AIDA64 3.00 benchmark tests

[R0H1T]

FinalWire announced today that AIDA64 (Extreme and Business Edition) has been updated to version 3.00. Being a major version number, the fresh revision brings to the table significant changes.

The developer makes available fully rewritten cache and memory bandwidth benchmarks, tweaked for AVX2, AVX and SSE capable AMD, Intel and VIA CPUs. The tests use as many as 32 threads to measure the bandwidth for read, write and copy operations.

For measuring memory latency, the suite uses a block-random method, which should eliminate the effect of hardware prefetchers and minimize TLB misses.

Improvements extend to optimization of 64-bit assessment routines for AMD “Kabini” and “Temash” APUs.

The new release has initial support for L4 cache bandwidth and latency benchmarks for the upcoming Intel “Crystal Well” processor.

Also worth mentioning is that AIDA64 3.00 can provide GPU details for AMD Radeon HD 7990 “Malta” and nVIDIA GeForce 700 Series.

Download

 

[R0H1T]

Can anyone do a quick roundup of the new tests on their Haswell system & if possible compare it with any of their older builds ?

 

[WhoBeDaPlaya]

I would if my test 4770K setup wasn't still ploughing through memtest86+ (32GB is taking forever!)

This reviewer's toolkit is handy : http://www.cpuid.com/news/76-reviewer_kit_for_intel_haswell.html

 

[bronxzv]

Can anyone do a quick roundup of the new tests on their Haswell system & if possible compare it with any of their older builds ?

btw this very issue is discussed here: http://www.realworldtech.com/forum/?threadid=133925&curpostid=133941

 

[R0H1T]

btw this very issue is discussed here: http://www.realworldtech.com/forum/?threadid=133925&curpostid=133941

Yes & apparently ~

> A major update is rolled out today, http://www.techpowerup.com/184875/finalwire-aida64-v3-00-released.html
>
> quite a bad timing now that all reviews are published
>

from the same forum ~
AIDA64 updated today

Perhaps some of the reviewers can update their charts accordingly :\

 

[Magic Carpet]

 

[bronxzv]

Yes & apparently ~from the same forum ~
AIDA64 updated today

Perhaps some of the reviewers can update their charts accordingly :\

I posted this there after seing your post here, thank you for the hint

 

[bronxzv]

thanks a lot, now that's fresh news!

 

[R0H1T]

Do you have any other system to do the tests on, other than IVB & using the latest AIDA64, it'll perhaps give us a perspective as to what else has changed.

 

[grimpr]

Quick run on the FX8350 box.

 

[Magic Carpet]

Do you have any other system to do the tests on, other than IVB & using the latest AIDA64, it'll perhaps give us a perspective as to what else has changed.

I'll bench a few more later today.

 

[R0H1T]

Possibly one of the slowest system going around D:

 

[SPBHM]

Quick run on the FX8350 box.

was latency so high (particularly the l3) on older Aida versions!?

 

[Magic Carpet]

was latency so high (particularly the l3) on older Aida versions!?

Looks like, 3.00 uses different methods, in all of my tests, I have got much worse latencies versus 2.85.

Coming up: Merom, Pentium 4 and Thuban

 

[SPBHM]

Possibly one of the slowest system going around D:

nope.

Looks like, 3.00 uses different methods, in all of my tests, I have got much worse latencies versus 2.85.

Coming up: Merom, Pentium 4 and Thuban

they definately changed a lot, and since you didn't mentioned Brisbane:

I wonder what's going on with the l2 write!?

 

[grimpr]

It says it clearly, thats where the latencies come from. Whats interesting is that they use all threads avalaible to measure real cache bandwidth, this is where the Core i7s & the FX will shine, the increase is big.

"For measuring memory latency, the suite uses a block-random method, which should eliminate the effect of hardware prefetchers and minimize TLB misses."

 

[R0H1T]

they definately changed a lot, and since you didn't mentioned Brisbane:

WTH the L2 Cache latency is ~7x worse now, quite similar to grimpr's 8350 & its L3 Cache latency, definitely poor(er) results for AMD

 

[grimpr]

WTH the L2 Cache latency is ~7x worse now, quite similar to grimpr's 8350 & its L3 Cache latency, definitely poor(er) results for AMD

Read again the description of the new test. Hardware prefetchers hide latency and Intel has the best ones.

"For measuring memory latency, the suite uses a block-random method, which should eliminate the effect of hardware prefetchers and minimize TLB misses."

 

[R0H1T]

It says it clearly, thats where the latencies come from. Whats interesting is that they use all threads avalaible to measure real cache bandwidth, this is where the Core i7s & the FX will shine, the increase is big.

Interesting, also worth noting is that your FX's L2 Cache read/write/copy numbers are off the charts even though the latency has nearly doubled

Read again the description of the new test. Hardware prefetchers hide latency and Intel has the best ones.

Yes I understand that but apparently such a drastic change could skew alot of calculations & other charts !

 

[grimpr]

Interesting, also worth noting is that your FX's L2 Cache read/write/copy numbers are off the charts even though the latency has nearly doubled Yes I understand that but apparently such a drastic change could skew alot of calculations & other charts !

All i know is that it uses all available threads at 100%, so the numbers are the total throughput of the chip with its load/store units and not per module or core as were the old version, still many unknowns though we need more details and benchmarks to come to a conclusion, big upgrade that changes everything.

 

[R0H1T]

All i know is that it uses all available threads at 100%, so the numbers are the total throughput of the chip with its load/store units and not per module or core as were the old version, still many unknowns though we need more details and benchmarks to come to a conclusion, big upgrade that changes everything.

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:

 

[guskline]

Thanks for the tip. I have a licensed version of Aida64. Well worth the $$.

 

[csbin]

X79 QUAD Channel

 

[R0H1T]

Better numbers in safe mode

 

[Magic Carpet]

As promised:

Please note, this is Single Channel.




Now the oldies:




Late Pentium 4 D Stepping. Not much difference there.