Hexacores and You
Seems like AMD and Intel are going mad with their multicore offerings which are now starting to creep into the many-core category. Intel has already released their Intel Core i7 980X and its chief rival AMD is set to release its set of hexacores next month. With all this multicore frenzy, one must question, is it becoming too much?

(AMD Opteron Istanbul die courtesy of Anandtech)
Any basic computer science student can tell you, the more cores you throw in, the higher the performance. While theoretically true, it's not that rosy in the real world. Yes, parallel processing in terms of CPU technology is better than the old serial processing as we've found out during the mid to late 2000s. We could only clock CPUs up to 3.8 GHz without having to use liquid nitrogen to cool down the processors. The idea of having more cores also sound great as it means you can just divide up your processes and just offload it to other computational cores on your computer. I myself own a quad-core system.
However, the issue is that what is traditionally processed by a CPU are serial tasks. To make this easier to understand, let me use an example. For instance, you're playing a game, and you want to fire the gun at a sheet of metal. In order for the end result to happen, the computer first needs to calculate the fact that you triggered the event of "pulling the trigger on the gun", then it must calculate the physical trajectory of the bullet then have to calculate the damage caused by said bullet before it prints a bullet hole on a sheet of metal. You can't really parallel such a process as you need A to happen before B can occur. Software developers especially game designers are having a really hard time trying to get their engines multithreaded with more than 2 cores. Hence even with 4 cores, most games will not show a performance gain compared to a decently clocked dual core. This issue isn't only affecting the gaming world but including other applications like Microsoft Office, and other productivity software as well due to the way they are designed.
So why the big push? Servers. Server applications like web servers, database servers and other transactional servers work great in parallel. In addition, graphical and video work run great using parallel processors. The only way I can make my desktop crawl is by transcoding a 1080p video on it since my transcoder can fully take advantage of my quadcore as it can split all the frames up and let each core to work on it and then at the end merge all the frames back together.
In my opinion, the only way for gamers to seek better performance by getting new processors is in the focus of processing more instruction per cycle (IPC) than just shoving more cores at the problem as it shows that it doesn't really make a difference. If you can have better IPC, even a lower clocked processor can handle as many data as those who have lower IPC and a higher clock. This holds true for multicore as most consumers don't really need anything more than a quadcore processor until in a few years time when game and other consumer-oriented developers can get multithreading working as they should in theory. I believe that AMD and Intel should spend less time shoving more cores and start refining their architectures to pack more punch out of architectural design than just more clockspeed or just plainly more cores at the problem which they seemed to do for the past decade.
As of writing, no one needs anything more than a quadcore, no matter how hardcore a user you are. Maybe in 2 or 3 years but not right now.

Comments
2 comments postedGook points and comments I totaly agree with your idea of what AMD and Intel should do with architecture sprucing.
As it stands they can improve, HT NB and NB and SB chips. along with the other vitals.
Well, yes of course they should invest time into their intercommunication bridges, lanes and buses. I believe if AMD truly wants to be competitive they should use bi-directional links or at least 32-bit wide uni-directional links than the current 16-bit unidirectional links which in my opinion isn't good enough for their processors right now. Yes, it adds complexity but the trade-off sounds much better. Hopefully they can revamp the HyperTransport links for Bulldozer.
As for Intel's solution, I'm not sure what they're really up to since QPI is a closed system based off PCI Express like DMI. So I'm not sure if they'll make it faster for Sandy Bridge.
Post new comment