No CPU fits my needs

[pointertovoid]

There are dozens of CPU models on the market, but they don't really fit my needs...

I need computing power for old software which is single-tasked.

Sorry designers, I know this is difficult because semiconductor technology and physics go a different direction, but these are my real needs.

I use regularly stone-old software. Some compute molecule conformations, others rocket fuel performance... (MS-DOS era with a graphical interface added!)

They aren't maintained any more, I won't get a multicore version of them, and alternatives for Windows are really scarce.

I bought a Core 2, a Duo because Solo don't exist, a E8600 because it was the fastest on single tasks. Fabulous beast, happy.

Since then, progress has been faint on single-task speed. Clock speed stagnates (Over- or not) and single-task computation power has gained just 20% over all these years, at least on i86 and SSE programs, an improvement too small to let me change my machine.

So if cycle efficiency can't improve more on old code (I know it's difficult) I'd like to have faster clocks. I don't care a bit about 4, 6 or 24 idling cores; I don't need hyperthreading; huge L2 L3 L4 cache size shouldn't bring much - I need one fast core.

And in case this is impossible... Choosing a dual core for single-task performance, not for low cost, I'll never use an integrated GPU, as I also have video games. But I'd like the CPU to run cooler, so I can put a more silent fan, or have passive cooling at the CPU as well. If a recent computer could warm my room as little as my Tualatin did, it would be fine.

Thank you for your attention, Mr. AMD, thank you for your attention, Mr. Intel, sorry for interrupting you.

Edited February 26, 2012 by pointertovoid

[allen2]

Just a few things:

- games love multi-core cpus.

- buying a good to a high end cpu for mobile computer might help increase the performance per watt. BTW that's what i did for most of my desktop/server computers to make them use less than 200W.

- maybe disabling other core in bios might help improve overclocking of a the cpu.

- The other way around would be use those multi core cpu and run two or more VMs (each with one cpu) and run different things in each of them this way you'll gain time.

[pointertovoid]

Quite precisely off-topic.

[Kelsenellenelvian]

Ummm is there a question here?

[CoffeeFiend]

I use regularly stone-old software. Some compute molecule conformations, others rocket fuel performance... (MS-DOS era with a graphical interface added!)

They aren't maintained any more, I won't get a multicore version of them

There's your problem.

There are limits that push modern CPU design in a different direction which modern software makes use of, along with many other options. Such "scientific" obviously software exists, and there's TONS of things to improve performance:

-newer instruction sets and optimizations that a MS-DOS wouldn't use (SIMD instructions for starters)

-running on HPC clusters, or having a "distributed" architecture

-GPGPU computing

-specialized hardware (based on FPGAs or ASICs)

If you insist on using ancient software there's nothing they can (nor will) do for you. And no, they won't base their CPUs designs based on the specific needs of one person in a million (there's much more money to make keeping the other 999999 happy). Especially when their problem is merely that they just won't use modern software, and that modern software happily makes use of modern CPU designs.

TL;DR: you only have yourself to blame.

[blackturbokitty]

Your single core software will run just fine on a multicore CPU. In fact, everything will run better, since the single thread software will only use one core. As someone mentioned before, games love multi-core cpu's, even the single threaded games since the game pretty much has it's own core most of the time while the other cores pay attention to the other background services. You'll notice your older games will be smoother and they will run insanely fast.

Edited March 7, 2012 by blackturbokitty

[pointertovoid]

You didn't get it.

Software to compute combustion in a rocket is not abundant, quite the opposite. It is single-tasked and this won't change.

A multi-core CPU does not accelerate single-task software.

CPUs did not improve significantly on single-task software since the Core 2.

Edited March 7, 2012 by pointertovoid

[CoffeeFiend]

It is single-tasked and this won't change.

Then you won't get any meaningful performance boosts ever. End of story. And it's very much self-inflicted.

You know, people with those needs could rewrite software in modern ways to make use of modern architectures. If they won't then they'll have to live with the consequences. Everybody else can seemingly adapt just fine. It's far too easy to blame CPU designers here when there's lots of options available to make faster software. Pretending it's 1980 and using single-threaded MS-DOS apps won't get you the performance of modern software and modern architectures.

[bphlpt]

You didn't get it.

No, we get it, but you don't seem to.

It used to be that CPU progress was measured by steadily increasing speeds, coupled with increases in width (8bit -> 16 -> 32 ->64), enhanced instructions to do more per instruction, etc. Software was ALL, or at least mostly, written as single-tasked so everyone benefited from these improvements and there was both market and demand for these more and more powerful chips therefore profits for the manufacturers. But then physical limitations became a problem. The combined factors of die size and complexity, feature size, and power requirements, along with the resulting heat and reliability issues, put a practical limit on how fast a chip could go and be priced so that people would buy it. As technology advanced to make feature size small enough they found that they could put more and more on the chip doing things in parallel at slower speeds and get more total work out of the chip at a given speed, power, heat, and cost than they could trying to do things faster and faster serially. But the software has to be rewritten to be able to take advantage of it. Sorry. EVERYBODY has had to adapt to that reality, from the home user to the scientific users of supercomputers. Parallel operations are currently the only practical way to get increased performance. From your first post in this thread you seem to understand these physical realities, so I assume this thread is really just a place for you to vent your frustrations at being constrained to using software from a supplier that is either not willing or not able to rewrite their product to take advantage of modern tools. This was a place to post your wishes, though wanting faster AND cooler flies in the face of possibilities as well. You've heard an expression similar to this before I'm sure:

Faster, Smaller, Cooler, Cheaper, Less Power - Pick ONE.

Market realities limit your options, no matter how legitimate your needs. And of course it makes no sense at all to complain here on this forum that you can't find what you need/want when it doesn't exist and currently can't exist. The only option you really have to go faster with your single-tasked software is super cooling and over clocking. You could probably double your speed by cooling with liquid helium, but this isn't the place to give you much advice about that either.

So all we can really say is:

"We understand your problem. We're sorry we can't help you. Have a great day!"

Cheers and Regards

Edited March 7, 2012 by bphlpt

[dencorso]

+1.

Liquid Nitrogen should be enough, though, and it is way cheaper.

Then again, MSFN is definitely not the place for advice on how to do it.