i7 6800k is slower than expectation!!

atomdie · October 22, 2016, 18:48

Hi guys!
I just build a new system with
- Asrock X99 Taichi
- i7 6800k
- 64GB Ram 2400Mhz Cas 15
- SSD 850 EVO 250GB+ 2TBB HDD

I tried to run Ansys Mechanical, and it was slower than I expected. The speed was only 65000 Mflops, meanwhile my E3 1230v5 with 4 cores + 32GB RAM 2133Mhz event could reach around 100000 Mflops. How is it possible? What did I do wrong?

Thanks a lot.

flotus1 · October 22, 2016, 18:55

I have a few questions:
How did you measure "Mflops" with Ansys machanical?
Did you compare the systems with a single-core, a 4-core or a maximum-amount-of-cores benchmark?
Is hyperthreading turned off?
Have you made sure that all components are running at their intended speed even under load?

atomdie · October 22, 2016, 19:02

Quote:

Originally Posted by flotus1

I have a few questions:
How did you measure "Mflops" with Ansys machanical?
Did you compare the systems with a single-core, a 4-core or a maximum-amount-of-cores benchmark?
Is hyperthreading turned off?
Have you made sure that all components are running at their intended speed even under load?

Hi flotus1,

I took a screenshot, so it is easier to understand.

As you can see from the picture, the Mflops is measured. I turned off hyper thread. Six cores were used with 100% load. What did I do wrong? I still cannot understand. I expect it should be at least 1.5x faster than my older system.

And here from E3 1203 v5 without turning off hyper thread.

flotus1 · October 23, 2016, 03:26

Which does not answer my last question.
You should at least monitor CPU frequency and temperature under load. In addition to that, you should run independent benchmarks that will allow you to compare your results. Use benchmarks that focus on CPU power and memory bandwidth independently, for example Cinebench and Aida64.

atomdie · October 23, 2016, 15:10

Quote:

Originally Posted by flotus1

Which does not answer my last question.
You should at least monitor CPU frequency and temperature under load. In addition to that, you should run independent benchmarks that will allow you to compare your results. Use benchmarks that focus on CPU power and memory bandwidth independently, for example Cinebench and Aida64.

Thanks flotus, after a long day I figured out what is wrong. The problem is window 7 version. I wiped out everything and installed new window 7. Now everything is fine, i7 6800k@4.0Ghz has speed around 150k Mflops. I am happy now

atomdie · November 7, 2016, 17:01

Quote:

Originally Posted by thickglass

I got almost the same specs as you do, its not that bad, I should be able to reach 190k easily

How could you do that?

mongo · November 21, 2018, 02:44

Quote:

Originally Posted by atomdie

Thanks flotus, after a long day I figured out what is wrong. The problem is window 7 version. I wiped out everything and installed new window 7. Now everything is fine, i7 6800k@4.0Ghz has speed around 150k Mflops. I am happy now

Hi, thank you for the data firstly. My cpu is I5 2500 running in win7 with 16Gb RAM. I use all cores to run a contact analysis, the ram is enough for the simulation. but the speed is only 30000 MFLOPS, compared with your speed 100000+MFLOPS, I think my CPU is running at non reasonable speed. then I test my cpu the cinebench , the result is normal. do you have any information about the CPU speed in ANSYS? Thanks

flotus1 · November 23, 2018, 13:46

Worst case would be that your memory is DDR3-1333 dual-channel.
DDR4-2400 quad-channel delivers almost 4 times the theoretical memory bandwidth. Combined with 2 additional cores, higher clock speed and a newer architecture on the I7-6800k the results don't seem too far off.
Cinebench is mostly a compute-bound benchmark, so not a good indicator for memory-bound applications.

October 22, 2016, 18:48	i7 6800k is slower than expectation!!	#1
atomdie New Member Deutschland / Germany Join Date: Aug 2016 Posts: 8 Rep Power: 9	Hi guys! I just build a new system with - Asrock X99 Taichi - i7 6800k - 64GB Ram 2400Mhz Cas 15 - SSD 850 EVO 250GB+ 2TBB HDD I tried to run Ansys Mechanical, and it was slower than I expected. The speed was only 65000 Mflops, meanwhile my E3 1230v5 with 4 cores + 32GB RAM 2133Mhz event could reach around 100000 Mflops. How is it possible? What did I do wrong? Thanks a lot. thickglass likes this.

October 23, 2016, 03:26		#4
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,400 Rep Power: 47	Which does not answer my last question. You should at least monitor CPU frequency and temperature under load. In addition to that, you should run independent benchmarks that will allow you to compare your results. Use benchmarks that focus on CPU power and memory bandwidth independently, for example Cinebench and Aida64. atomdie likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Problem with conexion under windows Expectation failed: PostGui_ogl.exe:	seba_dd	ANSYS	1	April 15, 2016 11:40
MPI slower than serial	LucaFen	OpenFOAM Running, Solving & CFD	0	November 24, 2015 06:08
Dual Nodes is Slower Than Single Node	Mrxlazuardin	FLUENT	0	May 21, 2010 01:48
Parallel runs slower with MTU=9000 than MTU=1500	Javier Larrondo	FLUENT	0	October 28, 2007 22:30
Fluent 6.2 slower than 6.122?	Riaan	FLUENT	4	April 15, 2005 17:44

October 22, 2016, 18:55		#2
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,400 Rep Power: 47	I have a few questions: How did you measure "Mflops" with Ansys machanical? Did you compare the systems with a single-core, a 4-core or a maximum-amount-of-cores benchmark? Is hyperthreading turned off? Have you made sure that all components are running at their intended speed even under load?

November 23, 2018, 13:46		#8
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,400 Rep Power: 47	Worst case would be that your memory is DDR3-1333 dual-channel. DDR4-2400 quad-channel delivers almost 4 times the theoretical memory bandwidth. Combined with 2 additional cores, higher clock speed and a newer architecture on the I7-6800k the results don't seem too far off. Cinebench is mostly a compute-bound benchmark, so not a good indicator for memory-bound applications.