PDA

View Full Version : AMD working on new PC rating method


^7_of_9
21st August 2002, 11:27 PM
Chipmaker Advanced Micro Devices is continuing its fight against the most common way of rating computer performance--a method that relies on what AMD calls the "megahertz myth."


Last summer, the Sunnyvale, Calif.-based company launched its True Performance Initiative, urging consumers to question the notion that a PC with a faster chip will always outperform one with a slower processor. Now AMD says it has joined with other members of the PC industry to develop a new measurement, one that would take various factors into consideration to more accurately reflect the overall performance of a computer.

"We've been working with industry leaders today to propose a solution...to come up with a better way for end users to evaluate what they're really getting," Patrick Moorhead, vice president of consumer advocacy for AMD, said Wednesday--the same day AMD introduced two new Athlon XP desktop PC processors. Moorhead said AMD is seeking feedback from software developers, as well as from other PC-component makers.



Consumers often compare processor clock speeds and prices on various new PCs. But the lowest cost PC with the highest clock speed processor might not always offer the best overall performance, AMD maintains. The company has argued that a less expensive machine with one of its own 1.8GHz Athlon XP 2200+ processors can perform as well or better than a PC using archrival Intel's 2.26GHz Pentium 4 chip.

But some PC industry players might ask why another performance measure is needed when a host of PC evaluations and performance information is available in the form of reviews in magazines and on the Web.

Currently, PC makers can cite a number of measurements, including benchmarks--tests that score performance based on how quickly the computer handles a certain task--processor clock speeds; and even the performance of various components, such as memory or graphics cards. Reviewers tend to use benchmarks, such as Business Applications Performance's SYSmark, to measure the performance of desktop PCs.

But AMD says any one of these measurements tells only part of the story and that trying to juggle all of them only muddies the waters for people intent on bringing home a new PC, especially first-time buyers. It proposes a method that would take all the different factors into consideration and produce an easily digestible rating that buyers could consider at the store, without referring to a host of intimidating reference materials.

"Light bulbs have better information about them at the point of sale than PCs," Moorhead said, adding that what's good for customers is good for the business--currently in the throes of a major sales slump. "A confused buyer is a buyer who sits on the sidelines. That's not good for the industry," he said.

While most hardware makers are usually eager to join in on new PC industry standards, those contacted by CNET News.com said it was too early to comment on the new proposal, which has not yet been finalized. Intel, for example, said it hasn't yet seen any new proposals regarding performance from AMD.

Moorhead said AMD hopes to release the new PC performance measurement by early next year. "We're in analysis mode" right now, he said.

jema
22nd August 2002, 06:18 AM
Seems like an impossible goal, for a start you would have to divide it into two groups, one where 3d graphic card performance matters and one where it does not.

jema

LinearB
22nd August 2002, 12:41 PM
Have to agree with Jema on this one. The performance of a processor is largely dependent on what you use it for. Even putting to one side a moment the 3d graphics issue, even across the distributed computing clients different processors and hardware configurations can have a huge effect on performance, and what about power requirements / motherboard & RAM costs / ease of upgrade etc. ?

I for one am happy for processor manufacturers to just let us know the price and it's clock speed and sites like ours can then spend hours discussing the relative merits of each one :D

mackerel
22nd August 2002, 04:46 PM
Originally posted by LinearB
I for one am happy for processor manufacturers to just let us know the price and it's clock speed and sites like ours can then spend hours discussing the relative merits of each one :D

For informed buyers that may be the case. But for the masses GHz rule and that hurts AMD.

Benchmarking a CPU will always be a hot topic for debate. Since AMD are trying to compare their CPUs with Intel, are they going to bench just the CPU or the whole system, at which point a whole bunch of other variables come in.

Even for the relativly simple case of a direct CPU to CPU comparison, choosing the code to use will be difficult. Given the fsb matters I can't see how they can escape benching at a system level.

This leads me to one option which I don't think CPU makes will like, as it relegates their product somewhat in importance. That is for an industry wide application level benchmark to be created and the scores for each PC to be compared. What form that benchmark could take is another discussion...

mackerel
22nd August 2002, 04:56 PM
Thinking more, maybe some sort of education of the masses is required.

Take cars for example. You don't just go out and say that one with a 3.5L engine is "better" than a 2.0L. You'll also look at other figures like fuel economy, acceleration, insurance costs, number of seats... so it's not beyond someone to compare figures.

So a system benchmark would produce a few figures since you'll never get agreement on how to arrive at a single unified figure balancing different requirements. Maybe have a score for each of 3D gaming, office stuff and multimedia procesing.

zeke
10th September 2002, 02:01 PM
How about rating CPU's in MIPS or millions of instructions per second?

"Light bulbs have better information about them at the point of sale than PCs," Moorhead said,

Oh, and always buy the 130 Volt filament light bulbs. ;) They last longer. :D

mackerel
10th September 2002, 04:43 PM
Originally posted by zeke
How about rating CPU's in MIPS or millions of instructions per second?

The trouble then is which instructions you pick. The P4 can do a bunch of SSE/SSE2 instructions in no time, and with its massive clock speed would grind AMD into the ground. But, if you use regular x87 FPU instructions, then AMD would kick back fighting.

Oh, and always buy the 130 Volt filament light bulbs. ;) They last longer. :D

Not if you plug them into good old fashioned 240 volts we get around here :D

jema
10th September 2002, 07:35 PM
Originally posted by zeke
How about rating CPU's in MIPS or millions of instructions per second?



Always way too simplistic, as processors have such a large cache of core memory, tests like these tend to operate purely in cache and thereforfe ignore the speed of system memory access. Let alone the issue that for a home user, it is normally pretty vital to have a good 3d card and for a business user, disk speed may, or may not be an important issue. You will never be able to measure PC performance in a single number... another example of this would be FPU performance, an old amdkII450 chip would be pretty fast for word processing etc, but slower than a PPro200 for games like quake and things like Seti as its FPU sucked badly :(

Real life performance is also effected by ammount of memory and whether the operating system can use it well. If you are processing digital images, a super fast CPU with 128Mb of memory would when dealing with images of 100MB+ work worse than a far slower CPU with 512Mb due to memory paging to disk.

So lets see now we have:

1) Core processing speed measured by some fair means, involving main memory access.
2) 3d performance. A contraversal area in itself.
3) disk speed.
4) FPU performance.
5) Amount of memory.

Unifying that lot into a single number or two is not going to be anymore true as a result than the intel ghz con.

jema

phoenix
12th September 2002, 02:56 PM
Originally posted by zeke
How about rating CPU's in MIPS or millions of instructions per second?



Oh, and always buy the 130 Volt filament light bulbs. ;) They last longer. :D


maybe by a millisecond or two, myself I prefer 250V bulbs :D

LinearB
12th September 2002, 03:42 PM
Originally posted by mackerel
Thinking more, maybe some sort of education of the masses is required.



Masses are swayed by advertising not education, For Mr or Mrs Joe Public, Either processor camp will provide all their needs, so it comes down to cost and what did next doors buy ;).

And when it comes to cars,

Men, does it come with alloy wheels ?
Women, Do they do it in mauve ?

/LB runs for cover ;)

mackerel
12th September 2002, 04:57 PM
Originally posted by LinearB
And when it comes to cars,

Men, does it come with alloy wheels ?
Women, Do they do it in mauve ?
/LB runs for cover ;)
Hey! My next car will be metallic purple if they can do the right shade!

Reminds me of something I saw on Top Gear GTi, they had some snooty woman review some big sporty Rover. At one point she pointed at the spoiler declaring it unnecessary. Typical - missing the point altogether! It's a sporty car, not a motorised shopping cart.

zeke
13th September 2002, 05:17 PM
Opps, sorry for my ignorance. You guys have 250volts running around your houses? Must be lots of electrical fires over there. :eek: Well most light bulbs here are 120volt but you can buy "rough service" or "country service" bulbs. The filament is larger and rated at 130volt.

I have heard that Britons have to pay for every phone call you make as well. Is this true?

mackerel
13th September 2002, 07:21 PM
The nominal mains voltage in the UK is 240 V, but with all the european stuff happening it has been creeping lower since bits of europe are 220 V and they are trying to standardise everything. Almost everything bought in the last 10 years will be happy anywhere in that range so it's no problem.

As for paid for phone calls, as standard, all calls are charged. Only relatively recently has there been any change, where if you pay for a higher level service, you can get some call types (local voice) for "free". And to make everyone unhappy, they often seperate out internet calls, requiring a seperate payment...

jema
13th September 2002, 10:21 PM
There are now more different phone tariffs than you can count! with all sorts of caveats, for example with NTL you can ring other people on NTL for free evenings and weekends, you can also sign up for "prefix" services (though not if you are on NTL!) that allow cheaper calls to certain places, for example Australia can be done for 4p a minute :) less than some standard rates for national calls!

jema

prokaryote
15th September 2002, 06:49 AM
Devise a by use rating system with "standard" equipment that is updated every couple of years or so.
So, to calculate the rating, an independent house would measure N pertinent parameters for each processor.
Each parameter would have to be made commensurate (normalized) by any one of dozen's of methods. This independent lab/house whatever would be like the JEDEC organization to arrive at the consensus details (what parameters, what equipment for each use category, etc.).

Processor ........Parameter A...Parameter B ... Parameter N
A
B
C
D
:
:
X


There would be a concensus about how important each parameter is for a particular use.
Rating Weighting by Use Factors
Business........20% 40% 5%
Gaming..........60% 10% 0%
Internet/email.10% 70% 10%
:
:
Scientific.......70% 5% 0%



Final by use rating = Sum(commensurate parameter (i) * parameter use(j) rating)

This final, by use rating would have to be normalized to some standard system.
So the end rating product that the public actually sees might look like this:


Business.........100
Gaming...........40
Internet/email..110
:
:
Scientific........70


Or if people need a visual, then a bar with an indicator arrow for each use category can be adopted


Business.........Lowest ---------------^------------------ Fastest
Gaming...........Lowest ------^--------------------------- Fastest
Internet/email..Lowest -----------^---------------------- Fastest
:
:
Scientific........Lowest ---------------------^------------ Fastest


Note that the system doesn't have to be 100% accurate in all cases, but should give the consumer a good idea if the bargin basement processor that they're buying is actually good for any of the typical uses, or if the top of the line is really that much better than some mid level processor.

zeke
15th September 2002, 10:42 AM
Or maybe just publish how long a Seti unit takes? :D Just kidding eh. That sounds like a good idea Prokaryote.

jema
16th September 2002, 03:14 AM
I'll play devils advocate and say it would never work, apart form gaming the categories are two vague.

Take Seti as an example. if you are a scientist at Borkely running the server, then you don't want FPU performance, you want server disk performance, businesses on the other hand have a great number of different needs....
As for gaming, the things to rate 3d cards on change yearly or quicker, so the ball park constantly changes, and it is contraversial, just what you should be measuring with a 3d card, e.g. how much to rate raw fill rate and how much to rate a great feature that is not used by anything but cutting edge games.

jema

prokaryote
16th September 2002, 04:39 AM
Originally posted by jema
I'll play devils advocate and say it would never work, apart form gaming the categories are two vague.

Take Seti as an example. if you are a scientist at Borkely running the server, then you don't want FPU performance, you want server disk performance, businesses on the other hand have a great number of different needs....
As for gaming, the things to rate 3d cards on change yearly or quicker, so the ball park constantly changes, and it is contraversial, just what you should be measuring with a 3d card, e.g. how much to rate raw fill rate and how much to rate a great feature that is not used by anything but cutting edge games.

jema

I think we're talking processor with some set of agreed upon equipment set. The specifics would be where the "standards committee" would come into play (i.e. JEDEC or some such for package, MPEG for video compression, etc.). Wouldn't be easy as every processor company would push their agenda, but it would make the marketing / consumer choice easier. Otherwise, it's no use and the masses will judge by what ever company throws the most money into advertising to tell them what it is that they want. The categories were thrown as an example, not as set in stone. I'm sure that the interested parties could hammer it out. The point being that use categories would be the "executive summary" that the public sees and those that care could review the parameter ratings and make up their own minds as to the weighting. Let's face it, most people wouldn't know the subtlties involved and frankly wouldn't care, they just want a general idea that they're getting a processor the fits their somewhat ambiguous needs. At least having a use rating would get them to think about what they're going to use the processor for and hopefully think about what kind of components that they're plugging it into. A use category could also be devised on the end machine, where it would be up to the OEM to adhere to the standard and provide a similar use rating for the machine. Granted, no relative comparison will satisfy everyone all of the time (or some, none of the time), but it gives the rest some semblance of what they're getting. Otherwise they're just taking a shot in the dark. Such uncertainty in the market place is not conducive to improving PC sales, since people will grow ever more reluctant to upgrade every few years if they don't have some measure of "worth" that the new technology brings. Imperfect information that is at least somewhat accurate is much better than the chaos of uncertainty, especially in hardening economic times. Just some thoughts.