Intel adopt PR

[Wombat]

There already is, SPEC. But that would make Athlons and P4s look puny compared to real processors.

[CHHAS]

The Cyrix chips were pretty decent, as long as you stuck to doing integer stuff, the floating point part was very weak.
I had a Cyrix 166+ at one point and in basic use it was as fast if not faster than my K6-166.

[dZeus]

K6(-2/-3) also had very crappy FPU compared to the Intel CPUs

[Wombat]

Yes, but they were tons better than Cyrix's. Then the Athlon had better FPU than the P3. Then there's McKinley

[Ribbit]

The problem with the Cyrix PR rating was that Cyrix stretched it for all they could get - comparing integer code rather than a mix, rating against Intel chips with 66MHz buses when 100MHz was the norm, that kind of thing. AMD have for the most part done it properly and fairly this time round, although I wish they wouldn't spend so much effort covering up the actual clock speed.

[Ribbit]

Originally posted by dZeus
K6(-2/-3) also had very crappy FPU compared to the Intel CPUs

It wasn't actually that bad, it just had very different performance characteristics. The Pentium (<4) FPUs were heavily pipelined, so they took a long time to perform an operation, but they could line them up and pump them out practically every clock cycle. The K6-x FPU was low-latency - it could get through each operation quickly, but couldn't line them up. It was possible to get good FP performance from a K6-x, but Intel chips had more throughput and more people had them, so that's where the rewards from optimising were.

That explanation was crap. Sorry...

[Wombat]

That wasn't a bad explanation, but the K6x FPU, even with optimized code, just couldn't put out as much as the Intel units.

[lurqa_MU]

'real processors'

Originally posted by Wombat
There already is, SPEC. But that would make Athlons and P4s look puny compared to real processors.

If you check spec.org you will see that today there is no single CPU that can match the performance of the fastest Athlon/P4.

UltraSparc, Alpha, Power*, MIPS they are all toast...

Naturally, once you look at SMP/Numa architectures the pictures changes.

Regards,
lurqa

[SpiralDragon]

arg... thy should be working on the botle neks first in my opinion.... what with all the old tech in a x86 machine... PCI, IDE, etc... etc.... do they have to milk every penny from this stuf up untill the point where they just have to introduce a technology that should have been introduced along time ago but now is nothing compared to what they haveand will not realeas for a long while :-( ..... i cant keep up with upgrading my pc every year.. its getting to frutrating....

[lurqa_MU]

Cyrix!

Originally posted by CHHAS
The Cyrix chips were pretty decent, as long as you stuck to doing integer stuff, the floating point part was very weak.
I had a Cyrix 166+ at one point and in basic use it was as fast if not faster than my K6-166.

Yeah. I am with CHHAS, Cyrix were the kings in integer performance.

Too bad that only the serious stuff would run well with the 'integer only' performance.

Regards,
lurqa

[Wombat]

If you check spec.org you will see that today there is no single CPU that can match the performance of the fastest Athlon/P4.

Huh? Check out the SPECfp marks. Here, try this eye chart: http://www.hp.com/products1/itanium/...e/speccpu.html

P4 2.53/533: 878
McKinley: 1GHz/200: 1356

[lurqa_MU]

spec.org

Originally posted by Wombat
Huh? Check out the SPECfp marks. Here, try this eye chart: http://www.hp.com/products1/itanium/...e/speccpu.html

P4 2.53/533: 878
McKinley: 1GHz/200: 1356

My comments were mostly ment on the *nix and mainframe CPU's listed there - there are starting to look really bad.

I hadn't noticed that McKinley results had been posted.

http://www.spec.org/osg/cpu2000/results/cpu2000.html

(I check spec.org mostly to see which CPU gets the highest score in the gcc test (integer)... )

McKinley seems to have an interesting performance profile. Strong fpu and perculiar low integer performance... It seems McKinely is more a workstation than a server CPU?. (That is unless they get MHz cranked up.....)

Regards,
lurqa

[Wombat]

It's good at both, though we did sacrifice int slightly in order to make the FP stuff the fastest on the planet. As a server, it does pretty well: our 4-way boxes did a nice job of beating up on Dell's 8-way Xeon setups, and our TPC-C numbers are nice too.
http://www.hp.com/products1/itanium/...nce/index.html The rest of the server info will have to wait though.

[lurqa_MU]

McKinley

Originally posted by Wombat
It's good at both, though we did sacrifice int slightly in order to make the FP stuff the fastest on the planet. As a server, it does pretty well: our 4-way boxes did a nice job of beating up on Dell's 8-way Xeon setups, and our TPC-C numbers are nice too.
http://www.hp.com/products1/itanium/...nce/index.html The rest of the server info will have to wait though.

Interesting. How symmetric are those machines - I mean are they traditional intel arch. using a shared bus or are they using a Numa like implementation?

I guess the real question is: will we see boxes with more than 4 McKinley's in them?

Regards,
lurqa

[Wombat]

Give me a while. I know all the answers to those questions, but these NDAs are a pain in the ass. It will probably have to wait a while until I see what's public these days.

One thing though, those boxes are HP boxes running HP chipsets, I think they're calling it the "zx1" these days.