Low-Performance of AMD Microprocessors May Be Conditioned by? Poor BIOS.

[s1nykuL]

http://www.xbitlabs.com/news/cpu/dis...Poor_BIOS.html

[The Stilt]

That article is nothing but pure speculation.
Until the real technical background of the x87 "tweak" is found, all of the speculation is a complete waste of time.

AGESA, together with the microcode basically dictates how the CPU is configured.
Unless the ODMs modify the AGESA itself they really can't make much of a difference.

Besides, this "tweak" (x87) does not have anything to do with either the bios or the microcode.

[AzureSky]

thought I would love better bios/uefi, Its not bios thats the major cause of the perf differences, the fx line are weaker in ipc, the arent bad, i love both my 8120 and 8350, great chips for the price IMHO, but they arent going to destroy intel chips clock for clock.

depending on the workload they can be the better value though....at least in my book....

what I would really love to see is board makers offer us a choice of UEFI or Award bios like gigabyte offer(features of uefi but UI of award), i prefer awards layout honestly.....and i have had a sabertooth for a long ass time now....

I would also like to see an unbias compiler come out thats at least as optimized as intels, so us AMD users could get AVX and FMA support(intels compiler blocks non-intel from getting avx, and the intel compiler patch cant fix that)

from what I have seen testing some small beta apps that have AVX and FMA support, in some cases, it can more then double the perf of my 8350......thats IN_FING_SANE....and i can see why intel wouldnt want that to happen when people use their compiler, after all, imagine if cinebench got working FMA and AVX for AMD chips.....sad since at times, amd can already stock for stock beat intel chips......that cost more....

[s1nykuL]

I would also like to see an unbias compiler come out thats at least as optimized as intels, so us AMD users could get AVX and FMA support(intels compiler blocks non-intel from getting avx, and the intel compiler patch cant fix that)

Are there no compilers that take full advantage of AMD feature set? It's been a long while since I did any serious coding.
I just work with PHP and HTML these days.

My last AMD chip was an XP3200. Prior to that a 1 gig T-bird, that I ran at 1.43 Ghz.
I have no favorites.... I would jump to AMD in a heartbeat if their chips outshone the Intel offerings.

[CrazyNutz]

Are there no compilers that take full advantage of AMD feature set? It's been a long while since I did any serious coding.
I just work with PHP and HTML these days.

Compilers do have options for many instructions sets & targets, however you could compile a program that only works on certain CPU's.

It's not so easy. Good optimization must be hand coded using ASM, especially when it comes to SIMD instructions.

The best way is to detect what instruction sets the CPU supports at run time and then the program can use optimized versions of functions written for that specific instruction set. But this is a lot of work & you will only see developers put this kind of work in if it is a necessity such as video codecs, or scientific computing etc. Also if the project does not have the budget in money and/or time it will not happen.

This is why it is still important for legacy instructions to perform well.

[AzureSky]

no, as far as I have seen, no compiler currently is optimized enough to compete with intels, and no compiler I have tested gives decent fma/avx for amd, intels compiler is the best on the market, but the problem is they cripple non-intel chips intentionally desite the deal they made with amd when they paid them all that money to stop doing that....

yes hand coded ASM optimizations are optimal, but a good compiler can make a huge difference,
http://www.agner.org/optimize/blog/read.php?i=49

skim that, you will see how huge a boost VIA for example gets from changing the cpuid to genuine intel.....

it has driven alot of developers to just use MS's compiler as its faster(compile time) and less bias, it dosnt lead to one vendor being faster then the other by sending cpu's differing code paths based on nothing more then vendor.

compiler level optimization isnt optimal but its better then nothing...and cheaper then hand coded asm.