No matter what CPU you use you absolutely have to have a high end performance motherboard.

I only have a Q9650 processor, but I believe my excellent performance on AVCHD is partially due to the MB and the MB bios being and memory management being carefully tweaked.

Also, I wouldn't spend time or money on the "economy" processors. You will only be back here later complaining how hard AVCHD is to edit.

Again in this case, we're talking about an "economy" processor that offer better video performance than a Q9650

Reply by: warriorking
Date: 9/11/2009 6:19:06 AM

Being one who moved from a Quadcore Q9550 to a i7Core 920 I saw a very noticable improvement in Video rendering power, I knocked over an hour off my AVCHD rendering times in most cases, even more in others....

Looking at the test results, 920, bang for the buck, is still a very good processor, being that it's constantly on the heals of the 870 in the bench tests and almost half the price. I'm in the process of upgrading my P4HT machine and am still leaning towards the 920 unless the price of the 870 drops considerably before the end of the year.

The 860 is faster than the 920 across the board at about the same price. Plus the 860 compatible motherboard is less expensive than the 920 board making the 860 with mobo a faster and less expensive solution.

- Mark

The question for me would be what can each do overclocked - the 920 can go to over 3.6ghz easily.

Hyperthreading is a mixed bag... its value depends on the application. Anyone with an i7 should post some results.

Hyperthreading was known, prior to Intel's implementation, as hardware multithreading. In a normal SMP system, each CPU is a completely separate CPU, usually sharing memory resources and at least one cache, but with their own private MMU, L1 cache, etc.

In a hyperthreaded CPU, there are basically just a duplicated set of registers, and a means of flipping that register context. When one thread hits a pipeline stall (maybe the CPU is waiting on memory or I/O or something), the other thread comes into context, and can hopefully get some work done while that first thread is waiting. The idea is that you can approach 100% use of the CPU resources... which may be 5%, 10%, or even 25% more than you might with just a single thread.

The use of the word "thread" is correct, here, too.. yo can't schedule different programs on the same core this way... there's still only the one MMU context. So, unlikely regular SMP that helps as long as you have an OS and other programs running, this only improves the performance of multithreaded applications. Vegas, for example.

There are times when it actually makes things worse, however. One easy thing to observe is the first killer... with two threads on the same CPU, you're filling the L1 cache with twice as much data, and potentially even twice as many instructions, as you would be with a single thread. This alone can actually mean you're falling off the L1 cache (thrashing it) enough to make performance worse than with just the one thread. This is why Intel's always had an "off switch" for multithreading.... er, hyperthreading.

While multithreaded preview would be great, save that for layering and other stuff that's more inherently CPU intensive.

Most middle-to-high-end GPUs already have video acceleration "in hardware" (eg, running on otherwise unused stream processors), and there's been a good video acceleration API in Windows since Vista shipped (the first one, DVXA, actually came out with Win2K, but DVXA 2.0 is reportedly much better, and that's the one that's new with Vista).

Either way, it's damn shame that only some 10%-15% of the computational horsepower available is being used on the most human-intensive part of the process. Faster output rendering is great, but I can usually do other things (sleep, etc) while that's happening. Interactive editing is real "Dave Time"... making that as fast as possible should be a primary goal of any NLE.

Posted by: srode
Date: 9/13/2009 12:28:39 AM

The question for me would be what can each do overclocked - the 920 can go to over 3.6ghz easily.

Short answer:
yes, the 860 can be overclocked to 4Ghz but you need a heavy duty heat sink.

Long answer:
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3634&p=18

but, as explained in the anandtec article, a significant increase in voltage to the entire CPU is required for any serious OC. How long would a CPU last under such conditions? I remember one OC article that got fantastic speeds with a good heat sink, but concluded with the suggestion that a CPU under those hi-volt conditions should last at least 3 or 4 months. huh!?

When I OC'ed my Q6600 with G0 stepping, I put on a pretty good heat sink to keep max temp. well below 60 degrees, but hardly boosted CPU voltage. That was almost 2 years ago and it's still going strong, despite a rise in temp. that has required taming it down a bit (yes, I did finally blow out the fins on the cooler with compressed air, but that didn't help; it's the film on the fins, I think, not the removable lint that's the culprit). Even with a slight decline in top speed, I've been quite satisfied with its stability and performance. Could I say the same two years from now with an 860 if I up the voltage?

Again in this case, we're talking about an "economy" processor that offer better video performance than a Q9650

Buster:

The $196 price of the i5 750 does indeed make this an economy processor. The performance of the i5 750 vs the Q9650 shows the performance to be essentially identical with no OC, it does not have any meaningful level of increased performance over the Q9650.



http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3634

Actually, the Anand video benchmarks show the i5 are a few seconds faster than the 9650, which is about 5% to 11% increase in performance.

Just for reference, my now 1.5 years old Qx 9650 clocked @ 3,8GHz runs the NewRendertest (in this forum) in 71 seconds. Another user running an i7 940 (@2.93 Ghz) with same OS (Vista64) and VP (9.0b) runs the same test in 65 seconds. That's about an 10% improvement at standard clocking. Not bad at all. Qx9650 is still going strong, but partly because of its overclocking potential.

Christian

"yo can't schedule different programs on the same core this way"Actually, you can schedule H.T. on today's cores. ;-) As earlier H.T. was a mixed bag, today's engineered threads are performing better. There are tasks that can not be broken up into parts, these are truly single core tasks, meaning a 4core cpu will perform better running them on one its cores than across a couple. There are "conditions" that permit threads to run on many cores, but currently few are designed and tested to avoid performance hits. Performance hits? most current cores are better design today and have copies of internal core assets that can be share with H.T. What hurts this design by Intel is the L1 cache for the instruction queue. The L1 cache for the core is used by both instruction(L1) and data(L1) streams, primary and secondary(hyperthread). The L1 cache should be split to effective management to different address ranges for the shared core assets. Some encoders use this weak point to duplicate the encoding instructions with different data sets with small address ranges..... this works well. But tasks/threads that are running in own address space can take major performance hits because L1 goes to L2 then L3 then off chip. The L2 to L3 is what usually forces the core to go the next thread....... this is very simplified but hints why not all tasks perform well.

Overclocking.. the real answer is "you can't know, you can only try it".

Increasing the voltage alone is always a bad idea.. this will lead to additional heat and additional thermal stress (the mechanical stress on the chip as it repeated heats and cools). That will affect the life of the CPU.. but if you improve the cooling, you may eliminate this problem.

Another factor is electromigration (http://en.wikipedia.org/wiki/Electromigration), or technically, faster electromigration. Increasing the voltage will cause the chip to age faster, essentially, even if you don't overheat it. Is this a real problem? You don't really know. Obviously, if your voltage boost changes the CPU's life from 20 years to 10 years, you probably don't care.. who keeps a PC for ten years (well, ok, my Mom...). But if it changes it from 20 years to 1 year.. you may have a problem.

The more practical problem with overclocking is reliability.. does it all really work at the extended clock speed? Will memory be exactly as reliable, or will it fault 100x as often, though still rarely. Or only on hot August days? Does the CPU really work at those speeds, or will it occasionally make a mistake.

There's plenty of reason to believe some overclocking is not a problem at all... if you're lucky. The speed of the CPU you buy is based on binning.. the manufacturer runs performance tests on a chip. If it passes the 3.2GHz test, it goes into that bin, otherwise maybe it goes to the 3GHz or 2.8GHz test. However, once they have enough 3.2GHz chips, they probably don't run any more through the 3.2GHz test.. they all go to the 3.0 GHz test. So some of the 3.0GHz or even 2.8GHz chips may have been fine at 3.2GHz, under all of the test parameters.

This also translates to GPUs, DRAM, and other chips. If the speed you're pushing for is very expensive and rare, there's a pretty good chance that every chip is tested at that speed, and all of the lower-grade parts are high-speed dropouts... you might hit 3.2GHz at a voltage higher than used for the normal 3.2GHz chip, this might well contribute to a short life.

Another way of looking at this... the high-end Intel CPU is usually around $1000 when new. In 2-3 years, that same chip will be $200-$300 or so. If you buy a $200 chip today and get a year or two of $1000 performance out of it, you're still well ahead of the game, even if it fails in a year. On the other hand, if it's flaky enough to cause an increase in BSODs and other errors, it's not worth it.

For the record, I don't overclock things.

apit34356... incorrect. Each process in a system (usually corresponding to "program", though of course, services and other entities in the system are also processes) has its own MMU context.

A thread is an independent simultaneous path of execution within a single process (and MMU context). By definition, every multi-threaded application starts up multiple threads... some of this more recently has been to take advantage of SMP systems. But it's been a fairly good way to design most non-trivial programs for decades. Some small bit of this kind of programming has gone on in OSs without threads even longer (like original UNIX), but the overhead of communications between full processes (through files or pipes, than thus, the kernel) makes this less useful than between threads (shared memory locations protected by semaphores).

Similarly, each physical CPU core has a single MMU context at any given time... both "hyperthreads" with a core are limited to that same MMU context -- thus, the same program.

Intel describes this further here: http://www.intel.com/technology/platform-technology/hyper-threading/index.htm?iid=support.

Of course, multiple threads in the same program ran run on as many cores as the code and the scheduler allow.. that's not even remotely a problem. And even programmers who didn't learn programming on AmigaOS or BeOS are starting to use more threads... that's also a good thing, and it's useful on any SMP system.

The hyperthreading limit is the other way around... while you can easily duplicate an MMU context on multiple cores, you only get one MMU context per core. Both threads on that core have the same one... that's why Intel called this "hyperthreading". Sure, you see eight cores under WIndows, but the scheduler knows half of these are "virtual", and only available for multithreaded applications.

These new processor chips are lower performance cheaper versions of the current i7 chips as I understand it. How does that equate to faster AVCHD editing, surely the i7-975 would be the go if one needed to crunch AVCHD as smoothly as possible

I agree on that.

One upside to the new chips is they may make the 975 price come down to the $575 price point they usually hit at some point. Or maybe a higher performance than 975 chip will be released soon.

I edit AVCHD footage off a HMC-150s for weddings and events. I am having good results on my old Q9650 for the most part. However, a 975 would be a nice improvement. I certainly don't want to spend a lot of time on a system that is not the current top of the line.

Yeah,

Overclocking is a double-egded sword. You should not do it if you don't know what you are doing. Or if you still do it, then you take a huge risk.

I never thought of overclocking before I bought a good quality mobo (from Asus) that itself had lots of headroom for overclocking. I would NEVER compromize stability for speed. You don't have to do that, necessarily, if you do the overclocking following some golden rules.

I checked the maximum allowed core voltage for the Qx9650 and NEVER exceeded it. That is important, the electromigration IS a real problem and must be taken seriously. Next I got the best possible CPU (forced) air cooler money could buy, and used premium heat transfer compound. I'm not a liquid cooling believer, those systems might be efficient, but not very reliable in the long run.

My DDR3 memory modules are also equipped with good heat spreaders. Next I spent couple of evenings testing and pushing the system to its limits (still without exceeding any max voltages). I was able to overclock the CPU up to 4.05GHz! Did some stability tests at this frequency and the system ran stable for one hour. However. it had a starting problem at very low temperatures. Yes - I put my PC overnight in my garage (it was winter) to cool it down to about 10 deg C. An overclocked system must be tested also at the lower extreme temps. I also monitored carefully all temperatures (both CPU, RAM and NB chips), even lended an IR-meter for a reliable check. Every temperature was within safe limits running the system at it highest utilisation in a continuos loop.

Backed down finally to 3.8HGz, to prepare some headroom for the oveclocking to make the system really robust. Backing down from your maximum stable frequency is a must. At least 5 to10% to guarantee a stable system over a longer time period.

I have been running smoothly since then for soon 2 years! The system is rock solid. The CPU internal cores never exceed 71 deg C under heaviest loading or highest ambient temps. I know that I might not anymore experience full 20 years of endurance, but who cares. It might be somewhat shortened, but have my doubts, since I am still running within dataheet specs!

This process took some time and experimenting, but the leap from 3 to 3,8 GHz with a similar 27% speed increase - was worth it!

And you need SOME luck to get a CPU that really has the overclocking potential. Not every CPU can be pushed this far.

Christian

I just received flyer in the mail from MicroCenter.

Boxed Intel Ci7 920 - Socket 1366 - $199.99. It is not listed on the web site, it is in the printed flyer.
Newegg.com (which usually has best prices) has it for $279.99.

I guess I will not be waiting for Christmas break to build my new i7 based PC.