2800+ athlon MPs with 434MHz FSB

[TotalChaos]

Does anyone think it possible to unlock two 2800+ athlon MP proccessors, using say PC3500 HyperX memory, and increase the FSB from 266MHz to 434MHz while lowering the multiplyer WITHOUT the chipset giving out? Assuming that this IS possible and runs STABLE; would a system configured this way be able to 'crunch' DivX encoding faster then my OCed P4 (3.25GHz 868MHzFSB with PC3500 HyperX mem)? I know the 2800+ CPUs are clocked at 2GHz, but isn't this suppost to translate into into 2.8GHz P4 equivalency. Also, when using two physical CPUs, resources (IE FPU, cache, ect..) of one CPU aren't divided among each thread like they are with HT CPUs. Just a thought I had.....if you ever read any of my other posts, you'll notice I have no patients!!! :angry: :devil:

[b00zed]

A couple of things should be noted.

The first is that the rating system used for the Athlon MP/XP was designed around office benchmarks, and was introduced when the best P4s were based on the Willamette core (100MHz FSB, 256KB of L2 cache.) These days the P4 has twice the cache and twice the memory bandwidth (200MHz FSB, 512KB of L2 cache.) The P4 is still even with an equally rated MP/XP in office benchmarks, but video editing appears to be more MHz friendly so you will find a 2800+ lagging well behind a 2.8GHz P4C. http://www.anandtech.com/cpu/showdoc.html?i=1834&p=13

Secondly, the Athlon MP runs at a 133MHz FSB. To run it at 217MHz (433MHz DDR) would require a massive overclock of the CPU(s) and northbridge of the motherboard's chipset. While you can get desktop motherboards that are designed or intended for overclocking, I doubt you're going to find a server/workstation board that puts much emphasis into anything but stability, especially since 217MHz isn't much over the maximum 200MHz that you get on an Athlon XP, but it is well over the 133MHz that all Athlon MPs run at.

Patience is a virtue, grasshopper.

[ppera2]

CPU's would run on that FSB very likely, but I doubt that you will find dual MBO which will run stable on that FSB.
Other thing is that most of dual board will work with any XP CPU, so unlock isn't problem.
Third thing is that P4 has yet another advantage: SSE2.
Perhaps better go on Athlon 64 system and wait for 64-bit optimized encoding software...

[b00zed]

Getting it to run 217MHz unlocked might be possible, but then it becomes a "how long is a piece of string?" type question concerning the multiplier that you can use.

I've noticed that Anandtech has switched from using Xmpeg to gordian knot for their MPEG-4 testing, so it's just a matter of wait and see since the previous link I posted concerned Xmpeg. Their first article using gordian knot is here (http://www.anandtech.com/cpu/showdoc.html?i=1927&p=6), but it's comparing Durons and XPs to Celerons...

[TotalChaos]

We all agree that it comes down to optimization and perhaps some MHz here and there right?.

I read articles on hardware frequently. I came across an article from tom's hardware

http://www6.tomshardware.com/cpu/20030217/cpu_charts-26.html

If you read the entire article you'll see intel win a few bench marks in one area and AMD win then in others. That link is to a benchmark on encoding MPEG-2 video. In it the Athlon XP 1500+ (running at 1.33GHz) beats the Athlon 1.4GHz. It would seem more then MHZ and FSB are determining the out come. Perhaps core improvments or new instruction sets. What I realy want to know is, would investing in a dual 2800+ Athlon MP (uses barton core btw) system be faster at encoding DivX then my current OCed 3.0GHz P4. (3.25GHZ 868FSB 2-3-3-7-1 timings)

Am I just going to have to purchase a quad opteron [<--lol] setup or somthing to get faster encode times????

[b00zed]

The XP included new instruction sets as you suggested, specifically Intel's SSE instructions. It also beats a 2GHz P4A running RDRAM. I don't know much about mainconcept but from appearances the Athlon likes MPEG-2 a lot more than it does MPEG-4...

http://www20.tomshardware.com/motherboard/20020211/dualathlon-05.html#audiovideo_benchmarks... The Athlon MP is very highly scalable combined with the right motherboard, at least when xmpeg is involved (I don't want to speculate what it would be like with virtualdub or its derivatives...) The Opteron is supposedly even more scalable due to its internal memory controller, but since it's similar on clockspeed and a lot higher in price, for both CPUs and boards, then I'd say the Athlon MP is probably better value for this type of work, even if Opteron does support SSE-2.

[TotalChaos]

I've tried to isolate exactly what are the differences between SMT and SMP. A hyper threaded CPU uses SMT and a dual CPU system uses SMP. When using an intel CPU, with hyper threading (SMT), multi CPU aware OS, and a multi threaded application, you would think the application would recieve just as much proccessing power as it would when used in a multi CPU system (SMP). Only it doesn't! Although two threads can be proccessed at the same time in a hyper threaded CPU, each thread must SHARE the same CPU resources...IE: FPU, cache, ect. In a dual CPU setup there are two seperate 'sets' of CPU resources. Threads could then get full attention from the CPU it's currently passing through. Also being that there is TWICE the amount of cache MORE coding and data could be kept closer to the CPUs. With dual 2800+ Athlon MPs that would translate into 1MB of total cache. The only bottle neck with these CPUs is the fact that they run on a pitiful 266MHZ FSB and RAM. If the CPU were unlocked, the FSB raise, and the multiplyer lowered these CPUs SEEM like a VERY economical resource for compressing....well anything for that matter. I guess the only way to find out is to actualy do it right? I enjoy this knowledge please keep it coming! Does anyone have any other info about this that could help in this decision?

[b00zed]

IIRC "SMT" is IBM's buzzword for their multiple CPU cores on a single die/package, though I'd say it's very similar to Intel's hyperthreading ("HT"). HT as you mentioned is merely one physical processor behaving as two logical processors, and appears as two CPUs with half the resources. Essentially it was intended to make use of the wasted resources as not every FPU and ALU is being used at any one time in a CPU, so HT was intended to use the excess capability to do something else. Sometimes it works, sometimes it doesn't, but it's in no way a substitute for two dedicated processors, and also increases the cooling requirements for that CPU.

The FSB is definitely a bottleneck on the Athlon MP but not because it's "only 266MHz" but because of the fact that both CPUs must share this bandwidth, and the fact that the Athlon doesn't suffer from low FSB as much as the P4 due to its substantially shorter pipeline. This is what hurts SMP systems, and also what makes the Opteron so scalable, you can if you like build a motherboard upon which each CPU has its own dedicated RAM due to its internal controller. The fact that THG's SMP Athlon test above yielded a 1-2CPU scalability of 80% suggests that even this amount of bottlenecking doesn't hurt too much when DivX through xmpeg is concerned. It is worth noting however that when THG tested the 2GHz Celeron and overclocked it to 3GHz, it performed very well compared to a 3.06GHz P4 in xmpeg+divx, but when Anandtech recently tested a 1.6GHz Duron against a 2.6GHz Celeron with gordianknot+divx, the Duron was trailing by less than 5%... It just goes to show how much you have to consider the repeatability of testing and how hard it is to compare different applications. Personally I'd be very happy with a dual Athlon setup regardless of overclocking based on what (albeit little) I've seen... :D

With respect to the Opteron, you can get some boards on which only the first CPU can directly access the memory, so the second one has to go through the first one whenever it needs to access memory, which is not very efficient. If you were ever interested in an SMP Opteron setup you'd be best served by a board that has dedicated RAM for each CPU. This path is not a cheap one though...

[TotalChaos]

Simply put....what do you [boozed] and anyone else out there think is the absolutely FASTEST way to enocode DVDs into Divx. Of course keeping GMC and B-frames on, while in standard mode using DivX 5.1.1! What hardware would be best????

[Perplx]

Well the absolute fastest ecoding would use as many processors as possible. Since you want divx 5.1.1 your left with windows (linux is at 5.05 and mac is 5.1) the most processors windows can support is 64 with Windows 2003 Datacenter and Im guessing each processor is at least 1Ghz a piece. I'm not sure what the cost is but I bet its alot.

But realistically AMD64 vs P4 you'll have to wait till there is the 64bit Divx version to compare.

[TotalChaos]

virtualdub only has 4 threads though. Could it even use more then 2 I think 2 of those threads may be for display purposes.