No more GHz

[ChrisLudwig]

I’m sure you guys have heard that Intel is dropping GHz as a rating of CPUs.

My question is: why don’t they use GFLOPS as the definitive measurement of performance? I understand this rating doesn’t account for everything, but it is the standard by which supercomputers are rated.

Like this: http://www.top500.org/list/2003/06/?page

 

[imgod2u]

Simply put, there is no one standard GFLOPS rating. For instance, take a P4 3.0 GHz. For 32-bit SP floating point numbers. It's x87 FPU is capable of a sustained/peak 6 gflops. Using SSE, that number is a peak/sustained 12 gflops. For 64-bit DP floating point, it's 3 gflops for x87 and 6 gflops for SSE2. However, using SSE and SSE2 must be in cases where SIMD applies. So the answer is, it varies too much. There is no one gflops rating. You could easily double or quadruple your gflops rating simply by changing the datatype.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

 

[kinney]

Sounds alot like stereo power ratings.
RMS/peak ect.

Once things get out of being 'set in stone' like mhz=performance (generally this has been true in the past).. I doubt we'll see too many companies agreeing on a standard rating because they can fudge with these numbers, unlike before in the 'MHZ days'.
They like that.

____________________________
<b>RESIDENT FORUM WARRIOR
<font color=purple>I just neutered the cat.
Now he's a liberal.</font color=purple></b>

 

[imgod2u]

I like the direction we're heading. Model numbers that aren't suppose to represent anything but relative performance among the same model. Now consumers will have to actually do research.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

 

[kinney]

Now consumers will have to actually do research.

Heh..
I wonder if that will actually happen... :frown:

____________________________
<b>RESIDENT FORUM WARRIOR
<font color=purple>I just neutered the cat.
Now he's a liberal.</font color=purple></b>

 

[imgod2u]

Probably not. Companies will now strive to come up with higher numbers with no correlation to anything.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

 

[kinney]

Radeon/Geforce 980098 anyone? :tongue:


edit to add- oops i misread you.
I disagree.. I think if they can sell more with merely a higher number.. they will.
You really do disagree with that?

I feel its already kind of happened with some of AMDs chips to an extent, and video cards like my original post.

____________________________
<b>RESIDENT FORUM WARRIOR
<font color=purple>I just neutered the cat.
Now he's a liberal.</font color=purple></b><P ID="edit"><FONT SIZE=-1><EM>Edited by kinney on 04/02/04 01:13 AM.</EM></FONT></P>

 

[P4Man]

SPEC would actually be somewhat more usefull, but even then.. performance depends on many things besides the cpu, for instance chipset, memory subsystem, and compilers ! It doesnt sounds practical to me to change your product number from 1175 to 1215 because you just released a new compiler

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[ytoledano]

doesnt sounds practical to me to change your product number from 1175 to 1215 because you just released a new compiler

I assume you mean that when Intel releases a compiler that uses SSE3 they'll change Presscott's name/rating? I guess then, that Presscott should've been given a different rating from the start since it's different than other P4s. Also, shouldn't chips be named not only according to their core, but also according to the chipset the use? (see Willamette 423/478).

Abit IS7 @ 275 FSB, VAGP @ 1.65, OCZ PC4200 RAM @ 550Mhz, P4 2.4 @ 3.3Ghz Vcore @ 1.625, GeXCube Radeon 9600P @ 450/626.
Thermalright SLK-947U, P3 HS @ NB.

 

[P4Man]

>I assume you mean that when Intel releases a compiler that
>uses SSE3 they'll change Presscott's name/rating?

No, compilers evolve/improve constantly, even without using newer architectural features. You might remember much of the "Pentium" optimized code also benefitted older 486's. This same has always been true, while maybe not as spectacular as the increase we see with newer CPU's, compiler improvements are responsible for a major part of the overall system performance improvements of the last decade(s).

SSE3 won't do much btw, not even for SPEC; if it does I'd go out on a limb claiming its probably because of compiler improvements unrelated to SSE3 but that are only enabled when SSE3 capabilities are detected. Intel tends to do that,locking out older processors (and thereby typically also competing processors) of generic improvements their compiler would also achieve on older ISA's by detecting the feature set, and using a slower generic code path for cpu's without the feature, and using newer faster codepath for cpu's with them. Ct recenlty showed how by "hacking" the intel compiler, they managed to produce code that ran significantly faster on A64 using a codepath that was uniquely reserved for "genuine intel cpu's", but that ran perfectly well on the A64 regardless.

= The views stated herein are my personal views, and not necessarily the views of my wife. =