of 4790, the answer is its higher clock speed (4.0GHz vs 3.6GHz).

System #1 is cheaper and faster (4 cores @ 4.0GHz vs 4 cores @ 3.4GHz), but we know that system #2 is made for such tasks. Unfortunately, Supermicro X10SAE is the only available mobo with C226 chipset in my country and I've heard some bad things about it.

There are some forums saying that systems with Core i7 CPUs and non-workstation mobos are stable enough for 24/7 use, but most of them are for gaming purposes and only one core (and not all cores at 100% load) is engaged.

Sorry for the long text, but I needed to explain my concerns and I have to choose between two systems #1 and #2. Which one should I choose and why? Any suggestions are welcome.
 
Solution
Non-workstation systems are fine for your purpose; especially overclocking hardware, they are designed to handle additional stress. ASUS makes great motherboards, so that is a good choice. You also picked a good aftermarket CPU cooler, so that's covered. Everything looks great, though you may want to consider using a higher frequency of RAM to increase bandwidth. Also, not sure of the size of simultaneous computing but additional capacity may be beneficial as well. Computer memory is a major component for efficient computing.

During the initial release of the 4970K CPU, we had similar combinations running for weeks at full load, but with Z97 motherboards and G.Skill RAM. Reliability of high end overclocking hardware is superb, designed to withstand extreme heat and cold.
 

Ali_N

Distinguished
Jun 9, 2015
63
2
18,535


 

Ali_N

Distinguished
Jun 9, 2015
63
2
18,535
Thank you for quick reply.

I chose 1600MHz RAM because I do not want to overclock and 1150 socket CPUs support up to 1600MHz. I can increase the memory size to 16GB.

Can i ask whick asus z97 mobo I should pick for such a task? Are you suggesting z97 because of only its quality and durability over h97 or something else?
 


Ali_N,

If you are running NAMD or MATLAB and can configure the custom multi-threaded algorithms, I would suggest that instead of building that for a similar price, a better approach, having better performance and higher reliability,is to upgrade a Dell Precision T5500 or, better, a T7500. In any event, it is essential to use ECC error correcting RAM as in long running problems for example protein folding a single bit error is magnified by extrapolation and days of processing is wasted.. These systems can use dual 6-core CPU's up to the Xeon X5690 (6-core @ 3.47 / 3.73GHz) and the T7500 can accommodate up to 192GB of ECC RAM. You can imagine that the advantage of 12 cores / 24 threads running simultaneously at 3.47GHz and at the double precision accuracy required is assured by the ECC RAM, is going to produce more accurate results, reliability, and especially, the 24 threads is 2X the 8 core maximum of an LGA1150 system. These systems are based on server technology (T5500: 875W and T7500: 1100W power supplies) and running without errors in that mode.I have had three used Precisions and used them for five years without a single failure.

About four months ago I bought a Precision T5500 for $190 including shipping:

Dell Precision T5500 (2011) (Original): Xeon E5620 quad core @ 2.4 / 2.6 GHz > 6GB DDR3 ECC Reg 1333 > Quadro FX 580 (512MB) > Dell PERC 6/i SAS /SATA controller > Seagate Cheetah 15K 146GB > Windows 7 Professional 64-bit
[ Passmark system rating = 1479 / CPU = 4067 / 2D= 520 / 3D= 311 / Mem= 1473 / Disk= 1208]

The condition was excellent and the system performed reasonably well on the Xeon E5620.

I purchased:

Xeon X5680 (6-core @ 3.33 / 3.6GHz) $200
24GB PC3-10090R RAM (DDR3-1333 ECC registered) $120
Windows 7 Professional : $0 (Dell sent me a replacement OS disk at no charge to load the original software. This is activated by the COA serial number on the system)

> and I had from a previous upgrade of another system:

Quadro 4000 (2GB): value $250
Samsung 840 250GB SSD : value $80
WD RE4 1TB (SATA II) value : $40

So, the total purchase of new parts, plus the value of used parts = $880

And the performance was very good:

Dell Precision T5500 (2011) (Revision 1) > Xeon X5680 six -core @ 3.33 / 3.6GHz, 24GB DDR3 ECC 1333 > Quadro 4000 (2GB ) > Samsung 840 250GB / WD RE4 Enterprise 1TB > M-Audio 192 sound card > Linksys WMP600N PCI WiFi > Windows 7 Professional 64> HP 2711x (1920 X 1080)

[ Passmark system rating = 3339 / CPU = 9347 / 2D= 684 / 3D= 2030 / Mem= 1871 / Disk= 2234]

This system had the best performance of a single CPU T5500 in Passmark of 199 systems tested

Subsequently, I upgraded my main system and last week the Quadro K2200 from that system was exchanged to the T5500:

With Quadro K2200 (4GB (Current configuration): [ Passmark system rating = 3490 / CPU = 9178 / 2D= 685 / 3D= 3566 / Mem= 1865 / Disk= 2122] [Cinebench 15 > CPU = 772 OpenGL= 99.72 FPS]

And I tested the new Quadro K4200 (4GB) in the T5500:
[ Passmark system rating = 3585 / CPU = 9346 / 2D= 683 / 3D= 4708 / Mem= 1850 / Disk= 2202]

One of the limitations of this kind of system, besides the RAM speed of 1333MHz is the SATA II 3GB/s disk system and I've purchased for $60 a Dell PERC H310 SAS / SATA 6GB/s RAID controller. I'll use the WD Black 1TB from my main system and add a second new one for RAID 1.

To add a 2nd CPU requires a CPU /Memory /Fan riser - $100-$120, another X5680 ($200) and I can populate the three RAM slots on the riser with the original 3X 2GB RAM that arrived with the system. The risers for the T7500 are more common and actually less expensive and have 6 RAM slots.

My recommendation for your use is to consider buying a Precision T7500 with a lower end CPU, ensuring the overall condition and the presence of the OS COA. You didn't mention where you are, but if you can buy from the EU have a look on Ebay.de. If you can buy the CPU's and RAM frim the US, the prices are much better.

Then purchase:

1. A CPU /Memory /Fan riser if not present

2. 2X Xeon X5690 ( 6-core @ 3.47 / 3.73 GHz) (About $250 each)- When buying, contemplate the fact the original cost was $1660 each.

3. As much RAM as is affordable. TheT7500 has 12 slots on the motherboard and 6 on the riser. The X58 chipset uses triple memory, so buy in sets of 3 modules. If the budget is a problem buy 3X 8GB and place in them in the proper sequence and use whatever arrived in the system to fill other slots. There must be at least one module in the riser board.

4. An LSI 9260-8i SAS/ SATA RAID controller (About $100-$120)

5. An SSD of 480 to 512GB. I recommend Intel 730 or Samsung 850 EVO . This will have a partition for the OS and applications, and another partition for the working files. (About $250)

6. 2X 2TB WD Black or Seagate ES.3 Constellation Enterprise HDD.

7. Perhaps you have a graphics card. If not, for monitor output : Quadro FX 3800 (1GB)

If you need more computing power, consider a used Tesla GPU co-processing unit. These can be quite inexpensive and were made to accelerate NAMD and MATLAB applications.

http://www.ebay.com/sch/Graphics-Video-Cards-/27386/i.html?_from=R40&_nkw=Tesla+GPU&_sop=15

This seems complex and wise and patient shopping is required, but overall, I think it is an easier task to upgrade a working system- which may done gradually- you could start with a single CPU with 24GB RAM and add the 2nd CPU and more RAM later. The system could be run with one HD without the RAID card and so on. When I bought the T5500, I might've loaded Windows and my applications and used it as it arrived, upgrading as each new part arrived. This is still an easier tack than researching, ordering, assembling, configuring, and testing a new system from components and the performance results and reliability can be excellent.

Cheers,

BambiBoom.


HP z420 (2015) > Xeon E5-1660 v2 six-core @ 3.7 / 4.0GHz > 32GB DDR3 ECC 1866 RAM > Quadro K4200 (4GB) > Intel 730 480GB (9SSDSC2BP480G4R5) > Western Digital Black WD1003FZEX 1TB> M-Audio 192 sound card > Logitech z2300 > Linksys AE3000 USB WiFi > 2X Dell Ultrasharp U2715H (2560 X 1440) > Windows 7 Professional 64 >
[ Passmark Rating = 5064 > CPU= 13989 / 2D= 819 / 3D= 4596 / Mem= 2772 / Disk= 4555] [Cinebench R15 > CPU = 1014 OpenGL= 126.59 FPS] 7.8.15

Pending upgrade: HP /LSI 9212-4i PCIe SAS /SATA HBA RAID controller, 2X Seagate Constellation ES.3 1TB (RAID 1)
 
Solution

Ali_N

Distinguished
Jun 9, 2015
63
2
18,535


Thank you for reply. I am running LAMMPS and Accelrys Materials Studio. Only 4 models of Dell Precision are available in here, and the cheapest one (T1700), which is similar in specification to my system #2, is 1.5x more expensive than it.
I can't decide between systems #1 and #2, because it is a tradeoff between speed, cost and reliability.
 
Ali_N wrote:

Thank you for reply. I am running LAMMPS and Accelrys Materials Studio. Only 4 models of Dell Precision are available in here, and the cheapest one (T1700), which is similar in specification to my system #2, is 1.5x more expensive than it.
I can't decide between systems #1 and #2, because it is a tradeoff between speed, cost and reliability.


Ali_N,

If it is not possible to purchase on Ebay within a fairly wide selection, then I think clearly the Xeon / Supermicro solution is preferable and with the maximum (32GB) ECC RAM. The Xeon E3-1245 v3 is excellent, producing a CPU average score of 9515 and ranking of No. 92 on Passmark. On Passmark, a Xeon E3-1245 v3 / Supermicro X10SAE / 16GB Intel HD P4600 ( integrated graphics) / Seagate ST240HM000-1G512 (SSD) - that is using integrated graphics :

Rating = 3109 / CPU = 10007 / 2D = 940 / 3D = 653 / Mem= 2407 / Disk = 3353

The "X10SAE" version produced noticeably higher CPU results than the "X10SAT" or "X10SL7-F".

So, I would say within the restrictions, you've made very good selections, with the Xeon system the clear favourite. The Intel 4600 IG is actually excellent in 2D and even in 3D equal to dedicated cards such as GeForce 620.

It's an intriguing sounding project- Graphene semiconductors?


Cheers,

BambiBoom
 

Ali_N

Distinguished
Jun 9, 2015
63
2
18,535
Ali_N,

If it is not possible to purchase on Ebay within a fairly wide selection, then I think clearly the Xeon / Supermicro solution is preferable and with the maximum (32GB) ECC RAM. The Xeon E3-1245 v3 is excellent, producing a CPU average score of 9515 and ranking of No. 92 on Passmark. On Passmark, a Xeon E3-1245 v3 / Supermicro X10SAE / 16GB Intel HD P4600 ( integrated graphics) / Seagate ST240HM000-1G512 (SSD) - that is using integrated graphics :

Rating = 3109 / CPU = 10007 / 2D = 940 / 3D = 653 / Mem= 2407 / Disk = 3353

The "X10SAE" version produced noticeably higher CPU results than the "X10SAT" or "X10SL7-F".

So, I would say within the restrictions, you've made very good selections, with the Xeon system the clear favourite. The Intel 4600 IG is actually excellent in 2D and even in 3D equal to dedicated cards such as GeForce 620.

It's an intriguing sounding project- Graphene semiconductors?


Cheers,

BambiBoom[/quotemsg]

Regarding 24/7 running of a system at full load, Xeon / Supermicro solution is preferable, as you stated. But, I am still hesitant about the quality of supermicro x10sae mobo. I'm not sure if a "Xeon / non-workstation ASUS mobo" system is a proper alternative.

Comparing Passmark rankings of e3-1245 v3 and core i7-4790k (92 vs 51) makes the decision even harder.

I really appreciate your detailed advice. Sometimes suggesting a solution quantitatively makes more sense. Can you give me the link of that Xeon E3-1245 v3 / Supermicro X10SAE benchmark?
 


Regarding 24/7 running of a system at full load, Xeon / Supermicro solution is preferable, as you stated. But, I am still hesitant about the quality of supermicro x10sae mobo. I'm not sure if a "Xeon / non-workstation ASUS mobo" system is a proper alternative.

Comparing Passmark rankings of e3-1245 v3 and core i7-4790k (92 vs 51) makes the decision even harder.

I really appreciate your detailed advice. Sometimes suggesting a solution quantitatively makes more sense. Can you give me the link of that Xeon E3-1245 v3 / Supermicro X10SAE benchmark?[/quotemsg]

Ali_N,

I have the highest confidence in Supermicro motherboards both in design and construction as they are specialists in server hardware. including Xeon E7 4 and 8 CPU solutions that may use 1TB of RAM- placed under the most demanding continuous operation with extremely high reliability. Also, I believe it is absolutely essential in this kind of work to use ECC RAM to avoid accumulated / extrapolated errors. ASUS makes very fine workstation motherboards as well, but regardless use ECC and I would configure the system to be able to use the full 32GB. If funds only permit purchasing 16GB, have it be 2X 8GB,

And, Supermicro makes a product line that is well worth consideration, the Supermicro "Superworkstations". These are cases in which the motherboard and power supply is mounted and configured, and including the cooling solution. These are such that the use need only add the CPU, RAM, GPU if any, and drives- essentially, plug in components. In this way a high performance, high reliability system- these are also said to be very quiet- may be configured quickly and using the highest quality parts. The power supply may be assumed to be of server quality. The unit for LGA1150 is the:

SuperWorkstation 5038A-iL > $399.99

http://www.supermicro.com/products/system/tower/5038/SYS-5038A-iL.cfm
http://www.superbiiz.com/detail.php?name=SY-538AIL

> and this unit includes the "Super X10SAE " motherboard. In the Passmark results, it is that particular "SAE" variant - up to about 25% higher- that produces significantly higher performance from the E3-1245 v3 on Passmark baselines. There were only three results however, so the sample is very small.

It's not possible to link to the Passmark results as these are the result of "Advanced" baseline searches in which each component- CPU, motherboard, GPU, and drives is entered and the results do not have a URL. If you send me a private message with your email address, I would be pleased to send screen capture images -*.JPG or *.PDF of these results.

Integrated Graphics: I understand your idea of economy in using integrated graphics, but I recommend you consider a dedicated GPU on any level- but the more CUDA cores the better. The Intel IG is a far better performer than one might think, but it does share the system RAM and subtracts some level of CPU performance from the computing tasks as it running the video subsystem continuously. I believe the Intel HD4600 may use up to 2GB of the system memory.

Even on systems using a dedicated card, replacing it with a higher performance GPU will show an increase in the CPU score. With my Precision T5400, using an FX 4800 (1.5GB) is accompanied by a CPU score of 8528, while changing to the more modern Quadro 4000 (2GB), and with everything else identical, the CPU score is 8625. With the HP z420, using a Quadro K2200, the CPU = 13941, and with the K4200, the CPU = 13989. The older the system, the more pronounced the improvement: On a Precision 390 (Xeon X3230 4-core 2.66GHz)), with a Quadro K600 (1GB), the CPU = 2953 and a half hour later with a Quadro K4200 (4GB) the CPU= 3467. The 3D score also improves from 862 to 4067 and that is demonstrating the dramatic corresponding relative increase in the computational power of the GPU to move polygons. The differences are actually more pronounced as Passmark scores are relative and as the average performance is higher, the score is reduced relatively. The same components' score of 6 months ago will appear lower today because the current weighted average increases.

I believe that the future of high performance workstations may well evolve the CPU into more a kind of GPU/ computing controller in which many, modular GPU computing units may be linked in parallel. This is the configuration of the fastest supercomputers are made today, for example the the Oak Ridge "Titan". There are of course GPU co-processors for workstations already- the NVIDIA Telsa and Xeon Phi and even separate enclosure to add a stack of them.

The current and high specification GPU accelerators are very expensive new- $2500+ but you can buy obsolete models on Ebay, sometimes inexpensively:

http://www.ebay.com/sch/Graphics-Video-Cards-/27386/i.html?_from=R40&_sop=15&_nkw=nvidia+tesla

http://www.ebay.com/itm/NVIDIA-Tesla-M2090-6GB-GDDR5-PCIe-x16-GPU-Computing-Processor-VIDEO-CARD-/281723734547?pt=LH_DefaultDomain_0&hash=item41980b0a13

> These are very reasonable as they are only useful in accelerating custom algorithm computing- exactly what you're doing. I mention this as a possible upgrade- along with the dedicated GPU to compensate for the proposed system limitation of only 4 Cores / 8 threads- the GPU and coprocessor is in effect similar to adding more CPU cores.

A very interesting project!

Cheers,

BambiBoom
 

TRENDING THREADS