Itanium Rc5 benchmarks!

[Fingers]

Sorry if i got your hopes up guys but for now with the current beta build of whistler for a 667 Mhz itanium we are only looking at 96kkeys/sec. see here for proof. sorry its in a foriegn language but i think most of you can make it out.

 

[Saus]

Gelukkig zijn er meer benchmarks dan alleen degenen die je overal ziet. Zo is er bijvoorbeeld de Distributed.net client, beter bekend als 'de koe'. Deze bleek gelukkig wel te werken op het systeem, hoewel de processor uiteraard niet werd herkend. Na de automatische selectie van de snelste core startten we vol goede moed een benchmark... Een paar seconden lagen onze kaken op de grond, want de Itanium haalde nog geen 96 kkeys/s, een score die zelfs door 486 verbeterd kan worden! De Rc5 cores van de client zijn natuurlijk zwaar geoptimaliseerd voor echte x86 processors, waardoor een goede score eigenlijk al uitgesloten was, maar zó slecht had waarschijnlijk niemand verwacht:



i'm going to translate this in pieces because my english isn't so good

so the story is every benchmarksoftware wouldn't work on a Itanium, so they tried the Distributed client, better know as the cow. the processor was ofouze unknown for the client so after the automatic selection procedure they started the fastes core, after a few seconds they saw that the Itanium not even do 96 kkeys/sec, a score what could be beaten by an 486, the scores for the RC5 clients are optimalised for an x86 processor, so a good score was not in the line of predictions but so bad, no one expected


tada

 

[Fingers]

thanks for the info.

 

[ElFenix]

isn't the client optimized for very sepecific processors? if this is the case no benchmark using RC5 is valid until code specific to that processor is written.

 

[ss59]

ElFenix is right. Until the code has been optimised for the itanium(which will probably take a year or more) we can't judge the chip one way or the other.

 

[ss59]

Not to mention the Windows kernel at work here probably isn't up to snuff yet.

 

[Fingers]

i figured but it's still kindaq funny to see such a high end CPU running so slow.

 

[QW]

Tweakers.net has already posted an English version of the Itanium Preview Here.
But the biggest problem of the low keyrate is that de Distributed Client isn't build for the Itanium let alone a 64Bit processor. Off course i hope that changes some dat

 

[thorin]

Well it selected some RISC core this is the first clue that it isn't going to be a valid indicator of performance. Second 96kkeys/sec guys think about it. It's a 64bit processor it should be able to do everything it needs without any ROTL or ROTR (rotate left/right) commands since it can handle the whole 64bit key at once.

When we get a real Itanium core it'll fly for RC5-64, bet on it.

Thorin

 

[LeBlatt]

I disagree when one says that because no rc5 core is optimized for a cpu, the rc5 benchmark for that cpu is meaningless.

It does mean that the itanium is unable to run some software that people currently use. So before anyone goes out and buys one, he has to test every programm he uses to make sure the new expensive chip isn't going to perform like a 486.

Hell, if I work with photoshop and make heavy use of a particular filter that is heavy on a particular set of instructions, and I find out the itanium sucks at it, will I wait 1 year for that filter to be optimized for my cpu ? I sure will get another platform to work on. That goes for all other software.

When the athlon was out, no core was available for it but it rocked rc5 already. I dont mean HW vendors must not innovate and stick to old standards, just that abrupt innovation can not win the market at once ; software has to follow. If you run too fast, you end up hitting a wall.