Ninja and Intel or AMD

[neb1998]

Has anyone directly compared intel versus AMD cpu's for backtesting? I generally run strategies with at least 20 inputs and sometimes higher.

I have 2 i7920's clocked around 4ghz and notice that with HT turned on that generally there is no performance increase and possibly even a decrease. This is not the case with simple systems, if i am just running like 5 inputs to test 4 exits or so the HT seems to help.

Seems the higher the generation size and population size (larger search space) that HT seems to not add any benefit.

With this being the case AMD chips become a very interesting alternative. As HT enhances many of the benchmarks that Intel versus AMD chips are compared on.

With HT turned off usually a 30% drop in performance on most benchmarks is exibited. With that in case AMD might actually be much better at $230 versus 600 for a 6 core intel chip.

Any feedback would be great.

[Trader.Jon]

oranges, tomatoes and apples

FWIW

I have two machines I use for optimizations.
#1 = ASUS mb stock Intel X9650 w 8gb ram + ram hdd + Win7 64 + NT7.1
#2 - TYAN mb, stock dual opteron 2382 w 64gb ram + sata hdd + Win7 64 + NT7.1

same strategy/parameters/data on an extensive optimization I worry about the intel freezing up ... on less extensive optimizations they are very equal, but I notice that the opterons 'jump' to the end of genetic optimizations more often than the intel (ie were at 30-45 min left and suddenly shift to 5 min)

YMMV, but I think a sweet spot would be 6 cores and 24-32gb ram to use the full capability of the current PHGenetic & NT7. For myself $$wise, if setting up another machine I would use the same opteron setup I have now but remove 32gb ram to the new machine.

I dont do benchmarks as there really isnt anything that is setup IMHO to test NT7 but the cpu charts http://www.cpubenchmark.net/high_end_cpus.html will show how the cpu compares to yours. It might be interesting to run an optimization of same strategy/data/parameters if it could be clocked by NT7 itself for results as a timespan.

Regards,
Jon