TECH Now that AMD is releasing their new chip in a few months..

[SeeVinceRun]

What are your thoughts? I'm having a hard time believing it will be any more powerful than sandybridge. Id love to see AMD back on top, but the realist in me says its not going to happen.

Especially when AMD is saying the new 8 cores are "50% faster" than the current gen. Phenoms. This seems like they are referring to multithreaded apps, not IPC.

What is OTs take on this?

 

[deusexaethera]

AMD is smaller and more agile. Intel is whipping its ass right now because they have more production capacity and their chipset offerings are better-integrated with their CPU offerings, not because their products are inherently better. Everyone seems to have forgotten how AMD was kicking Intel's ass for several years with the Athlon XP and Athlon 64 CPUs, while Intel was scrambling to change direction very, very slowly.

My point is, AMD helps drive the market forward because they're Intel's only real competition, and occasionally they come up with a really great idea that changes the game. Don't write them off just because Intel is moving more product right now.

 

[SeeVinceRun]

AMD is smaller and more agile. Intel is whipping its ass right now because they have more production capacity and their chipset offerings are better-integrated with their CPU offerings, not because their products are inherently better. Everyone seems to have forgotten how AMD was kicking Intel's ass for several years with the Athlon XP and Athlon 64 CPUs, while Intel was scrambling to change direction very, very slowly.

My point is, AMD helps drive the market forward because they're Intel's only real competition, and occasionally they come up with a really great idea that changes the game. Don't write them off just because Intel is moving more product right now.

I really hope they do come out on top in this next generation chip war. I always seem to find myself rooting for the underdog. In fact, I think all my CPUs in the house right now are AMD.

I remember the Athlon days where a 1.8ghz XP chip was killing a 2.8ghz HT P4. I'm just worried that AMD is getting too caught up in massively multicore CPUs (of course, not a bad thing) and forgetting that IPC matters as well.

But all things said, I like the design of the bulldozer modules, its very forward thinking and COULD pay off greatly. But, the design of the first Phenom quad core was also very elegant and look how that turned out.

 

[deusexaethera]

IPC really doesn't matter for the vast majority of processes running on any given machine. I think the ideal setup would be eight cores: six "cheap" cores running at low speed, and two "expensive" cores running at high speed. But for that to work, the decades-old argument about what constitutes a "heavy" process vs. a "light" process would have to be resolved, and that will probably never happen.

 

[CodeX]

By IPC you are referring to instructions per cycle and not inter-process communication... right?

Instructions per cycle is of course important, but generally no more or less important than clock speed or number of cores... they all factor together to give you the general performance of the processor. What I mean is, if you double the clock speed but half the IPC's it is exactly equivalent to doubling the IPC's but halving the clock speed... neither is more important than the other, and which one is preferred often has to do with engineering constraints more than anything else.

You should not underestimate the importance of SMP though... most users only use their CPU for somewhere between 0 and 1% of the time they are using the computer, and a good portion of that is in context switching. The more cores you have the less context switching must be done and the more responsive each running application appears to be to the end user.

I'd rather have 10 cores running at 500mhz than 1 running at 5000mhz in terms of average user experience... obviously if you do non-standard things like video compression or 3D rendering this would not be ideal though...

 

[deusexaethera]

2 cores at half the speed is not the same as 1 core at twice the speed. The single faster core has to swap threads more often, and it takes a fixed number of processor cycles to do that, so the operational overhead is lower for a single-core processor. Also, the single faster core requires faster data storage to keep it supplied with instructions to run, otherwise it will sit idle while waiting for data to transfer. Single-core processors are less efficient overall than multi-core processors, hence why nobody bothers with single-core processors anymore.

However, what you say about non-standard tasks vs. standard tasks is why I think the best arrangement would be a few slow cores plus a couple fast cores.

 

[SeeVinceRun]

Wouldn't an ideal CPU not be 6 cheap cores/2 expensive cores, but something along the lines of GPU for heavy FP calc and x86 cores for everything else? Excuse me if im way off base there.

I thought this was where the APU was going, but now it just seems to be there to just get rid of onboard video and nothing else.

 

[CodeX]

2 cores at half the speed is not the same as 1 core at twice the speed.

I didn't say that, I said that 1 core at 2x IPC is roughly equivalent to 2 cores each with 1x IPC... you're right there is overhead to switch contexts, which I also mentioned in that post.