Dual Xeon CPU vs Single Intel i7 Low Benchmark Results

[JAO]

Hi,

Wanted to ask if someone out there has had this or a similar experience. I recently did an Intel i7 build which I called it UAS1 this particular build is significant and runs well. However, I have come to a point that I need more horsepower and the ability to process jobs a lot quicker. I then opted to look on the passmark website to gather better hardware and create a much better build then the intel. I then went with a dual Xeon build which I called UAS2. Please refer to the attached file for each of the machine's specs and Passmark benchmarks.

I am now a bit concerned that the second build has much better hardware but the benchmark results are lower. Not only are the results lower but the overall performance use of UAS2 machine is noticeable. I wanted to know if anyone has had an experience like this or if there is a way to improve the over all performance of the second machine to get significant passmarks scores.

Much appreciated to all that can assist in this matter.

[David (PassMark)]

What is really important is know how your particular application runs and what hardware it can make use of. What does "process jobs" really mean?

Maybe your software it limited by the speed of your internet link? If so then getting dual Xeons isn't going to help at all.

Also in terms of single threaded performance you have gone significantly backwards. 2226 with the i7 and only 1767 for the Xeon. So if your jobs are only single threaded you have 27 CPU cores sitting idle. Likely the Xeon is a worse CPU until you get to maybe 6+ threads. But starts to look significantly better at 10+ threads.

3D and 2D scores suffer slightly as a result of decreased single threaded performance.

Also the memory score looks worse. ECC is known to be slower (but more reliable), so likely you lost some performance there as well, NUMA doesn't help performance either.

[JAO]

Thanks David for you post and insight.

Sorry to have taken so long to re-post, I have been pretty tied up.

In reference to “process jobs” I meant to say that the application that I am using is doing high number calculations and algorithms which is forcing the application to take several hours to complete the analysis and finish the run. In this aspect there is a slight improvement when running the application but very slight. However, I have done a simple benchmark in opening the application on each of the two builds and I have noticed a faster response on the first build by :30 to :45 seconds faster.

As far as I know, the software is not running through the internet link however it uses the internet for the licensing model just to make sure it is registered.

I did not look at this line item in reference to the single threads, very important factor. I am considering changing the two existing Xeon’s for one of these two (Xeon E5-1650 v3 @ 3.5GHz (6 core) or Xeon E5-1630 v3 @ 3.7GHz (4 core)). I have included the comparison of each CPU side by side and the 4 core is actually very close to the i7 chip. Now, with two of these chips I would have a total of 8 cores, combined from two chips, which would be the same as the i7 being only one chip. Your thoughts on this?

Was not aware of the 3D and 2D performance based on single thread, so if I understand correctly by having a chip with more threads it hinders the capability of the GPU, correct? Another point of information which I am not to sure on since this is my first dual CPU build is that I have the current GPU placed on a x16 PCIe slot however it is connected to the bus of CPU2. Does this have any effect on performance? Would it be better to have it on the PCIe from CPU1 bus?

I had my doubts on the ECC performance and went into the bios and turned it off. I re-ran the performance test and I did get an improvement to the performance score. The scores I initially sent is with the ECC turned off.

Thanks again for your insight.

JAO

[David (PassMark)]

I meant to say that the application that I am using is doing high number calculations and algorithms which is forcing the application to take several hours to complete the analysis and finish the run.

If you are trying to build a computer primarily to run a single application then you need to profile this application (whatever it is) to see how it uses the resources available. Does it use the disk? How many cores does it use / can it use? Is is limited by RAM speed? Is it 3D? Is it limited by the 3D performance? How much RAM does it use? Does it use the internet? Otherwise you are just guessing.

so if I understand correctly by having a chip with more threads it hinders the capability of the GPU, correct?

No, not correct. BUT, chips with large numbers of cores tend to have lower clock speeds. This is done to keep the heat under control. So having a large number of cores usually means poor single thread performance. See also this chart,
http://www.cpubenchmark.net/singleThread.html
The fastest single threaded chips have 2 or 4 cores.