MFLOPs instead of PR rating? Poll

[tart666]

What do you guys think of using MFLOPs instead of the AMD's PR rating?

(or any other STANDARD bench for that matter)

Rpeak and Rmax seem to have been stable benches for a few years now, at top500 at least. So I say, who thinks that AMD should use that?

Consider this: you could compare different architechtures/single vs dual/ diff memory size /even CPU vs GPU (for same task) all in one number. Just like the different supercomputers are compared. It will not encompass everything, but anything is better than bare MHz that everyone is used to. And the enthusiast can allways look at the full Linpack plots/ more granular benches.

This should be easy to explain to the public, which already seems to understand the rpm/bhp/torque scheme pretty well

put in your 2 ¢

T

 

[drogue]

to be honest, i dont know what a mflop is, so i cant really vote. anyone care to enlighten me?

 

[Budman]

Originally posted by: drogue
to be honest, i dont know what a mflop is, so i cant really vote. anyone care to enlighten me?

A megaflop is like the movie Waterworld,that was a megaflop.

 

[zzzz]

FLOPs: Floating point operations per second

 

[bacillus]

A megaflop is like the movie Waterworld,that was a megaflop

hehe

 

[WarCon]

If your going to go to the extreme of requiring that all companies display their "ratings" equally then some third party needs to write a benchmarking suite that takes into account the specific benchmarks that most consumers use and rates it with the most weight and so on and so forth until a mutually agreed upon benchmark is developed.

The downside of this is that the manufacturers will beef up only the section that gives the highest rating since consumers will buy based upon those decisions. I personally would rather let them duke it out and keep innovating to beat each other.

If one company feels PR ratings is the way to go then more power to them. I will still buy based on performance and stability and overclockability (because I am an overclocker now........damn disease I tell ya).

 

[andreasl]

Unless you're going to use these desktops for HPC (High Performance Computing) type work I don't see the point in using MFLOPS to measure them. MFLOPS was the measurement with which super computers were measured back in the days and it has sort of stuck by ever since.

Desktop computers should be measured owith desktop type benchmarks and workloads. The ideal thing would be for the industry(OEMs) to agree on a set of benchmarks and then rate each system configuration accordingly. Fat chance of that happening however.

 

[Duvie]

First of....

Are you going to use mflops like in sandra that calculate p4's with sse2??? cause if you don't your benchmark theory really sucks...

A 2.66ghz which would clearly kick the 2200+ butt since the 2.533ghz already beats it, would onluy have 1385 non sse2 mflops to the athlons 2505...hell it would still get beat by a 1.2ghz tbird...If you use sse2 the p4 gets the unfair advantage...Overall a sucking way to pr rate...

Some company or consumer reports type company should write a benchmark program to test the cpu in a host of test...


I dont like any of the choices...no vote here...

 

[DAPUNISHER]

Yeah, all the choices are suck, and that AMD/Apple/Alpha should die all together choice is just wrong on so many levels and leads one to surmise that you aren't an economist

 

[tart666]

and that AMD/Apple/Alpha should die all together choice is just wrong on so many levels and leads one to surmise that you aren't an economist

huh? don't you mean my incompetent opinion suggests that I AM an economist?

BTW, new poll option for y'all Alpha lovers

T

 

[DAPUNISHER]

Sorry that new option is suck too BTW I wasn't implying you are incompetent, I was inferring that were all those companies to go "belly up" that the lack of competition would be bad for the consumer (me and you) so you probably weren't a financial expert, that is all.

 

[mechBgon]

Watch this space. :Q Not sure what to make of that headline... I guess we'll see when the story goes up.

aaaand it's up. Sounds like a challenge for Intel's marketing folks to come up with a good explaination for how it can be faster even with less MHz:

That is why in the end of the year Pentium 4-M family with the working frequencies lying between 2GHz and 2.6GHz will be replaced by Banias based processors working only at 1.3-1.6GHz. How will Intel marketing people manage to convince the users that the new Banias will be faster than the old Pentium 4-M, is a secret. At the same time Banias 1.6GHz is expected to be faster than Pentium 4-M 2.6GHz and to dissipate considerably less heat. For instance, TDP (Thermal Design Power) for Banias CPUs will make 20-25W, while the fastest Pentium 4-M models will have TDP=35W.

 

[Damascus]

I have a good idea. SPECint and SPECfp should be used.

 

[Lonyo]

How could Intel and AMD do PR ratings? There would be no standard (AMD compares to Intel speeds does it not?) so if Intel and AMD both went to PR it would be incredibly confusing. Could they do a MFLOP test without using extra instructions like SSE/SSE2 etc? That would negate the point of them as a marketing aid but might give a better indication of pure speed.

 

[CTho9305]

mflops is worse than PR and worse than mhz. I dont think any synthetic rating works. When I buy a CPU, I read benchmarks of software I use (at the time unreal tournament and 3d studio max mainly). If a 500mhz athlon were better than a 1ghz p3 at THOSE TASKS so be it. Now, since most people dont have the time to find that kind of info, the PR rating is pretty good. Read AMD's paper on how they determine it - using a suite of real apps.

 

[Soulkeeper]

they should just have model names only
like cars and never tell you the speed
hmmm maybe not but still maybe that would force people to look at benchmarks

 

[SSXeon5]

Originally posted by: Damascus
I have a good idea. SPECint and SPECfp should be used.

agreed

SSXeon

 

[andreasl]

Originally posted by: SSXeon5

Originally posted by: Damascus
I have a good idea. SPECint and SPECfp should be used.

agreed

SSXeon

Do you use any of the following programs on your PC? If you don't, do you think SPEC2K is still useful to compare desktop PC performance?

SPECINT
164.gzip 1400 Data compression utility
175.vpr 1400 FPGA circuit placement and routing
176.gcc 1100 C compiler
181.mcf 1800 Minimum cost network flow solver
186.crafty 1000 Chess program
197.parser 1800 Natural language processing
252.eon 1300 Ray tracing
253.perlbmk 1800 Perl
254.gap 1100 Computational group theory
255.vortex 1900 Object Oriented Database
256.bzip2 1500 Data compression utility
300.twolf 3000 Place and route simulator

SPECFP
CFP2000 contains 14 applications (6 Fortran-77, 4 Fortran-90 and 4 C) that are used as benchmarks:
Name Ref Time Remarks
168.wupwise 1600 Quantum chromodynamics
171.swim 3100 Shallow water modeling
172.mgrid 1800 Multi-grid solver in 3D potential field
173.applu 2100 Parabolic/elliptic partial differential
equations
177.mesa 1400 3D Graphics library
178.galgel 2900 Fluid dynamics: analysis of oscillatory instability
179.art 2600 Neural network simulation; adaptive resonance theory
183.equake 1300 Finite element simulation; earthquake modeling
187.facerec 1900 Computer vision: recognizes faces
188.ammp 2200 Computational chemistry
189.lucas 2000 Number theory: primality testing
191.fma3d 2100 Finite element crash simulation
200.sixtrack 1100 Particle accelerator model
301.apsi 2600 Solves problems regarding temperature, wind,
velocity and distribution of pollutants

 

[mechBgon]

I think everyone here knows how to find out which platform suits their applications best, by doing a little reading at major tech sites and comparing notes with others in the Forum. And the fact is, platforms do well at some things, not so well at others, and each manufacturer wants to look good, so they'd naturally want to base a common PR scheme on something their system was good at. Make sense?

The PR rating scheme is really not meant for us, I don't think. It's meant for the everyday person at CompUSA, CircuitCity, Best Buy, etc, who isn't really aware of the strengths and weaknesses of the different platforms. Is PR fair to them? It's misleading, yes. They think they got a 2000MHz system when they bought an AthlonXP 2000+. But pairing up a Pentium4 with a PC133 SDRAM platform isn't any more fair, since the everyday buyer has no idea that it will be significantly slowed down by this substitution, and Celeron4's on i845GL or i845GLL, with not even an AGP slot for upgrading, is not "fair" in my eyes either. Just my 2c worth: caveat emptor... and that goes for both brands.

AMD joined BAPCo recently. Maybe that's a step towards some commonly-agreed-upon ground for system performance. If a shopper can someday go to CompUSA and see a HP that gets 430 BAPMarks, then go over to Gateway's outlet and find a system that also gets 430 BAPMarks, and can get it home and actually run the bundled BAPMark benchie for themselves and it does get ~430... then we'd be getting somewhere, yes? And for us tech-heads, we can still scoff at the BAPMark and do our homework on how the platform does in our own favorite application, whether it's Doom3, AutoCAD or whatever.

 

[BowlingNut]

i agree that the Pr rating, MFLOP, and MHz/GHz needs to go...but using a single set of benchmarks just wont cut it either.

we all know that some benchmarks favor p4s, and some favor athlons b/c of differences in the design of the processor and optimizations in the code. so why not have some benchmarks written that has 1 set of tests that favor a p4, 1 set that favor the athlon? the utility would run both sets of benchies and then average the scores? i'm sure that somewhere some coding genius could do that. this would be the best "overall" performance indicator i can imagine. its no different in theory than the 3dmark bench used, everything tested in FPS, then report in "marks" - and as far as i know, no one complains that NVidia has an unfair advantage in 3dmark over ATI b/c of architectural differences. furthermore, it'd be hard to cheat since you'd run both an optimized test and an unoptimzed test giving you mean results.
anyone else care to jump on my bandwagon?

 

[BowlingNut]

in addition to my post above i suppose each "set" of benchmarks would have to have a representative of all types of applications: database, graphix, cad, animation, number crunching, etc..

 

[AntiAMD]

I don't think there ever will be a "one size fits all" benchmark for cpu's, but traditionally the industry would see how many instructions a cpu could chew through per second. If I am rendering something in lightwave, mflops would be very important and I would value a cpu that could produce as many floating point calculations as possible. But, if I am web browsing, using office apps, or even playing a game, mips would be more important to me.
Way back when, and I'm talking back to the days of 386/486 or even early pentium chips (60/75/90 mhz), we'd all compare mips. It was a decent way
of showing raw cpu power, and could could be used to compare to Motorola 68k cpu's. Being a diehard Amiga fanboy, I thought my 68000, running at a mind numbing 7mhz, turning in 4 mips was smokin'! (4 mips back in '87 probably was smokin'

 

[BowlingNut]

Originally posted by: AntiAMD
I don't think there ever will be a "one size fits all" benchmark for cpu's, but traditionally the industry would see how many instructions a cpu could chew through per second. If I am rendering something in lightwave, mflops would be very important and I would value a cpu that could produce as many floating point calculations as possible. But, if I am web browsing, using office apps, or even playing a game, mips would be more important to me. Way back when, and I'm talking back to the days of 386/486 or even early pentium chips (60/75/90 mhz), we'd all compare mips. It was a decent way of showing raw cpu power, and could could be used to compare to Motorola 68k cpu's. Being a diehard Amiga fanboy, I thought my 68000, running at a mind numbing 7mhz, turning in 4 mips was smokin'! (4 mips back in '87 probably was smokin'

well, why not report them in mflops and mips then? just have a big drive to educate consumer what mflop measures/relates to, and what mips measures/relates to...and then jsut report speeds in mflop/mip just like how car manufacturers report gas mileage in city/highway?

 

[AntiAMD]

BowlingNut; That would make waaay too much sense, never work

 

[Boonesmi]

if any universal benchmark was agreed on it seems like both intel and amd would start optimizing their cpu's for the benchmark

but who cares about a synthetic benchmark?