Centrino kicks rendering butt (1.3GHz is 75-100% faster than my 2.4ghz P4)

[lesterc]

:eek: :eek:
thats very interesting
hmmm

thanks for the info :beer:

[jbw]

Centrino is the best of both pIII and p4, apparently?
You've got better performance per clock cycle plus you're using a faster front side bus and support for simd and simd2 instructions. Apparently a 1.6 centrino will outperform a 2.4ghz p4 machine?

ta,
jbw

[Aearon]

yeah you get the best performance per clock cycle, at least in some applications (..that benefit from a large cache), but that is not essentially a sign for better engineering.. the p4 was simply designed with high clock speed in mind, so it's just a different approach..

the centrino doesn't have a higher fsb btw..

and as i said, in 3dsmax even this 1.3GHz centrino outperforms my 2.4GHz p4, this doesn't hold true for all applications though

[Matt]

Bah, I haven't looked at today's prices, but you can afford like four athlonXPs for the price of a single Intel CPU, lol.

[jbw]

the centrino doesn't have a higher fsb btw..

Oops... my bad. :blush:

ta,
jbw

[GregHess]

Make sure that...

On the P4 System....

1) Your using Max 4.26 or Max5.1.
2) If your using Max5, that you turn SSE on in the render dialog box (its disabled by default).
3) Your using up to date Intel INF and the Intel Application Accelerator (if they apply to your system).
4) You use Sisoft sandra to verify your system is running within reported specifications. (http://www.sisoftware.net/)
5) You use a suitable benchmark scene...like islands.max at 2048x1536 to compare the two systems.

Verify all that and then report back please.

Centrino is fast...but its not twice as fast as a chip running at over 2x the megahertz.

[Thalaxis]

I personally LOVE my Centrino machine... but one must keep in mind its design goals. It's never going to match a P4 + 800 MHz FSB + HyperThreading. It shows more of a focus on power savings than on performance, and it shows.

That said, it's a great chip; it has SSE2, a 400 MHz FSB (just like the one on the original Willamette core P4), and a 1 MB cache.

And in fall when it goes to the 90 nm process, it will get a 2 MB 2nd level cache and a 1.8 GHz clock speed. Apparently, Intel is planning on marketing that as a blade solution, so the renderfarm wishes might not be far off. :)

BTW (just to prevent confusion for people doing any shopping), Centrino is not a chip; it is a chip (Pentium-M, code named Banias in its first iteration, Dothan in its second, described above) combined with a supporting chipset with a specific feature set.

So if you see a laptop with a Pentium-M, you're looking at the same processor, just teamed with a different chipset from Centrino.

[Aearon]

GregHess: I understand your disbelief, but it sounds like you automatically assume it's my fault, think for a moment: i already said it's slower in other apps.. so it's not a general problem of my p4 system. you've made some valid points though..

1. I'm using 5.1 on both, they are configured 100% the same
2. I've been using max for quite some time and my system is setup right as far as i know, i've been using brazil 1.0 for the test btw.
3. first things i installed after windows
4. both are
5. a suitable benchmark scene is whatever i use in production, why would i "synthetify" a benchmark when i can just use my production environment which is where i'll use the systems

Both systems are optimally configured and what i reported are simply the results i get. Look at some benchmarks again, in the reviews i read it's much faster than the p4 in specific applications..

Matt: You have a point, it would still be interesting to see the centrino on the desktop market though, there's no denying that. And btw: I'd prefer a 2.4C @ 1000FSB to a AMD system at any time ;)

[GregHess]

Fist,

I wasn't assuming you were wrong...just you didn't give enough information on the two systems or configurations to lend much credibility to the benchmark.

I was trying to get some questions answered so your setups would be better understood.

According to data online a 1.6 centrino is pretty close in performance to a 2.4B P4 (www.anandtech.com) but when certain tests are run...it can be almost 50% slower.

I'd encourage you to run the 3dluvr max benchmarks (which contain scenes which are readily available on the max cd's) to verify your P4 2.4B is running within specs, as well as submitting your centrino data in the benchmark list.

http://www.3dluvr.com/content/maxbench.php

http://www.3dluvr.com/content/article/3

I must admit that I am rather skeptical...having be exposed to some of these new chips...knowing quite well that they aren't running twice as fast as the workstations sitting next to them.

[Aearon]

GregHess: ok got you. I'd probably be as sceptical, if i find the time i might run the 3dluvr benchmark suite, but i think i already did all that's needed to verify the performance of my p4 system (it performs on-par in Sandra)

But I'll also compare my rendering results with some online data soon if i find the time.

[GregHess]

Another easy and simplistic benchmark to run (download and exe) is the cinema IV benchmark.

http://www.maxon.net/pages/download/cinebench.html

And here's a thread with submitted data.

http://www.cgarchitect.com/ubb/ultimatebb.php?ubb=get_topic;f=2;t=000146#000000

Btw this also runs on mac's.

That would be another good test to run on the centrino and the P4 system.

[mark_wilkins]

Using an independent benchmark may give you a useful hint that there's something out of whack with your configuration. I think that's why Greg recommended it.

-- Mark

[elvis]

anandtech reviewed the centrino when it first debuted and they got similar results (in that clock for clock the centrino beats a P4).

the amazing thing is the centrino is a much lower voltage part (meaning cheaper to run and cooler), which makes it a great alternative for multi-procesing parts, especially blade servers and other high-density CPU solutions.

the obvious step for me is large scale render farms based on centrino technology. it's a shame intel are hholding centrino back for the laptop market when there's such a potential right under their noses. i guess they're afraid to lose money on their own product. imagine the P4 bombing because another intel product outsold it! it's a shame marketing once again overrides engineering and the quest for a better CPU.

[Aearon]

Greg: thx again for the tips. i tried your 3dluvr benchmark suite (in parts) and got results similar to the 2.53GHz system, so i guess my system is alright, but the centrino system isn't faster in those benchmarks... i think it's brazil.. mhh

elvis: yep exactly my thoughts

[GregHess]

Fist,

Could very well be. Vray/Brazil and other third party renderers are MUCH more efficent then max's scanline renderer. The difference is astounding.

For example with Dual Xeon's...enabling hyperthreading under max5 gives around a 5-8% gain in performance.

Do the same thing under vray and your performance jumps over 25%.

Thanks for running those tests.

[Aearon]

ok i think we cleared it up than. brazil being my most used renderer i think the centrino is definitely a nice addition to my renderfarm :)

[Thalaxis]

Originally posted by elvis

the obvious step for me is large scale render farms based on centrino technology. it's a shame intel are hholding centrino back for the laptop market when there's such a potential right under their noses. i guess they're afraid to lose money on their own product. imagine the P4 bombing because another intel product outsold it! it's a shame marketing once again overrides engineering and the quest for a better CPU.

Your wish might well come true this fall, when Intel launches the
next version of that processor.

[elvis]

Originally posted by Thalaxis
Your wish might well come true this fall, when Intel launches the
next version of that processor.
very tasty!! :thumbsup:

[Thalaxis]

If you think that's tasty, wait until you see the specs. The power
consumption is LOWER, the die is smaller, the 2nd level cache is
2x larger, and the clock speeds are higher (starting at 1.8 GHz).

And apparently they are launching them ahead of schedule.

[Aearon]

well it's basically just larger cache and a die shrink.. but smaller manufacturing processes actually give you even more advantages in the notebook sector, so centrino gets even better ;)

[elvis]

it's no secret intel relied on the scalability of the P4 to get them up into the multi-gigahertz range, and therefor be a competative product. AMDs "PR" rating scheme is proof that MHz don't mean anything (as anyone reading this thread probably knows too well).

it's good to see intel go for a better CPU design instead of numbers and marketing. i'm pleased they've finally engaged their brains and not their wallets.

[Aearon]

true: the p4 design was probably influenced by marketing
wrong: the design is bad

just because it's less effective per mhz doesn't mean it's bad, this is intended... the top performing p4's are faster than the top performing athlons right now, so the design works... the amd 'raw power' approach seems more solid but it's not essentially the better design

i say: whatever works..

[Thalaxis]

Originally posted by fist
true: the p4 design was probably influenced by marketing


Wrong. The P4 design was influenced by the fact that it is easier to increase clock speed than it is to find parallelism in x86 code, in addition to the fact that the clock speed method worked exceedingly well for the Alpha.


wrong: the design is bad


Agreed. Neither is the Athlon's... though I would definitely say that AMD showcased some impressive engineering to get as much performance as they did out of their lower clock speeds.

Given equivalent manufacturing technology, I think it would be very much up in the air as to whether or not the P4 could outperform the Athlon, but the manufacturing edge that Intel has is giving them a lead that AMD can't quite match with even their engineering.

I would not be surprised if the Athlon (even less surprised for Athlon64/Opteron) outperformed the P4 if backed by the same manufacturing prowess, though.

[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.