In defence of multi-core

http://www.bit-tech.net/columns/2006/11/05/In_defence_of_multi_core/1.html

 

Good article. I gotta agree with it. If Intel say we're going to be at 80 cores in 5 years, even if we're only at a quarter of that in 5 years - that's massive change. Parrelism has to be the future since CPU's don't seem to want to go hugely faster then they already do these days anyway.

 

I spose that Ferrari you promised is in the post to replace my Clio Im driving to work with?

 

There is still some life left in current technology, but CPU's still have a long way to go in my opinion. I believe intel were playing around with using Light processors. some sort of optics i guess but the point i'm making is there are other technologies out there which could potentially make a single core a heck of alot faster. For the moment multi core seems to be the way to go. What the future brings is really unknown.

For the moment quad core seems a little pointless although it obviously has its uses. I suspect it will only really be useful for higher end users to begin with doing video encoding etc. I for 1 would love to get my hands on one

 

He promised you a Ferrari ? Can i have one too?

 

I've recently decided to update my ye olde PC - almost all had to go since Im still stuck in AGPville with my sn45 shuttle - I went conroe because its the best around I also took measures to get a motherboard that would accept the new quad cores should they be worth it when they come around! The valve stuff basically convinced me to upgrade, my main machine is for games and multi-core wasnt worth it but now it seems it definitely is.

 

Of course, I don't deny that they do. But around the 5-6Ghz mark seems to be the absolute maximum possible for the high clocking chips right now. It's not much of an improvement really, so paralell is for sure the way to go.

 

I am ready for 80 cores now! I can easily think of uses such as replacing most addon products we currently use like graphics cards, digital sound processors, and tv cards for recording and encoding multiple tv shows in h.264 while watching another. and all this while doing storage processing for downloading vids to a raid5 array, dictating an email using advanced speech recognition, and ripping a dvd in mpeg4. I can definately find a use for more cores just on the basis of multitasking and replacing most other processing devices we currently use on motherboards to make a much smaller, less complex, cheaper, cooler pc. I say bring on the teraflops!

 

love it.

 

Well written. My complain about technology is that there has been 1GB models of ram for over three years now, and that still all I can find even though mobo's support up to 32 GB's on 4 Dimm's. I just consider it a conspiracy from Msoft for the need of system ram for Vista.

Bindibadgi, please change your avatar....

 

i have simple use for as many cores you give me any DC project out there

 

Good article

I don't understand why some people are not getting it, at the moment it's probably the only cost effective way to progress cpu's processing power until somebody actually makes a working (and affordable) replacement for sillicon based chips.

There must have been plenty Why-ners ranting away when the first personal computer was invented... "you honestly think theres going to be a use for that!?". Sometimes things just have to be made before real uses are actually realised for them. I know quad core is in no way as important but any step, whether it's success or failure is part of the evolution of computers.

 

Actually not really. The first computer was invented to improve the efficicency of enigma cracking IIRC. So there was a purpose for it when it was invented. Your point is right though.

 

to Nature: Please consider that Bindibadgi is a moderator and his avatar is cool, in contrast to yours that is a simple and boring pie chart. *me hopes that Bindi makes his magic on your avatar*

as for "you honestly think theres going to be a use for that!?" consider this:

http://www.killernic.com/KillerNic/

i am not saying that multi core is not the path to the future, i want a multicore machine.

sorry: I WANT A MULTICORE MACHINE AND A FERRARI.

edit: specofdust there were some computers made before WW2 even if they were all mechanical.

 

Here's my question though: I am going to be upgrading my PC this X-mas as it is 2/3 years old and hasn't been updated since I bought that state-of-the-art 6600GT.

My problem now is this: I can buy a 7950GT with a 4000+ single-core, or I can make do with a 7900GS and include a 3200+ dual-core processor. Where's the tradeoff? Are current applications going to be able to utilise that second core?

Tell me this, Mr. Harris (or anyone else), is multi-core viable for the gamer on a budget buying in a month's time? Or is it wise to save cash now on a slower single-core CPU and wait for cheaper multi-cores?

I know I'm using my own situation as an example but my argument is that I wonder where multi-core becomes useful right now, and not a year in the future.

 

smoguzbenjamin has a point there.....

 

The price of dual cores is significantly cheaper then it once was. You can get them for <£100 (X2 3800+) so I'd say that they're worth it.

 

Well, personally I've been a "loud critic" of multi-core. And the Short of it, is "I think you are wrong Wil"

The criticisms (at least the rational ones) of multi-core are not "We don't need faster CPUs". That is the argument you are essentially debunking. Of course we do. The actual argument against multi-core is that a multi-core chip is not your traditionally "faster" CPU. It's "faster*", with the little asterix disclaimer.

2x1GHz cores does NOT equal 2GHz. 4x1GHz is even less equal to 4GHz.

Multicore only equals faster, for the types of applications and uses that can benefit. Those uses are by no means uniform and all encompassing. Multi-taskers (but as the # of cores goes up the benefit goes down as one can only multi-task so much) and multi-media encoders for the most part. Current games hardly at all, and hopefully more in the future (*crosses fingers for Valve*).

Multi-core is not the "next natural step in CPU design". It's a cop out. It's a financial move pure and simple. It is arguably unavoidable (moores law can only fight the laws of physics for so long), but that doesn't mean we should have our head in the sand about the disadvantages. It's cheaper to make two slower cores, than one faster. 2 slower cores are LESS robust than 1 faster one. If it were cheaper to make single-core chips, then multi-core wouldn't even be on anyones radar.

So you can't equate people who ask the "Why?" of multi-core, with those that are against progress.

No one is discounting the usefullness, but we can't simply treat the linear scaling of cores as the same with the MHz race of the past. If I had the choice between an 8x1GHz chip, or a single core 5GHz, I'd take the single core hands down. The performance benefits of multi-core come with a catch, and that catch is often overlooked. Thats all the criticism really is.

Aggies

 

Stop asking, the answer is NO.

But your argument is flawwed aggi: because a 2GHz CPU can't do twice as much as a 1GHz CPU. It always the law of diminishing returns and a combination of clock speed and cores allows someone to maximise those diminishing returns. You can't tell me you dont multitask? Remember the days in Win95 when you had to do only one thing at once?? Now everyone has MSN, torrents, Firefox, music/video, email, anti-spyware, anti-virus, internet connection etcetc. Where will this be in another 5 years?? If you dont look ahead and offer people the power to do something, they wont, and negative thinking never got us anywhere.

 

I could be wrong on this one, (although I dont' think I am, heh), but given the same architecture, a 2xclockrate increase by definition means a 2x increase in speed(instruction completion rate)? (Edit: Upon further reflection the issue of Clockspeed vs performance is really a minor point. I use GHz as simple a simple way to gauge single-core performance. So if it doesn't scale linearly, I just mean performance, not specifically the clock-rate. So you could say "a dual core chip is not equal to a twice-as-fast single core chip". The GHz numbers make the math alot simpler to follow though ) I still think the point itself holds though

I'm not talking about GHz ratings in general (how AMD is slower than Intel, and the new conroes are clocked lower than AMD but are faster), just how they are a good (literal) performance indicator between relative members of the same architecture.

I would agree about diminishing returns though, as it becomes harder (takes more $$$) and harder to get faster and faster. It's not the performance that diminishes relative to Mhz though, it's the performance increases that diminish relative to money/time (R&D) spent creating it. Which is why I mention the little caveat how multi-core is essentially unnavoidable/moores law thing.

It has been a while since my CPU architecture course, so I could be totally out to lunch on this one , but unless I'm missing something obvious I'm pretty sure this is the case?

Multi-core "required" is at least a year out.

My guess is that current/soon Multi-core "using" games (Crysis, ET:QW etc) will only see performance improvements of %30-40 at best (probably more around %20) (compared to the theorhetical %100).

However CPU prices are such that it's is rather short sighted to not go multicore at the moment, if you are upgrading. (I'm upgrading from a 6600GT and a64 3200+, to a 4400+). Unless you plan to upgrade again in a year.

In general the trend is: more GHz to keep up with faster graphics cards (faster GPU's need faster CPU's to achieve their potential), more cores to support the increased functionality in games. (Ie. physics, fancier effects, etc). So GHz affects FPS while multi-core affects gameplay, if you want to look at it like that. This won't really come into play until maybe a year out or so.

However for non-gaming uses dual core is a great benefit. Just for OS responsiveness alone (the small pauses you get all the time) it's great to smooth out the experiences. I use yahoo mail beta, which is all javascript and still buggy, so it very often will block and demand 100% cpu, which momentarily locks up my machine. Dual core will help to smooth out stuff like that.

The equivalent dual core chip of a singlecore is about 2x the price, or $100(CAD) more. (Just put in an x2 3800 for someone recently).

If you plan on upgrading again in a year, and you want to get the most impact on your gaming TODAY, and you are also on a budget, then money spent on more cores is probably better spent on more GHz. Ideally, if you had no budget, you'd go for a faster chip, that is also multi-core. (The diff between my +3200 and +4400 is 2.0GHz and 2.4Ghz (which should overclock to ~2.6Ghz) which is the top of single-core AMD speed anyhow.

I'm waiting for a DX10 card, so I figure multi-core will be usefull to allow that thing to run as fast as it can. Plus I need it for day to day OS smoothing out. And I'l hopefully start encoding my TV show DVD's to mp4 for easier viewing

So yeah, the benefits for gaming of dual core are still a ways out. If you are on a budget, need to upgrade now, and will probably upgrade again in a years times, put the $$ in as much GHz as you can. However in the future it'l be GHz for fps and Cores for gameplay (like how the Valve article explains).


Aggies