Sandra Arthmetic/Multimedia on LGA2011 - Hyperthreading effect?

K

Krobar

Limp Gawd
Joined
Feb 24, 2010
Messages
345
Can anyone supply some with and without hyper-threading numbers on this benchmark on some sort of LGA2011 platform? (Single or dual CPU is fine)

I'm configuring a dual E5-4617 (An oddity without HT support) system and am getting low results in arithmetic test of 241 GIPS and 140GFLOPS and just want to make sure it is lack of hyper threading causing these low scores and not a system issue.

Aggregate multimedia is low too at 438/218mpix/sec..
 
Yeah I will have to do it late tonight when I get home. If I forget to just PM me and remind me as I have been super busy lately.
 
Sandra Results are a bit random with this setup:
Memory, Latency and Bandwidth - Very Good
Arithmetic - Seems Low (Similar to E5-2620)
Multimedia Integer- Seem Low (Same as E5-2620)
Multimedia Float - Seems Terrible (Half of E5-2620)
.Net Integer Dhrystone - Looks Abnormally high (Better than E5-2687)
.Net Integer Whetstone - Looks Low (Same as E5-2620)
.Net Multimedia - Look Low (Similar to E5-2620)
Inter core efficiency - Very weird result, does not look like SNB-E at all
Transaction Throughput - High (On Par with 2687W)
Encryption Bandwidth - Very High (Better than E5-2687W)
Encryption Hashing - As Expected

Cinebench 11.5 Multicore - 15.2
 
Sandra Results are a bit random with this setup:
Click to expand...

I would bet the benchmarks are for the most part not using all of your cores. Look at task manager when you run the benchmark. How many cores are at 100%? If the benchmark is not using enough threads then your dual processor will get beat by a higher clocked lga2011 with less cores.
 
It uses only one CPU, and it looks like it doesn't use CPU AVX for floating point benchmark.
 
Raghar said:
It uses only one CPU, and it looks like it doesn't use CPU AVX for floating point benchmark.
Click to expand...

Thanks for the tip Raghar, SP1 hand not properly taken hold. Multimeda float tests are now looking healthier but still below expectations; 456/422Mpix/sec.
 
Here is a link for the 3930K @ 3.2Ghz. I did not make any special system changes other than 3.2ghz cpu and HT disabled. Turbo was also disabled.

Keep in mind I did not sanitize my operating system of all my background services and programs. Just a simple click and wait bench as is.

HT Disabled:119.28 GFLOPS / 70.08 GIPS

http://www.sisoftware.eu/rank2011d/...988e9d4e0d9e0d1e7d4f280bd8dabceab96a680f3cefe

Ht Enabled:156.69 GFLOPS / 117.06 GIPS

http://www.sisoftware.eu/rank2011d/...988e9d4e5d2e4d5e1d2f486bb8badc8ad90a086f5c8f8

4.5ghz OC with HT: 220.41GFLOPS 162.33GIPS

http://www.sisoftware.eu/rank2011d/...988e9d4e0d4e5d6e5d3f587ba8aacc9ac91a187f4c9f9
 
Last edited:
Thanks Tangoseal,

Your figures are the other way round to mine, FPU as expected but Integer results far too low. Looks like HT has a heavy effect on the integer score.
 
Krobar said:
Can anyone supply some with and without hyper-threading numbers on this benchmark on some sort of LGA2011 platform? (Single or dual CPU is fine)

I'm configuring a dual E5-4617 (An oddity without HT support) system and am getting low results in arithmetic test of 241 GIPS and 140GFLOPS and just want to make sure it is lack of hyper threading causing these low scores and not a system issue.

Aggregate multimedia is low too at 438/218mpix/sec..
Click to expand...

RAM is important too .. as in how fast can it feed the cores...

how much and what speed ram are you using?

I have 32GB of 1600

Also that chip does not have hyperthreading which does make a difference in arithmetic benchmarks
Your integer is going to be higher because while your dual chip system has 12 threads they are 100% real cores. HT units are about 25-50% as effective as a real true core to my understanding depending on how they are utilized.

The good news is that I am getting ready to place an order for a dual E5-2640 system for a client so that will be 24 threads of XEON power. I will do this same bench and post up here on the results.
 
This setup is on its initial install of 64GB RAM (Each CPU is running 8 4GB sticks at 1600) which should be enough for the CPU based benchmarks. RAM performance looks good (72Gbyte/sec).

E5-2640 is the same CPU with Hyperthreading enabled but a lower clock speed so I would be interested in your results. I only recently found out that Win 7 SP1 adds AVX support which has quite a strong effect on the Multimedia FPU benchmark.
 
Yes this new build is a workstation that will be using Win7 SP1 since it is going to be a mathematical computational engineering HPC for a engineering firm.
 
Krobar said:
This setup is on its initial install of 64GB RAM (Each CPU is running 8 4GB sticks at 1600) which should be enough for the CPU based benchmarks. RAM performance looks good (72Gbyte/sec).
Click to expand...

Can you set up 16 GB RAM drive, and post its benchmarks... Just curious.

Also set Linux into dual boot and test it from Linux just to find possible hiccups caused by W7 and dual CPU system. I assume you have at least pro version of W7.
 
Last edited:
Just wanted to let you all know that in this case the issue appears to be Sandra. Linpack performance was fine on the E5-4617s after ISOC was disabled.

For the record:
Linpack - 278Gflops
3D Mark - 7400

Sandra results remained a but weird. The R based programming routines the system is going to be used for achieved 6 times the performance of the old single Xeon X5472 system the user had.
 
Krobar I got the chips in... Dual 4620's. I will be building the system tomorrow night or Wednesday night and then I have to do a burn in for 12 hours to make sure she's stable then I will linpack and sandra it before I deliver to the new owner.
 
You must log in or register to reply here.
Back
Top