Can Ca Handle Dual Processors?


servoos
 Share

Recommended Posts

My PC is far from new but has six cores in its CPU. Chief Architect Software Engineers and Architects have several times stated that "the more cores, the better". Chief Architect is not, I believe "hyper-threaded" but does benefit from the presents of multiple core CPU's, SSD.s and a well made, up to date Video card and RAM.

The only usual limiting factor is one's budget, the more expensive machines give greater and greater performance but as it increases, the difference is measurable only by sicentific testing devices.

Get the best you can afford and then be happy.

 

DJP

Link to comment
Share on other sites

So the 2670 has 12 cores. and the 2697 has 14. The turbo speed is also different. It is hard to judge what would be best without doing some benchmarking. But 24 cores vs 14 should give you more throughput for ray tracing assuming the memory can keep up. Either option would be the envy of most users of Chief though.

Link to comment
Share on other sites

  • 2 months later...

I know I'll probably receive a lot of flak for this but here goes!!! I think there are some misconceptions concerning multi-core processors and their benefits. First, a single core processor running at the same clock rate as a multi-core processor has the same throughput. The additional cores have no real impact in regards to shear input/output processing power. In theory the single core processor should actually beat the multi-core as it does not have to spend time managing the cores. The real benefit of multi versus single is their ability to more efficiently handle multi-tasking demands. These days we are usually running several programs concurrently, Chief, MS Office and our browser. It is under this scenario that the multi-cores have an advantage. Once an instruction set enters the processor it can be directed to one or more of the cores, It's as if each program has a dedicated piece of the processor. However this should not be construed in the same way as having parallel processors, it's a hybrid approach. The cores are within the processor, they in themselves are not fully independent processors. No matter how many cores are available instructions can only move into and out of the processor at the clock cycle (frequency) rate.

 

Here's an overly simplified analogy. Envision two 5 mile stretches of highway, each has a single lane on ramp at the beginning and a single lane off ramp at the end. Between the on/off ramps one of the highways has one lane while the other splits into 4 lanes. In both cases the on and off ramps determine the rate that vehicles can enter and exit the highway, however once on the highway the four lane one will be less congested. Once the on or off ramp reaches it's maximum flow rate it does not matter how many lanes you have in between.

 

Graham

 

 

 

Link to comment
Share on other sites

The parallel data access to memory is certainly an issue with multiple processors and cores.

 

In our current batch of multicore hardware memory speed is a key bottleneck. Which actually means that in some cases one core in a multicore machine may be faster than a single core machine for the simple reason that they need to put faster memory and wider data channels into the hardware to support more cores.

 

In addition there are on CPU caches that allow data to remain in the CPU for the duration of a task. You will note that new processors are sporting rather large caches these days.

 

Finally, there are some process that require a lot of computaion with relatively little supplied data that make ideal candidates for parallel processing. Ray tracing is one of these.

 

There are a lot of cases where there are N^2 computations for data of size N.

 

So while to a large extent what you say is true, more cores can be faster. However, software needs to be made to support the cores in order for it to take advantage of the benefit.

 

In Chief we do support more cores in a lot of cases, although not all that theoretically can take advantage of more cores. Ray tracing as I mentioned is one. We also do this in a few other computational bottlenecks such as generating lines when we send a 3D view to layout.

 

You can easily test the benefit of this in Chief by turning off the "Optimize for Multi-Core CPUs" option in preferences and doing timing comparaisons of certain operations. While it will never be N times faster where N is number of cores, we often get N - approximately 1 or 2 for a few cases. Others are not as good but if you can get a 2X improvement on a 6 core processor, it is still worth doing.

 

This is especially important in today's computer environment where processor speed improvements have slowed over time. Eventually they will still need to improve if we are going to make things faster because not everything can be done in parallel. But a lot can and we will certainly continue to improve performance where we can.

 

What you have described is conceptially the big technical problem with parallel processing that has taken a long time for computer manufacturers to deal with. Some of the early hightly parallel super computer research done 20 years ago or so has made possible the hardware we have today that actually does to a limited extent overcome the bandwidth problems that you describe.

 

 

No worries though. Your analysis is a valid concern that still limits overall computational speed.

Link to comment
Share on other sites

Thanks Doug, really appreciate your expansion on this subject. I fully agree that multi-cores are for most users the best available solution. For all of us with the X7 upgrade and multi-cores we can easily demonstrate this by setting the Raytrace core usage to max. and then try running another program at the same time as a Raytrace, can get real laggy. Great way to simulate one core versus multi-core under multi-tasking situations.

 

When I upgraded this was immediately evident as X7 is defaulted to use maximum cores. Initially I thought some virus or background process was the culprit. Ran a few process monitoring programs that indicate Raytrace was consuming 100% of my processor all the time. As soon as I saw the setting in "Preferences" "Render" "Raytrace" I knew exactly what was happening. Freed up one core and everything is back to normal.

 

I know from this forum that many users are experiencing this, maybe a general notification is needed to help resolve this and alleviate a lot of undue stress. Especially for those who less familiar with this subject.

 

Graham

Link to comment
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
 Share