PDA

View Full Version : Lightwave benchmarks- Dual Opteron vs. Dual Xeon


Shade01
06-26-2003, 04:05 AM
http://www.tomshardware.com/cpu/20030422/opteron-23.html#3drendering

Look how horrible any Athlon is.

Down at the bottom you can get to the entire review

MorBioS
06-26-2003, 04:28 AM
I liked of opteron

Lw is not a 64bit architecture then opteron can't show how is powerful.

Triple G
06-26-2003, 04:57 AM
I think if the Opteron had some 64-bit apps to run on it, we'd all be saying wow...look how horrible all the Xeons are. :p Time will tell...

NanoGator
06-26-2003, 05:14 AM
Funny you should mention that. My laptop is a P4 1.7ghz Dell. My desktop is a dual Athlon 1.4. Today I did a benchmark comparing the two, trying to even the playing field a bit by setting the affinity on the dual to only one proc and setting the threads to 1. Despite that, the 1.4 athlon blew the doors off my p4 1.7.

I'm totally puzzled by this. My boss has an IBM laptop that's the same mhz rating, and when I did the same style of benchmark on it, the way the math worked out was that 1 P4 megahert was equal to 1 Athlon megahert, give or take a couple of percent. This really cheeses me off because I'm this close to buying a p4, but now it seems that a couple hundred bucks more gets me a much more series dual athlon machine.

I'm in luck, though. A p4 3.06 ghz just showed up at my company. Presumably, it's top of the line (full fsb, ht, etc). I'll be able to find out if those Intel favored benchmarks really mean anything or not.

Triple G
06-26-2003, 05:36 AM
Originally posted by NanoGator
A p4 3.06 ghz just showed up at my company. Presumably, it's top of the line (full fsb, ht, etc).

Actually, I think their "top of the line" is a 3.2GHz chip now. But anyway, good luck...I'd be interested in your results. With all the shakings going on lately in the hardware world, I'm having a tough time figuring out what my next machine should be. I've been leaning toward a dual Xeon setup, but I don't know if I want to stick with Mac and the new G5s, or if I should seriously be considering the Opteron. Well, by the time I'm ready to buy, the G5s will have been out for 4 or 5 months (assuming Apple ships them on time), so I guess the smoke will have cleared a bit and it'll be easier to see where everything stands. *sigh* decisions like this are never easy...:hmm:

NanoGator
06-26-2003, 05:54 AM
64-bit is off my radar until LW is optimized for it.

Bah I want dual-core processors.

mattc
06-26-2003, 06:43 AM
Far be it for me to comment on something like this but I'd point out a few things...

1. The awful athlon scores are due to the lack of the SSE2 SIMD instruction set in LW.

2. The Opteron should score alot better since it does have these instructions. Currently they're probably held back by the relatively low clock rate (1.8Ghz versus 2.8Ghz).

3. LW is already 64-bit clean. Arnie Cachelin posted a message about that a long time ago. It's doubtful whether this would give you the speedup you think you'll get anyway.

4. Dual core processors are certainly a good idea. That's effectively what HP did with their last generation of PA cpu's. I'm pretty sure that's what IBM has been doing with it's Power4 architecture as well.

Remember kids, 64-bit fun is not the answer to everyone's prayers nor is it a solution to the problems that current machines face in trying to extract better performance.

Kind Regards
Matt

Shade01
06-26-2003, 06:46 AM
Actually, the Athlons ranked pretty bad for all of the 3d programs compared to the P4's.

NanoGator
06-26-2003, 06:55 AM
Originally posted by Shade01
Actually, the Athlons ranked pretty bad for all of the 3d programs compared to the P4's.

Including LW, which made my benchmark results baffling today heh.

Im hoping the p4 in my laptop just has some limitation i don't know about.

Shade01
06-26-2003, 07:07 AM
Well, the laptop P4's don't act the same as desktop P4's. To conserve power consumption and avoid overheating, the laptop P4's always throttle back to a lower speed. You could lose as much as 200mhz worth of speed when it does that. It almost never runs at the true stated ghz.

takkun
06-26-2003, 07:17 AM
Originally posted by mattc
Far be it for me to comment on something like this but I'd point out a few things...

1. The awful athlon scores are due to the lack of the SSE2 SIMD instruction set in LW. Not true, LW has been optimized for SSE2 since 7.0b. But you proabably meant that Athlons don't have SSE2.


I'm with Nanogator, I really don't care about 64-bit CPU's till 64-bit programs come out.

Shade01
06-26-2003, 07:22 AM
I think you'll start seeing a lot of 64 bit programs next year. The last 32 bit P4 is the 3.2 ghz. All the new P4's will be 64 bit. Like wise, Microsofts next OS will be 64 bit.

mattc
06-26-2003, 07:37 AM
Takkun,

Oops, yes, that is what I meant. Sheesh, I really must read what I type more carefully :)

Regards
Matt

lightwolf
06-26-2003, 09:43 AM
A couple of comments on the Opteron vs. P4 / Xeon "war".

Currently there is no compiler truly optimized for the Opteron, like the intel compiler which is heavily optimized for the P4 / Xeon (and this ain't gonna change until AMD release their own compiler, which I doubt, or intel adopts the 64bit extensions of the Athlon).
So, the Opteron is between a rock and a hard place. Either 32bit optimized code using the intel compiler (which is being used for LW), or "potentially" faster 64bit code (the CPU has more registers and a couple of new instructions in 64bit mode) compiled on a worse compiler (like gcc linux -> see G5 debacle).
Current processors all rely heavily on optimizing compilers if you want the last drop of power squeezed out of them (due to the long pipelines, memory access issues, paralellism within the processor etc...).
As far as the benchmarks are concerned: Tom's hardware only benches one LW scene. If you go over to www.blanos.com you'll see that depending on the scene either AMD or intel lead (intel on raytracing and GI, Athlon on "basic" rendering and volumetrics).
Have a look at this:
http://www.anandtech.com/cpu/showdoc.html?i=1818&p=11
The problem with SSE2 is that both on the P4 and the Opteron the SSSE2 unit is roughly the same speed when measured in clock cycles, but the P4 gains due to the higher clock speed.

Cheers,
Mike

bri
06-26-2003, 09:47 AM
Originally posted by Shade01
I think you'll start seeing a lot of 64 bit programs next year. The last 32 bit P4 is the 3.2 ghz. All the new P4's will be 64 bit. Like wise, Microsofts next OS will be 64 bit.

Nope, next p5- prescott is still 32 bit. More cache, more MHz, better Hyperthreading, but 32 bit. The only intel 64 bit proc is itanium. MS next os is longhorn, and it will be in 32 & 64 bit flavours. MS is finishing an AMD64 version of winXP for xmas, and w2k3 for opteron is currently in beta.

The main benefit from 64 bit isn't 64 bit memory allocation, is the extended registry set. Adding those 8 reg's to the 8 existing ones adds 20-40% performance, but the soft MUST know that it can use 16 regs and not 8, so optimizing is a must, as it was with sse2. Sadly, despite amd's efforts, it's not just recompiling.

i'm thinking about taking one nforce3 with a 240, cheap as i can, and try to optimize pov-ray in suse 64, just to see the real performance of those babies in real 64 bits.

And... c'mon, thg is well known for their incompetence with benchmarks. they don't use the benchm scenes, and i presume they leave 1 thread rendering as default, even with dual procs or hyperthreaded ones, thus under-using the power of their machines :hmm: anyway, i hope lw8 appears with a 64 bit version, or at least lw 8.5c:p :D

Shade01
06-26-2003, 10:03 AM
bri, you're missing an entire family of Intel chips. True the P5-itanium will be 64 bit, but it won't be ready until next year, for $1,338 to over $4,000 a cpu. Intel is producing a lower end line of the itanium, code named Deerfield. The Deerfield chips will still be 64 bit but they will be marketed under the P4 family of processors and will see release in a month or two.


Anyway, I think Intels 64 bit processors may mop the floor with Opteron. In addition to the extended instruction set, they will have the next generation of hyperthreading technology and come standard with 9mb of cache(!)

takkun
06-26-2003, 10:25 AM
My friend pointed out yesterday that they shouldn't call the next intel chip the pentium 5, it's a bit redundant- pentium already means five (i.e. 586) so pentium 5 would be five-five. :shrug: I suspect they'll keep the name though, it's too well known.

OT: What do you guys think about programs that go over the version 10 mark? When they get close to 10 should they rename their product and start the version numbers over again? Corel Draw is the only program that I can think of that's gone over 10. Lightwave is getting closer to ten so I wonder what they'll do?

lightwolf
06-26-2003, 10:28 AM
OT: RenderMan is at 11 !
to be honest, I don't care. Switch to Morse code, binaries or hex ;) (Lightwave 0xa would be ten, 0xb eleven...)
Cheers,
Mike - feeling geeky today

bri
06-26-2003, 10:37 AM
well shade, i think you're a bit missinformed, but anyway i don't wanna start a roadmap discussion. p5 IS a new iteration of p4 with a new sse3 instruction set aka pni (prescott new instructions), double L2 cache (1mb) and more L1, improved hiperthreading and made with 90mm tech. just take a look:
http://www.theinquirer.net/?article=10121
http://www.theinquirer.net/?article=7881
http://www.xbitlabs.com/news/cpu/display/20030602181959.html

and i'm aware of the future cranked down itanium procs, but they won't be marketed as pentium-i-don't-know-what-fancy-number-i-should-put-there, but as itaniums for workstation, as a posible replacement for xeons (or opterons, ya know;) )

mattc
06-26-2003, 11:33 AM
Thing is, the Itanium (and any iterations of it) is still so dependent on static scheduling at compile time. Thoguh there's been talk of Intel adding dynamic scheduling to the Itanium in an attempt to get over the poor IA32 performance.

On the other hand, if Newtek were to 'tune' LW to operate on the Itanium you'd probably see some serious performance increases. However, it's alot of work and highly dependent on the Intel/HP compiler technology which seems to be where they're having alot of problems.

Shade01:

Intel's hyprthreading technology is pretty poor compared to the the R&D that actually exists. On the other hand, I did find some research that indicates that simulatenous multithreading (HT) may not actually provide the performance that it initially promised.

Lightwolf is right about the lack of Opteron specific compiler optimizations though. Though in the case of LW it's pretty much all down to the SSE2 side of things so the clock speed is obviously going to make a difference.

Anyway, keep an eye on comp.arch and the various IEEE/ACM papers. These are usually a good pointer to what's coming along.

At the end of the day, there's onyl so much you can squeeze out of traditional cpu architectures like the Athlon/Opteron and the Penitum. Sooner or later, you're going to hit the performance wall. Deepening pipelines, adding more functional units on the die and so forth are all nice techniques for increasing instruction level paralellism, but unfortunately, it ends up becoming expensive in terms of R&D and real estate on the die....

Kind Regards
Matt

swampthing
06-26-2003, 01:51 PM
I wouldn't use a laptop/desktop comparison as a basis for what hardware i'd choose.

Laptops are completely different beasts and really shouldn't ever be benchmarked period because their number one consideration is NOT speed, it's mobility. Speed always takes 2nd place in laptops to mobility. If the processor needs to be throttled back in order to keep it cool under load, they do it.

Comparing laptops performance to desktop is like apples and oranges.

NanoGator
06-26-2003, 05:12 PM
Originally posted by swampthing
I wouldn't use a laptop/desktop comparison as a basis for what hardware i'd choose.

Laptops are completely different beasts and really shouldn't ever be benchmarked period because their number one consideration is NOT speed, it's mobility. Speed always takes 2nd place in laptops to mobility. If the processor needs to be throttled back in order to keep it cool under load, they do it.

Comparing laptops performance to desktop is like apples and oranges.

When I first did the experiment many moons ago, it was with a different laptop. My AMD got it's butt kicked.

The only reason I did it originally was because I was curious what the P4 optimizations did for LW and I don't have a desktop P4 to compare against. I still don't. I'm just a little blah because I can't get my personal laptop to do what the company one did. ;)

DaveW
06-26-2003, 05:33 PM
Originally posted by Shade01
Actually, the Athlons ranked pretty bad for all of the 3d programs compared to the P4's.

Not really. It's obvious that the Athlons are a little slower than the P4's, but this is a lot more noticeable in LW. In the C4D and 3ds MAX benchmarks the Athlons are only a little bit slower than the P4's they're supposed to be comparable to, and in some cases faster than the comparable P4. There are a few anomalies though, with some 2600+ scores higher than 2800+ scores, and also they test 4 single processor 3ghz P4/Xeon systems so you see all four of them always on top of the Athlons, so it appears that the Athlons are doing worse than they actually are.

Baldric
06-26-2003, 06:09 PM
Here's why the P4 scores higher than the Athlons.

A ways back a bunch of Intel engineers went from software company to software company freely tweeking each companies specific app to work much much better with SSE instructions. Newtek was one of those companies and LW was the software.

Not all areas of LW rendering were affected. Pure raytracing i.e. hall of mirrors, will probably be faster on an athlon whereas more complex scenes using radiosity, soft shadows etc, will make use of the SSE enhancements and render much faster. Toms Hardware refuses to run the scenes that would balance out the scores between the INtel chips and the Athlons. Tom's has been notified by a number of people a number of times yet refuses to change it's way??? I dont know why, I can only speculate. But be aware that thos scores are HEAVILY tipped in favour of Intel.

Now, dont fault Intel, they did a great thing in helping companies tweak their software. But do fault Toms for piss poor benchmarking.

NanoGator
06-26-2003, 06:13 PM
I did a benchmark where I took the 'textures' scene and covered all of the boxes with a high res texture. The AMD smoked my P4.

Strange, iddn't it? I assumed that because the P4 had superior memory bandwidth and a higher cache (my AMD only has 256k) that it'd rock at that test.

This is frustrating. I can take solace in knowing that it doens't really matter anyway. Most of my renderings are done overnight. :applause:

CGTalk Moderation
01-15-2006, 12:00 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.