LOOKING for a single core no HT CPU

[BOXXlabs]

I'd go with a Pentium Pro 200Mhz



I've found that multi tasking CPU's are over rated and weak in some 3D workstation production operations. So I'm going to build a box with the fastest non multitasking/ HT / multi core CPU still out there, any thoughts on a fast single tasking CPU still alive out there ?

TIA

[davegraham]

lol.

Cyrix, baby, Cyrix.

get an AMD single core 3200+. that'll hold you over really well.

dave

I'd go with a Pentium Pro 200Mhz

[Novakog]

I'd go with a Pentium Pro 200Mhz

Nice.

There are definitely way faster single-core processors than the 3200+, such as the 4000+! Seriously, the fastest single-core performance these days is probably still going to come from a dual-core (Core 2 Extreme probably). But if you are morally against more than one core, maybe an Opteron 156?

[davegraham]

actually Core 2 Solo's would be the fastest available in terms of FPU/ALU power. just have to find a board that supports it (there are a few). but Opteron 256s definitely would be power houses (ne'er mind the ECC/Reg RAM requirement, of course) and the fact that you can't use them in single processor boards. :)

dave

[Novakog]

Are Core 2 Solo's out yet? Err, and woops, thank you, I meant Opteron 156, edited my original post.

[davegraham]

Are Core 2 Solo's out yet? Err, and woops, thank you, I meant Opteron 156, edited my original post.

yup, they've been released for quite some time.

dave

[Tarrbot]

BOXXLABS states:I'd go with a Pentium Pro 200Mhz
Is this with the 1mb onboard cache or the 512kb? :)

[muzbee3d]

A note to :
Adam Glick
Engineer, R&D
BOXX Technologies


Well since you'all do build such great hardware how about a
race against us home grown guys!

If I or anyone else here has a system with a single class CPU as I asked help on in this thread that can beat your BEST BOXX in a 2 out of 3 win running three specific stock calculations within 3ds MAX 9 winner(s) get your machine!..... FREE !


SO Adam, will BOXX and their hardware accept this simple little challenge? http://forums.cgsociety.org/images/icons/icon7.gif

[imashination]

I've found that multi tasking CPU's are over rated and weak in some 3D workstation production operations. So I'm going to build a box with the fastest non multitasking/ HT / multi core CPU still out there, any thoughts on a fast single tasking CPU still alive out there ?

TIA

Maybe the best answer would be to ask why you think a single cpu system would be better, or what problems you have with multi cpu/core machines?

[davegraham]

A note to :
Adam Glick
Engineer, R&D
BOXX Technologies


Well since you'all do build such great hardware how about a
race against us home grown guys!

If I or anyone else here has a system with a single class CPU as I asked help on in this thread that can beat your BEST BOXX in a 2 out of 3 win running three specific stock calculations within 3ds MAX 9 winner(s) get your machine!..... FREE !


SO Adam, will BOXX and their hardware accept this simple little challenge? http://forums.cgsociety.org/images/icons/icon7.gif

won't happen.

dave

[lots]

You're serious?

The fastest single core performance these days comes from multi core parts.

[Novakog]

A note to :
Adam Glick
Engineer, R&D
BOXX Technologies


Well since you'all do build such great hardware how about a
race against us home grown guys!

If I or anyone else here has a system with a single class CPU as I asked help on in this thread that can beat your BEST BOXX in a 2 out of 3 win running three specific stock calculations within 3ds MAX 9 winner(s) get your machine!..... FREE !


SO Adam, will BOXX and their hardware accept this simple little challenge? http://forums.cgsociety.org/images/icons/icon7.gif

And what do they get if they win ;)?

Seriously though, having dual-cores won't slow you down on anything, it can only help. There are actually very rare rendering cases where using Intel's quad-core is actually worse because of the two seperates dies, so some memory intensive tasks can mess it up, but a dual-core will *never* be slower than the same processor single-core.

[davegraham]

And what do they get if they win ;)?

Seriously though, having dual-cores won't slow you down on anything, it can only help. There are actually very rare rendering cases where using Intel's quad-core is actually worse because of the two seperates dies, so some memory intensive tasks can mess it up, but a dual-core will *never* be slower than the same processor single-core.

our undying love and adoration?

dave

[BOXXlabs]

-never said we had the fastest machines...just the best.

[davegraham]

...There are actually very rare rendering cases where using Intel's quad-core is actually worse because of the two seperates dies, so some memory intensive tasks can mess it up...

just to correct a misconception:

the issue isn't the two separate die (dice) that the cores reside in. the issue is the usage of a single pipe to the memory controller hub (MCH). This affects the total memory bandwidth that feeds the cores.

cheers,

dave

[Novakog]

I don't really know that much about the hardware itself, but are you saying that one pipe isn't enough for 4 cores, or that one pipe isn't enough for two dies (not dice ;))? In our case, we actually disabled the cores in combinations, and using just cores 1 and 2 (or 3&4) was about twice as fast as using a single core, but using only cores 1 and 3 (or 1&4, 2&3, 2&4) was about twice as slow as any single core. Using all four cores was slightly slower than a single core. So the speed was actually dependent on whether the cores used were on the same die or not.

And can you point me to where I can find a Core 2 Solo? I could only find Core (1) Solos.

[davegraham]

I don't really know that much about the hardware itself, but are you saying that one pipe isn't enough for 4 cores, or that one pipe isn't enough for two dies (not dice ;))?

semantics on the die/dice thing. ;P

i'm saying that the efficiency of that design does not lend itself to handling high bandwidth memory streaming.


In our case, we actually disabled the cores in combinations, and using just cores 1 and 2 (or 3&4) was about twice as fast as using a single core, but using only cores 1 and 3 (or 1&4, 2&3, 2&4) was about twice as slow as any single core. Using all four cores was slightly slower than a single core. So the speed was actually dependent on whether the cores used were on the same die or not.

you actually just proved my point from above. what you did was increase the efficiency of the pipe by preventing the cores from requesting data through that memory channel. see my illustration here (http://forums.cgsociety.org/showpost.php?p=4510250&postcount=9)for an example.

when the quad cores were first released, Intel engineers themselves were recommending the 51xx series OVER the 53xx series precisely for this reason: memory allocation and usable bandwidth. This has little to nothing to do with core packaging in dies.

I'd be very interested to know how you enabled/disabled your cores outside of physically manipulating the processor itself.

And can you point me to where I can find a Core 2 Solo? I could only find Core (1) Solos.

sorry, core solos were 1st gen. core products. they utilize the same packaging and FPU/ALU units with one core physically disabled. thanks for the correction.

dave

[muzbee3d]

Clearly today the problem lies with the software, some critical 3d animation production functions still run on a single thread. It ticks me off to see 50% of the machine doing nothing ! This is a real waste of production man-hours, productivity and workstation investments waiting hours for a single calculation to finish. I’m just looking for a solution to off-load these jobs to with a faster and efficient output without paying extra for under performing hardware. I’m sure the day will eventually come when software will fully task the system 100% but in the mean time a I think a tweaked AMD FX 57 will likely match or out compute a dual quad core in these raw number crunching tasks.

[davegraham]

Clearly today the problem lies with the software, some critical 3d animation production functions still run on a single thread. It ticks me off to see 50% of the machine doing nothing ! This is a real waste of production man-hours, productivity and workstation investments waiting hours for a single calculation to finish. I’m just looking for a solution to off-load these jobs to with a faster and efficient output without paying extra for under performing hardware. I’m sure the day will eventually come when software will fully task the system 100% but in the mean time a I think a tweaked AMD FX 57 will likely match or out compute a dual quad core in these raw number crunching tasks.

Probably. :) man, I LOVED the single core FX series back in the day.

I'm currently running dual FX-70s at home right now along with some Opteron 2210s.

cheers,

Dave

[BOXXlabs]

This is a purely academic discussion IMHO - here's why:

Intel 51xx Xeon, 53xx Xeon and Core2 cpu's all share roughly the same memory bus ( front side bus) bandwidth (from the CPU socket to main RAM) - around 5000MB/s to 8000MB/s - or roughly 5GB/s to 8GB/s. (with a theoretical maximum bandwidth of 10.5GB/s.) This is real-world maximum memory bandwidth and is limited to any higher speeds mostly because of the fact that :

A - the memory controller is located on the motherboard (Northbridge chipset) and not on the CPU itself

and B - The intel design uses a single 64bit pipeline for each CPU socket (no matter how many cores it has)


On the other hand AMD CPU's, because they are designed with the memory controller built directly onto the CPU -and use the more efficient and faster Hypertransport memory/data bus platform, enjoy a real-world memory bandwidth from 8GB/s to 13GB/s
(with a roughly 25GB/s theoretical maximum memory bus speed)



But here's the bottom line - The fact that the intel chips are limited in terms of memory bandwidth does NOT SEEM TO HURT IN-APPLICATION PERFORMANCE. This seems mostly due to their use of a large 4MB L2 cache and "smart caching" routines in the microcode. This smart caching process "prefetches" often used (or likely to be used) data and keeps in at-the-ready directly on the CPU itself. WHen the application needs the data, it's right there ready to be utilized in the calculation by the CPU.


While the next generation intel chips will enjoy on-chip memory controllers, this issue should really be a non-issue when you look at overall in-application performance...

As for the comments on disabling various cores, I believe this has more to do with how much/how efficiently the application has made use of multithreading. If an application is not well multithreaded, you can indeed see better performance by turning off the other cores... I doub't the performance issues the earlier poster was seeing is due to memory bandwidth problems...


my $.02

Adam
BOXXlabs

[Novakog]

Well I am dragging this thread off topic ;) - Dave, I will pm you.

[davegraham]

Well I am dragging this thread off topic ;) - Dave, I will pm you.

replied. :)

dave

[davijin]

Clearly today the problem lies with the software, some critical 3d animation production functions still run on a single thread. It ticks me off to see 50% of the machine doing nothing ! This is a real waste of production man-hours, productivity and workstation investments waiting hours for a single calculation to finish. I’m just looking for a solution to off-load these jobs to with a faster and efficient output without paying extra for under performing hardware. I’m sure the day will eventually come when software will fully task the system 100% but in the mean time a I think a tweaked AMD FX 57 will likely match or out compute a dual quad core in these raw number crunching tasks.

Yeah the thing is it won't, think about it. in a dual core system one core can focus on computing a task while the other can run the system, background programs etc. while a single core system is going to be doing your task and smaller tasks inbetween slowing the main computation down. plus money wize its not like you are spending a drastic amount more for dual-core vs single core. plus you might be screwing youself in the future because since dual/multi core cpus dominate the market more and more programs are going to be coded to take advantage of multi cpus.

[TestType]

Clearly today the problem lies with the software, some critical 3d animation production functions still run on a single thread. It ticks me off to see 50% of the machine doing nothing ! This is a real waste of production man-hours, productivity and workstation investments waiting hours for a single calculation to finish. I’m just looking for a solution to off-load these jobs to with a faster and efficient output without paying extra for under performing hardware. I’m sure the day will eventually come when software will fully task the system 100% but in the mean time a I think a tweaked AMD FX 57 will likely match or out compute a dual quad core in these raw number crunching tasks.

You do realize that when the CPU monitor shows the machine is running at 50% capacity it means that one of the two cores is running at 100% capacity. If it was 100% then it would mean both cores are running on full load. If you had an identical processor, just a single core, the CPU monitor would tell you it's at 100%, but it would give you the exact same result as a 50% dual-core.

Of course more software will be written to utilze multi-core processors, but right now they're mostly good for multi-tasking save for a few apps. But don't think you're wasting money though if you claim to never multi-task, single cores are still beat by a dual-core running only one core at full capacity.

[Tarrbot]

muzbee3d, I'm not sure of your technical background so please read this. (http://hubpages.com/hub/Single__Dual_or_Quad_Core_Which_Is_Best_For_You)

Let's assume you're running the one single-threaded app at 100% on a single core processor. It's doing the best it can do. All you can do is put a faster single core processor in there to increase it's effectiveness at single-threaded applications...

... or

you can put a dual/quad core processor in (or a dual or quad processor board with procs in it) and now you're getting the single-threaded app taxing one of the cores (or procs) at 100% while the other core(s) are sitting relatively idle.

What can you do during that idle time? You can open up Photoshop and tweak textures or a modeling app and futz with a model. Or even pop in a game and play the game while you wait.

Multiple cores/procs bring in better system response time and allow you to do other things while the single-threaded app does its own thing hogging up all of the cycles on one of the cores.

Also, as has been noted, your system will likely have the mundane operations pushed off onto one of the other cores and allow your app to use fully 100% of the core instead of 80-90% of the core in a single-core, single-processor system.

[muzbee3d]

http://forums.cgsociety.org/images/icons/icon13.gif Good input from all you guys ! Thanks for your help.......

With this DATA chart I rest my case. Cheers.

http://en.wikipedia.org/wiki/Million_instructions_per_second

[Tarrbot]

muzbee3d, nothing in that link actually tells us anything about your case.

MIPS/FLOPS are fairly meaningless numbers unless you work in a lab running calculations on processors all day. If anything, this actually hurts your case by showing that higher MIPS and MIPS/MHz scores are on dual-core processors not single core processors.

So, what exactly is your case? Because I'm not following at all here other than you're pointing at a non-relative chart with a bunch of semi-meaningless numbers on it.

[davegraham]

yeah, that article is kind of meaningless. to gauge efficiency and relative "power" of a processor, we typically look at IPC (instructions per clock) as one of several metrics. anyhow, i think this thread is dead.

dave

[imashination]

With this DATA chart I rest my case. Cheers.
http://en.wikipedia.org/wiki/Million_instructions_per_second

Do you buy cars based on the rpm of the engine?
Do you buy a bathtub based on how fast the water drains?
Do you buy buy house plants for the number of leaves?

Youve picked an utterly meaningless number which has no actual basis on how fast it will go. From the very page that you linked to:

"Critics of the term refer to it by backronyms such as "Meaningless Indication of Processor Speed" or "Meaningless Information on Performance for Salespeople" or "Meaningless Integer Performance Spec"."

[BOXXlabs]

agreed.

...this said, we do use Sisoft SANDRA 2007 as a normal part of our internal benchmarking process.

Sandra 2007 does have a standard CPU performance test that measures raw integer arithmetic (Dhrystone) performance in "MIPS" and floating point arithmetic (Whetstone) performance in MFLOPS.

It is a good (and accurate) RELATIVE BENCHMARK test -meaning that if you run the same SANDRA CPU test on multiple machines, you will get accurate, repeatable performance metrics which are directly related to the system's processing speed RELATIVE to another system. These numbers can be compared against each other to attain meaninful performance data.


Adam
BOXXlabs

[davegraham]

there are definitely more "accurate" cpu benchmarks out there with the key reproducability indicators needed to make it a reliable benchmarking standard. But, can't beat Sandra 2007 for what it is: a neatly integrated benchmarking suite.

cheers,

Dave

[BOXXlabs]

I agree, Dave.

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.