Hyperthreading + dualcores = multithread sw = nice for BP6?

[Wolfram]

What do you think: Will the presence of hyperthreading and dual cores encourage software developers to write multithreaded software, from which our good old BP6 might benefit?

[purrkur]

Yes, hopefully that will be the case.

However, I am against HyperThreading actually. I would say that a dual-core cpu is the poor mans SMP system and I like what I have seen so far of that technology but I dislike HyperThreading. I have seen cases where this technology has reduced performance and caused instability in some programs. I have also seen it increase performance in a few applications but the performance gain is so little that I would rather just buy a faster CPU if I needed those extra 10-15% of CPU that I find HyperThreading giving you on a good day.

At the end of the day I would say that HyperThreading is a marketing ploy from Intel used to try to justify the horribly high prices on their higher-end CPU's. Also, in order to fully explore this technology you need a compiler that has built in support for it which means that most software doesn't really take advantage of it. I think that this technology will turn out to be the next MMX type of technology from Intel. The idea is good and it looks good on paper, but in the end of the day you gain close to nothing so marketing uses it to justify prices. The bad thing is that the market buys it, just like they buy the "GHz means everything" ploy that Intel has been running before they brought Pentium M and Dothan technology to the market.

As you may see, I am AMD all the way, except for my wonderful BP6...

[hyperspace]

Yeah, I agree. I think HT was...

Hey, look what we can do with the idle registers and CPU cycles we have with these faster processors. Virtual processors are just that, not real.

Multi-Core is the wave of the future.

[Wolfram]

But will dual cores be treated like they were single cpus? Or will optimizations for dual cores have to be different from optimizations for smp systems?

[hyperspace]

But will dual cores be treated like they were single cpus? Or will optimizations for dual cores have to be different from optimizations for smp systems?

If the industry does it right, mainboards and software will be "re-designed" or optimised for this new architecture. I suspect that Multi-Core will be more marketing hype initially.

[purrkur]

The advantage with multi cores is that you actually have two different cores (with own ALU's etc). If the chip design is done right then each core should have dedicated onboard L1/L2 cache. You don't want them sharing cache which could reduce performance for the two cores (in cases you really use the cache) because of sharing issues. In that respect you can probably get the same performance as a dual CPU machine that doesn't have dedicated memory per CPU (like the BP6). The ultimate gain of having real dual, quad or whatever cpu's is that on higher end systems you can design the board so that each cpu has its own dedicated memory bank. The less the CPU's or cores have to share, the better the performance.

In theory it is possible to design each core with an address bus to their own dedicated memory but I but I don't think that is happening in the designs that will be coming up next year.

Also, multithreading isn't always the answer. More threads means more overhead for the kernel which isn't always desirable. For example, at work we have an application running on Solaris that is single threaded. We ported it over to Windows and made it multithreaded in the process. We see absolutely no performance gains when running on Windows. This may sound strange but we have thoroughly tested the app and the gains are simply not there. In fact, we have ported the app over to Linux (where it is also single threaded) where we can make direct comparisons on the same hardware and I would say that Linux has the upper hand, although not by any noticable means.

[Wolfram]

Just read that dual core Opterons and Athlon64s should run in already existing boards, only a BIOS update would be required. So no dedicated memory for now.

http://www.planet3dnow.de/vbulletin/sho ... did=170830 (German)

For the multithreading thing: What type of application was that? Probably not something average home users like me would run, right?

[purrkur]

For the multithreading thing: What type of application was that? Probably not something average home users like me would run, right?

You are right. It is a server application that connects to a database (Sybase on Solaris and Linux and MS SQL Server on Windows) and takes care of client subscriptions (among other things). Our largest customers run maybe 250-300 connections to it at any given time.

[Holodeck2]

multi-threaded apps are HARD to write!
you need to optomize like crazy or else you'll get a preformance loss

multi processor programs are harder to write. to make it efficent you need to bypass a lot of intergrated OS stuff to get it to work the way it's designed to work. I kill entire forests just trying to keep track of everything that's happening in the program

[Wolfram]

So I guess you'd say dual cores will increase the general responsiveness of a system, and a few programmers may write multithreaded software if dual cores become a widespread feature, but that's it?

[purrkur]

So I guess you'd say dual cores will increase the general responsiveness of a system, and a few programmers may write multithreaded software if dual cores become a widespread feature, but that's it?

Responsiveness is a good thing If one cpu is working hard your system has another so your system won't (for the most part) bog down if you try to do other work.

I think multithreaded software will become more popular. It is just a matter of having the hardware support for multiple cores/cpu's, and better support for writing multithreaded code in compilers, etc.