Hello, I wish to go home earlier  :'( . I spend lots of time waiting for simulations to get done  :( . Each cycle is lots of time  >:( . Keeping me in the office for longer and getting the design much slower  :-/ .

I wish to ask ALL of you to give an answer to this questions:

-Using multi-threaded parallel simulator APS, which is the best hardware to run on in terms of performance?

-->In specific: Is the current flagship server processor from AMD Opteron 6282SE better than Intel's Xeon E5-6290? (assumptions: Utilizing all the available cores/threads on the given processor and also with similar HDD and Memory).

-->Or, please answer the following question: What kind of floating point calculations the latest of Spectre/APS use? Does it use 128 bits or 256 bit floating point calculations? If 128bits, then AMD should fair well in the simulation performance, if 256, then AMD would suck big time in comparison with Intel.

So, please help if you know.

Thank you alot alot alot for the help.

basem

Thanks for the info, but does Spectre/APS really not fully use the hardware on modern processors? If it is true that Spectre will not use more than 128bits of floating point calculations then AMD will be MUCH better than Intel, but I have my fundamental doubts... (multiplying two 64 bit numbers will result in 128 bit number here AMD is good, multiplying three or more 64 bit numbers will result in 192 or more bits number here AMD is not good at all (50% loss of performance))

To all of the members reading this:

Did you or someone you know conduct any recent comparison between AMD and Intel computing platforms running Spectre or APS? With circuits having dynamic range of over 85dB (more like 104dB). The analog precision setting on the simulator is one order of magnitude higher than the defaults and the runs are on conservative setting.

Please forward all questions to all you know that might know about this.

thanks,

basem

You are very correct I am wrong it would over flow only, I do not know how x86 works and I do not know the IEEE 754 floating point standard and I do not know how x86 implements IEEE 754. I also do not know how simulators use the hardware.

Nvidia CUDA cores? Can I run APS on CUDA cores without doing anything special in software tweaks? I never knew that APS would use the GPU on its own....please correct me if I am wrong.

Without knowing IEEE standard and how x86 implements it, I wish to know how these different hardware would affect the simulations of analog circuits (Spectre/APS, etc), also, EM simulations. The number of "cores" on AMD is double that on intel, but, there is significant difference between the two that makes AMD not really "double" that of Intel. On the other hand, I do not know that Cadence has its simulators utilize the GPU for calculations. Let alone the vast majority of the broken software industry.

Can I use GPU with APS? If yes, how? Just install it? Which is better intel or amd? which path should I try? any one out there that knows the best way to go home earlier and get things done sooner?

thanks for the help

Thank you for the clarification and knowledge sharing. Circuits vary from having many many many nets resulting (as I understand from your post) in very large matrix to be solved with fairly small simulation coverage to having few components but which needs to be simulated extensively with very large verification simulation coverage.

What I am trying to say, sometimes the matrix is HUGE (I guess) and simulation is limited by the simulation time and available hardware. In other times (which is more often during design work), the simulation is repeated many many times with different configurations and settings on a moderate sized circuits. Design time can be reduced perhaps if GPU is used and cluster computation here is of no/little benefit (tried). Large full chip sims perhaps do not benefit from GPU as much and benefit from cluster computation (also tried).

To answer your questions:
-Intel: 256KB for each of the eight cores (inside each of them and very fast seem to be), and 20MB of L2 cache (they call it smart, so I guess its shared among the eight cores in a smart way?).
-AMD: The chip has 16 cores in eight modules. Each core has 48KB of L1 cache; then 1000KB of L2 cache running at 2.6GHz. Then shared among all the cores, there is L3 cache of 16384KB.

All the reviews I read online on these processors reveal that Intel is nearly 50% faster than AMD, however, none of these applications are circuits simulations which may demand very different hardware setup, not sure at all. That is why I asking here...

thanks a bunch

basem

No. But there is plenty of benefit in using multi-threading already in APS and it's unlikely more would be gained by using GPUs. Especially as often people are running simulations on servers without a graphics card.

Basem,
   What are the simulator options you normally run with?

-Ken