Are High-End CPUs Worth The Money?

Rampaging Goatbert (aka Jeff Feld) has posted a story at Newsforge about something you may want to argue about with your boss or significant other. Specifically, whether high-end CPUs are worth their high prices. Personally, I look even lower on the processor food chain, but watching those price-curve inflection points makes the runner-up chips pretty tempting. Your mileage will almost certainly vary.

Depends on what you use it for

[Blue Aardvark House]

If you're into high-end graphics or internet gaming, then the fast chips may be a good buy. But if you're doing spreadsheets or simply sending e-mail, than a slower chip will serve you well for years to come.

It's almost analogous to buying a car.

google

[fjordboy]

I think that google's massive (over 10000 units) server farm (all x86) proves that the high end cpu's aren't worth it. Multiple low end CPU's do the same (if not better) job of one high end CPU. I think Google proves this point.

More to it

[Uttles]

It is not fair to compare a number, like clock speed and then say "oh well only 70 mhz more costs 25% more money." These processors have extremely complicated designs and the newer ones are much more efficient in every way. You might see a 70 mhz gain in speed on a piece of paper, but the reliability, speed, and robustness of the new processors far outweigh the price increase.

Re:NO

[Agent Green]

I have to agree with you on this one.

But to stay on-topic, when I bought my A-Bit BP6 a couple years ago (jurassic computing, I know), I got it with two 366s, considering their value was good and the price was more or less middle-of-the road. It didn't take all that long of a period, maybe 6 months or so before I was able to get a couple of 550s to replace them...all without breaking the bank.

So getting the faster chip isn't worth it initially, but the motherboard that handles it is worth every cent. After all, prices will fall as the newer, better, faster hardware items come out.

Well Duh!

[gmhowell]

What hard hitting journalism. An amazing display of analytical prowess. I've had better stories rejected.

Of course the top of the line stuff is too expensive. What the hell is there even to discuss with this article?

(At home, I have a Celeron 466 or so on my Linux box. a PIII 600 or so on my 'doze box for games. Big frickin' deal, right? For the price of a processor upgrade, I can be running 1GB of ram in both systems. Through in another 100 bucks, and I've got more disk space than on the file server here at work (which is no slouch for what we do)).

Guess what? Processors don't really matter anymore. Neither does any of that hardware. What in the hell is anybody doing with computers that requires all of this horsepower? Yeah, something will come out. But what, and from whom? Don't we have enough cycles to have incredible voice interfaces? No, because everybody (and by that, I mean Joe Six Pack, aka, my mom) needs M$ bloatware to do anything. It's because Quicken wants to do so much that it takes many megs of RAM to load. Why???

Slashdot latest headline:

Top of the line stuff gives marginal improvements for mega price increase.

Christ, we knew that back when it was a 486-20 mHz vs a 486-25 mHz (and probably earlier). Christ on a crutch, how is this news?

I think I know how stories are picked: each one is printed out. One of the editors grabs a stack and wipes. Whatever story isn't covered in it gets posted.

Excuse me, I must go beat my head against the wall.

(And please, anybody who wants to mod this down, I would much prefer it if you answer my question: why the fuck does this matter?)

A Chain Is As Strong As Its Weakest Link

[Steve B]

Most systems with high-end CPUs have the real bottleneck somewhere else (memory, motherboard, graphics). A lot of systems out there would benefit more from another 128 MB RAM than another 0.2 GHz of CPU speed.

Re:google

[JWhitlock]

I think that google's massive (over 10000 units) server farm (all x86) proves that the high end cpu's aren't worth it. Multiple low end CPU's do the same (if not better) job of one high end CPU. I think Google proves this point.

How many fps does Google get?

The article is in the context of buying a PC for personal use, and benchmarks using FPS, ray-tracing, kernel-compiles, etc. The idea is to pay attention to incremental performance (1.33 Mhz to 1.4 Mhz, .07 Mhz) versus incremental cost ($33? $100), and make sure it's worth it. Bottom line, buy cutting edge, get screwed on price.

Re:Performance

[Anonymous Coward]

I've always been a large fan of using an army of small, low-powered boxes instead of one big expensive box. For one thing, if something breaks, everything else still works. For another thing, it's generally cheaper this way.

How many fps do you get on distributed Quake?

Oh, that's right - there is no distributed Quake. Sometimes, you have to put all your processing power in one computer - which is what the article was talking about.

The case for High end CPUs

[3ryon]

There is another reason to buy the high-end CPU that I haven't seen listed. If you are going to own the computer for 3+ years you'll get more milage out of that faster CPU....typing this on a three year old 233.

Bureaucracy

[Xenu]

If you are working for a government or large corporation, you may be better off getting the expensive, cutting edge machine if you are going to be stuck with it for the next 5 years.

Typing this on a blazing fast P5-233, and this is the _fast_ machine in my office.

Look at total system cost

[rgmoore]

It's rather silly in a case like this to look at just the price of the processor, disk, etc. You have to look at the price of the whole system and decide what kind of tradeoffs you need to make. Is $33 worth it for a 5% increase in processor speed? That depends on how much the whole system costs; if the system costs more than about $700 then the $33 is less than 5% of the system price and it may be worth it to pay more for the extra speed.

The case when this really kicks in is with expensive proprietary software licenses. I've seen various programs that I might want to use in my work that have license fees in the thousands of dollars. In some cases that's the price per box, but in others there's actually a per-CPU license. If you're running somthing that costs $5000 per CPU, it makes sense to spend some fairly serious cash on getting the fastest possible processor.

Re:Just Computer Hardware

[YIAAL]

This is what happens when you have a capitalist government. A capitalist government? We've never seen one of those, that's for sure. This is what happens when you have a (more-or-less) capitalist economic system. Poor people get a lot for a bargain. Rich people pay a premium to feel special. The premium that the rich people pay help companies charge less for the lower-line products that poor people buy. Everybody's happy. Rich people aren't forced to contribute (except by their egos) and no political bureaucracy dissipates the money being redistributed because it makes economic sense for companies to act this way. Hurray for capitalism! Thanks for pointing this out!

A lot of people will pay the price for the best...

[rekoil]

The reason for this pricing is to maximize the profit reaped from the "money is no object" buyer - the one who will say "I want the fastest chip you can put in a PC" and not worry about how much it costs. You'd be surprised how many of them there are, and how much of a chip manufacturer's profit comes from these buyers. These are the same people that spend $400 on a video card to get them 50% more frames per second than a $150 card. Again, you'd be surprised how many of them there are.

A little perspective

[Courageous]

I cost my employer a little more than 4 cents
a second to employ. So, if a CPU costs 40
dollars more, a mere 17 minutes saving to my
time pays for the difference.

C//

Re:Just Computer Hardware

[cancrman]

True regarding the performance of the two cars. But if you drive a Lexus people think you're successful. If you drive a Toyota you're just another schmuck. The problem with your arguement is that you are disregarding how the two separate brands are percieved. Automakers do this with countless models. Honda/Acura Accord/TL, Nissan/Infiniti Pattfinder/QX4, Toyota/Lexus Land Cruiser/RX430. If you're worried about bang for the buck performance on a car, go out and get a camaro (ick).

Anyway, your car comparison doesn't fly when put in the same context of the article. If you buy a 1.4 Athlon vs a 1.33 Athlon you'll end up saving about 6 second of kernal compile time (read the article). If you buy a Lexus instead of a Toyota you're buying status.

Pete

Price??

[crandall]

High price? What are these people talking about? My 1.4 ghz athlon 266bus cost me about 190$ USD. Compare that to same time last year, when the top end AMD processor was 400$, and the top end Intel processor was 600$. Processors expensive? Maybe if you live in a trailer park.

High end cpus are a waste of money

[plopez]

Most of my work these days is in databases. It is MUCH more cost effective o spend he extra money on the fastest memory, memory buses, disks and controllers you can get. Most Intel type cpus spend an inordinate amount of time waiting on IO. Faster IO means a faster system. In addition, on the low end Wintel systems, SMP is a joke. I have yet to see a system running with more than ~75% utilization per chip. In addition, in database systems, Oracle and MS both charge by number of processors and MHZ. So going single processor with a slower chip can save a considerable amount of licensing costs!

For a DB system the rule is 'fast disks, fast memory, fast buses, fast controlers', for heavy network traffic (lots of web hits), get the fastest networking you can afford.

And remember, MHZ is only part of the equation on processors. If you really need (and few people really do) a fast chip, good and large L1 cache is a bigger win than raw MHZ.

My $.02

Never buy the newest, never buy the cheapest

[bADlOGIN]

With the newest technology, you're the pilgrim taking the arrows to see what quality control may have let slip (early adopter syndrome). P60's that double as hotplates, Zip dirives with the "click of death" come to mind. If you buy the cheapest available (most often the oldest), you run the risk of technical relevance and quality of support (why is it so darn cheap again?). I like sega, but if you don't own a Dreamcast, do you want to sink $49 into one at Xmas "just 'cause" when that could be a Playstation 2 or GameCube game?

I don't think Joe Average consumer goes wrong with any technology buying somewhere to either side (or on) the middle of the road. Taking the leading edge or the trailing edge is the sure way to get taken as a consumer.

Re:You cost $300,000 a year?

[Anonymous Coward]

I doubt his paycheck, even before taxes comes close to 300K/year.

But it costs your employer much more then your salary to employ you.

There are costs for benefits (insurance, 401K, etc.).

There is the literal overhead, ie. the roof above you and all the other facilities, utilities, supplies, etc..

Then there is operational overhead, your supervisors on up the chain, secretarial, administrative, etc..

And finally profit to keep the shareholders happy.

My pre-tax pay is about 75K/year, but if you add up my charging rate, it comes to about $200K/year. Wish I took all that home!

So 300K isn't unreasonable if he's highly paid and maybe works in a very high cost area in a high overhead industry.

And yet, my boss pisses away hours of my time instead of getting me the hardware to do the job right :-(

YES!

[Bitmanhome]

Like the noisy AC said back there, ALL CPUs ARE CHEAP TODAY! If the computer is for work, then even the most expensive computer will easily pay for itself. If the computer is for home and doesn't earn you any money, then the only question is: How much do you wanna spend?

It's not a question of price/performance, it's of price/happiness. If the dollars make you happier, then keep them; if the megahertz makes you happier, then buy them!

-B

The Rule of thumb for CPU's

[leonbev]

A long-standing guideline for purchasing CPU's has been to buy 1 notch below the absolute latest in technology. That way, you can get about 80% of the performance that the newest product offers, at about 60% of the cost. That way, you can get the best price-to-performance ratio, and have some money left over for other computer components. The "cutting edge" technology almost always has at least a 20% price premium attached to it, and should be avoided whenever possible. Save that money and spend it on something else, like more memory or storage.

Re:Performance

[Deflatamouse!]

One addition to the operational costs: older machines (esp. 486s and early Pentiums) runs HOT! They are not as power efficient as some of the newer chips in terms of performance/power usage ratio. So if you take into account the amount of electricity sucked up by these little machines over its life span, you will probably want to trash your old systems and buy a new one (not P4s though!)