AIDA64 3.00 benchmark tests

[Edrick]

 

[Edrick]

 

[TuxDave]

Yup seeing the IVB numbers above I can see why this is a welcome change however, as you said, there needs to be some more clarity especially wrt the accuracy of these numbers & the error margin :hmm:

If anything, this thing needs one more significant digit on the latency number so you can correctly translate memory latency into # of cycles. Rounding kind of screws it up right now. The L1 read bandwidth isn't quite at the theoretical number but it's pretty close.

 

[bronxzv]

The L1 read bandwidth isn't quite at the theoretical number but it's pretty close.

indeed, at ~31.21 B/clock (AIDA64 v3.00.2500, per core) it's at ~97.5% peak, L1D write bandwidth with 15.61 B/clock is also at ~97.5% peak

it will be neat if they also report scores that truelly maximize the L1D throughput, i.e. at 2:1 load:store ratio, the aggregate L1D bandwidth (all cores) will then be reported at around 770 GB/s

now, can someone tell me where I can see *Haswell scores* with this deeply revised version v3.0.2500 ?

 

[R0H1T]

If anything, this thing needs one more significant digit on the latency number so you can correctly translate memory latency into # of cycles. Rounding kind of screws it up right now. The L1 read bandwidth isn't quite at the theoretical number but it's pretty close.

Yes but I figured it'd be within the margin of error especially since theoretical numbers aren't achievable in real world scenarios so even benchmark tools can be off by as much as 5~10% depending on how good they're coded.

 

[tw33k]

 

[Redentor]

An interesting interview with Jozsef Barna (AIDA64 Italy) about these benchmark: http://www.bitsandchips.it/software...64-i-benchmark-su-cache-e-imc-hanno-un-perche

Interview is in English.

 

[Magic Carpet]