PDA

View Full Version : Dual core questions


adrencg
08-15-2005, 07:56 PM
I've posted recently along these lines, but I went to Siggraph and have since built my own Dualcore AMD machine. I found that Lightwave 8.3 takes advantage of a dualcore chip in every way...workflow was my main concern. In addition to rendering being twice as fast, I'm finding that moving polys around the scene is much, much quicker.

There was a BOXX machine at Siggraph which had a quad dualcore machine running Lightwave, and while it was amazing to see, I had to wonder later as to whether or not all 4(actually 8) threads were being utilized by LW during general scene manipulation. I know that during rendering, all 8 threads were working.

Can LW 8.3 only see "one dualcore processor", or can it see as many as you've got onboard when it comes to workflow? Maybe a dumb question, but we're buying all new hardware in my 3d department. If we can get a dual processor board with 2 dualcore cpu's, and have LW take advantage of all of the power all the time, then it's an option we'll probably go with.

Anyone know what I'm talking about here?

Mike

thx1138
08-15-2005, 09:05 PM
Using multiprocessor systems (SMP) for Lightwave will give you the most improvement when rendering. That's when every little drop of calculation power is sqeezed our of your system. The biggest improvement is going from 1 to 2 cpu's. Going from 2 to 4 is less dramatic and going beyond 4 even less. Even more if you look at the costs for Quad or 8-way systems. They're just not worth it.

Dualcore sits somewhere between Hyperthreading and SMP. It's performs better in multithreaded situations but doesn't come with the cost of SMP (motherboard etc). But there are some drawbacks. The L2 cache is shared and therefore only half of what you get in SMP systems. And cache is where the CPU really kicks into lightspeed no matter how fast your DDR2 and FSB are. This is also why the Xeon's have always been faster (and hotter) then normal P4's since their cache runs at the same clockspeed as the CPU memory bus. On a P4 and AMD the cache runs slower (means less heat buildup).

Lightwave will see as much CPU's you can throw at it, as long as the hardware and your OS sees them. Remember Windows XP only supports 2 CPU's. You'll need 2003 Server for 4 CPU's. Not sure how XP will see a dual-dualCore system. As 2 or as 4 ?
If it sees it at 4 and only use 2, your better off getting a single-Dualcore system I guess.

As for the workflow. Anything above 2 CPU's (or cores) won't make much of a difference. Not if you look at the price. With dual-core's this will drop, but it still won't give you miracles.
Also dualCore is new stuff. Best to wait a while until it becomes mainstream. If you need something now. Going with regular SMP Xeon dual-based systems will give the best price/performance.

Just my 2 cents.

MadMax
08-16-2005, 04:41 AM
Dualcore sits somewhere between Hyperthreading and SMP. It's performs better in multithreaded situations but doesn't come with the cost of SMP (motherboard etc). But there are some drawbacks. The L2 cache is shared and therefore only half of what you get in SMP systems. And cache is where the CPU really kicks into lightspeed no matter how fast your DDR2 and FSB are. This is also why the Xeon's have always been faster (and hotter) then normal P4's since their cache runs at the same clockspeed as the CPU memory bus. On a P4 and AMD the cache runs slower (means less heat buildup).


I'd say you need to do your research better as virtually nothing of what you said is correct.

first off, L2 cache on dual core AMD processors is NOT shared. they come in either 512k or 1mb per core. So it is not half of what you get in SMP based systems.

Also, cache on AMD's are not slower as you suggest, it runs at clock speed.


Lightwave will see as much CPU's you can throw at it, as long as the hardware and your OS sees them. Remember Windows XP only supports 2 CPU's. You'll need 2003 Server for 4 CPU's. Not sure how XP will see a dual-dualCore system. As 2 or as 4 ?


XP Pro will see a dual dual core just fine. Task manager shows 4 cores. It uses all four cores without issue. Server 2003 Enterprise was only necessary with the box Quad dual core.


Also dualCore is new stuff. Best to wait a while until it becomes mainstream. If you need something now. Going with regular SMP Xeon dual-based systems will give the best price/performance.


Sorry, not even close to being accurate. wait for it to become mainstream? I guess if there were any real reason for a reply like this I might agree. Fact is, that is just bad advice. A dual Xeon abolutely does NOT provide a better price/performance.

Niklas Collin
08-16-2005, 08:00 AM
Indeed, dual core systems are the way to go right now. When it comes to the price/performance -ratio I'd say dual cores beat dual xeons hands down. Did a quick search and found out that you can get 3.2 GHz xeon (single core, single processor) with about the same price as dual core 4400+. With the fact that the dual core AMD is also 64-bit processor and actually even the single core will beat the xeon in most cases then it would be nice to see how xeon can be a better solution? And when you take into account the fact that dual processor motherboards cost a LOT more than single processor motherboards I don't really see why xeons would be a good solution.

I'd say that MadMax pretty much summed it up. Dual core systems don't have shared L2 cache, it runs at clock speed. Actually in many tests Dual core processors have been FASTER (by a very small margin yes, but still) than normal dual processor systems with equal hardware in other ways. No way in hell I'd buy a dual xeon system right now, it just doesn't give enough bang for the buck.

MadMax
08-16-2005, 08:31 AM
FYI, take note of the specs in my sig.

That is a dual Workstation board with a pair of dual core CPU's.

thx1138
08-16-2005, 09:36 AM
I'd say you need to do your research better as virtually nothing of what you said is correct. first off, L2 cache on dual core AMD processors is NOT shared. they come in either 512k or 1mb per core. So it is not half of what you get in SMP based systems.
Also, cache on AMD's are not slower as you suggest, it runs at clock speed.


Guess I didn't know AMD was ahead of Intel on the cache side, although I wouldn't classify that as 'virtually nothing'. Thanks for the update.

Sorry, not even close to being accurate. wait for it to become mainstream? I guess if there were any real reason for a reply like this I might agree. Fact is, that is just bad advice. A dual Xeon abolutely does NOT provide a better price/performance.

So you're saying DualCore perfoms the same or better as SMP ? And with SMP being more expensive in general that would make Dualcore the better choice ? Now I'm having doubts about accuracy or I'm not sure what you mean.

MadMax
08-16-2005, 10:28 AM
So you're saying DualCore perfoms the same or better as SMP ? And with SMP being more expensive in general that would make Dualcore the better choice ? Now I'm having doubts about accuracy or I'm not sure what you mean.


That is exactly what i am saying.

With a choice between a dual core 4400+ or a dual CPU Opteron 248, I'd take the dual core setup. It's going to be faster.

Add to that a lot cheaper. Dual CPU motherboards are 279.00 on the low end, up to 500+ on the high end. I can buy a solid single CPU board loaded with features for 120.00 - 175.00

registered ram for the SMP is more expensive, even if only slightly. Also the extra error checking makes it just a touch slower than standard DDR ram.

a 4400+ costs around 550.00, an Opteron 248 for example costs 345.00 each, making for almost 700.00 for an SMP system.

Now regarding dual core architecture, the cores are tightly integrated. They communicate directly on the SRQ (System request Queue) Compared to Intel with zero core integration, and their dual cores need to send requests out to a slow FSB, then back into the CPU to be recognized.

The integrated memory controller and major northbridge functions integrated into the CPU provide for exceedingly fast and efficient performance.

If you look at just a couple of basics from my comparison above, I am going to save several hundred dollars by buying a dual core system over SMP.

And this doesn't even address the issue of 64 bit. Remember, LW 8.5 and 9 are in the mix here as 64 bit apps. As I understand it, the 8.5 64 bit version is going to release at the same time as the 32 bit version. If Siggraph was any indication, that is going to be very soon now. I won't hazard a guess on 9.0, but I bet it won't be too far off.

lots
08-16-2005, 12:23 PM
You dont need MadMax to tell you a dual core will be faster than two single cores in dual operation. Though he is right ;)

Take a look at a recent article put up by anandtech. They compare the dual core X2s to the single core dual opteron work stations, and the X2s win out in almost all cases. This shows just how well designed the K8 architecture was for dualcore operation.

Hansybear
08-16-2005, 01:58 PM
Interresting thread. I'm currently sitting on a Intel P4 3,06 with HT.... and I'm tearing my hear off every time i render in lw. I don't know much about cpu's and stuff, but i have turned off my ht for the moment, because I didn't notice any difference :hmm:
Anyone have any suggestions on how to improve my rendering time?
It's worst with radiosity and antialiasing. I just leave the radiosity settings to standard.
Edit: One more question: I have only 1 gb of mem (dual channel). Can more ram improve rendering time?

adrencg
08-16-2005, 03:08 PM
So you're saying DualCore perfoms the same or better as SMP ? And with SMP being more expensive in general that would make Dualcore the better choice ? Now I'm having doubts about accuracy or I'm not sure what you mean.

You have to take into account the fact that Lightwave 8.3 is not written to use dual cpu setups for every phase of 3d work in layout (meaning poly pushing as well as rendering). We all know that dual cpu machines have been used for years for a rendering boost, but not general workflow.

Mike

thx1138
08-16-2005, 04:32 PM
You dont need MadMax to tell you a dual core will be faster than two single cores in dual operation. Though he is right ;)

Take a look at a recent article put up by anandtech. They compare the dual core X2s to the single core dual opteron work stations, and the X2s win out in almost all cases. This shows just how well designed the K8 architecture was for dualcore operation.

I thought we were talking about CPU's in general, not just AMD's. If the numbers prove DualCore X2's to be faster than regular Dual's, I'll follow that. But what about Intel DualCore's ? If they prove to be slower than Intel Dual's the general statment that DualCore is the way to go becomes somewhat tricky.

Guess there's only one way to find out. Convince my boss to get me a DualCore :D

lots
08-16-2005, 05:37 PM
Well I was referring to AMD.

Intel's dual core implimentation, is not all that wonderful. It's a hack job designed to make intel some cash while they redesign thier chip (probably for proper dual core implimentation). Also, since Xeons can already operate without the need for registered memory, much of the benifits seen in the AMD designs are probably not factors on Intel's platform.

My guess would be that dual core Pentiums will perform very similarly to thier single core dual Xeon counterparts. Perhaps being slower in many cases due to the ineffective multi core design of the Intel chips.

But why would you not buy the best ;) AMD definatly has a lead in this respect.

Jarrede
08-17-2005, 12:00 AM
I have a question regarding this topic, since volumetrics+transparency can only render with 1 thread enabled, in that situation, would you see the benifits of a dual proc?

lots
08-17-2005, 05:00 AM
If your scene is heavy on the volumetrics side, then you probably wouldnt see the benifits of more than one thread. However, if the scene is more balanced, less volumetrics and more multi threading friendly rendering techniques, then you would see a benifit.

Some tasks are just more easily computed by very fast single core chips. This is where the top of the line Athlon FX comes in.

Of cource we're over looking something here. If you pass your render job off to screamer net to be rendered, and set each CPU to render two separate frames at a time, you would in effect be rendering twice as much, even if it is volumetric and slower on the slightly slower dual core CPUs. You'd still acomplish twice as much in the same time frame, and probably out do the fast single core idea. Granted, none of this matters if you are only rendering a single frame, but for animations this is great :)

jasonfox
08-17-2005, 09:19 AM
I've recently discovered a problem with hypervoxel(sprite) heavy renders when using ScreamerNet on HT processors !

I have a small render farm consisting of mostly Hyperthreading processors with a couple of older 2.4gig non-HT processors. I was rendering a hypervoxel heavy scene and noticed that the older processors were racing ahead of everything else, this was very confusing and I initially thought it may be something to do with large Napalm files and firewalls (the 2 older machines are the only ones with no firewall installed), so I messed around with that idea for a while with no success.

Eventually I brought up task manager on one of the HT machines and forced the screamernet node to only use one thread, WOW ! it suddenly took off and started rendering at least 5 times faster !!!

Maybe something for Newtek to address !

adrencg
08-17-2005, 03:30 PM
If your scene is heavy on the volumetrics side, then you probably wouldnt see the benifits of more than one thread. However, if the scene is more balanced, less volumetrics and more multi threading friendly rendering techniques, then you would see a benifit.

:)

I just happen to be doing something with heavy volumetric clouds(we know how slow HV clouds render). Setting to 2 threads on my dual core was about a 30% improvement over 1 thread. Oddly enough, setting to 8 threads cut the render time in half. I don't actually have the true "threads" there, but it seems to calculate more efficiently.

Mike

CGTalk Moderation
08-17-2005, 03:30 PM
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.