AMD vs. Intel 3d Rendering Benchmarks

eternalseven · Jan 8, 2002

Is THG taking the proper time to provide Raytracing 3d Rendering Benchmarks? I see Lightwave benchmarks but looking at the numbers I have trouble believing Intel can stomp on AMD so easly especially since Raytracing Production quality images require a Floating Point Usage out of the CPUs.

In short, Tom needs to do benchmarks on 3d rendering using Maya on complex scenes with Raytracing to show that AMD is better. At least I think AMD should be better, maybe he already is including Raytracing stuff on software that was optimized for P4 and that's why Intel is beating AMD in the discipline nowadays.

Can anyone verify this for me?

Thanks

~ Eternal Seven

Guest · Jan 8, 2002

That is because Lightwve 7.0b is about not only any optimized Intel (not simply P4) performance but also crippled AMD performance.

Check it <A HREF="http://www.tomshardware.com/cpu/01q4/011116/duron1200-09.html" target="_new">here</A>, see it? A Celeron 1.2ghz beats even Athlon 1.4ghz and XP 1500+ in the tricky Lightwave 7.0b final rendering test in the typical showcase Skull scence.

What feature(s) helps a Celeron 1.2ghz to excel here? SSE-2 optimization? No. Celly doesn't have a clue of what SSE-2 is about. SSE? No. XP has SSE, but still been stomped. Higher FSB? No. Celly still runs at the meager 100mhz FSB. Higher Clockrate? No. Bigger L2 cache? No. Hardware Prefetch? No. XP also has it.

Newtek Software makes sure that only Intel CPU's see the gain in the 7.0b upgrade. Meanwhile, what we can see from Maya 4, 3ds max 4.2.6 and even FlaskMPEG clearly tells us that most optimizations for P4 will also work for Athlon TB/XP or PIII or Celeron.

If you look into Lightwave, strange things happen. If you do benchmarks on both 7.0 and 7.0b side by side, you will see that Athlon TB/XP rules in 7.0. However, in 7.0b AMD sees no gain at all, while both P4 and Celeron 1.2ghz have an similar amount of gain in "performance". (For some benchmark results, check <A HREF="http://www.extremetech.com/article/0,3396,apn=6&s=1598&a=16008&app=4&ap=5,00.asp" target="_new">here</A>).

I am quite sure at one point that Lightwave 7.0b as a benchmark is very close to cheating. It's not Tom's problem. It's Newtek's problem.

P.S. some other facts:
1. Maya 4 compared with Maya 3, in the final rendering stage, P4's performance improved by up to 42%, while Athlon (TB), PIII and Celeron (CuMine) also see up to 30% gain. Athlon XP has hardware prefetch, so should gain even more.
(Data from <A HREF="http://www.maya-testcenter.de" target="_new">here</A>)

2. Dicreet says that 3ds max 4.2.6 will improve P4's performance up to more than 30%. While I had a chance to test it on a Celeron (CuMine), it also showed an easy 20% gain, let alone PIII's, TBirds and XP's.

Maybe Newtek is really so NEW and so different.

-----------------------------
Some are ignorantly happy,
While some, happily ignorant.

eternalseven · Jan 8, 2002

Ok, why is it then that Tom is using a benchmark on Lightwave when it clearly cripples an Athlon? I always thought that there was something strange about the outcome of those benchmarks because I know a little more about raytracing and cpu usage then most other people do. Thanks for the reply, Tom (or whoever is running those benchmarks) should look into getting a more realistic set of software and scenes to test those properties of the CPUs.

Personally I view 3d rendering to be one of the most CPU intensive disciplines, not only that but I think to ordinary users seeing those benchmarks is just as aplicable as compiling a Linux Kernel, and those benchmarks currently show a strange picture on THG.

~ Eternal Seven

Copenhagen · Jan 8, 2002

A couple of month back I e-mail'ed the guys at Newtek responsible of Lightwave. I asked the following question:
"Does Lightwave 7b, support SSE in combination with the new AMD Athlon XP which also implements SSE ?
In other words: Is Lightwave 7b able to take advantage of AMD Athlon XP's SSE support ? If not, do you have any idea when it will ?

The reason for my question is that there seems to be a huge performance difference between the Intel P4 and the AMP Athlon XP, in favor of the P4."

The answer was:
"I'm not positive if it does or not. I know they made a few optimizations for the Pentium side. I forwarded this email to engineering, so should be able to tell you soon."

Guess what; I never heard another word from those bastards!
Certainly, it could be an oversight on their part or maybe the guys don't want to give my a direct lie in writing, so they chose to simply ignore my question.

I wouldn't be at all surprised if Lightwave is demanding a "GenuineIntel" identification in order to utilize SSE. We'll see ...

/Copenhagen - P4 Willamette 1700MHz@2109 MHz, Vcore 1.75V@2.20V on Abit TH7II-RAID.

eternalseven · Jan 8, 2002

This still doesn't answer the question why THG is using Lightwave for 3d rendering benchmarks. Clearly something doesn't add up here. On one hand he doesn't use overclocked Athlons in his benchmarks (most recently) because that would be cheating: but on the other he uses software optimized for p4 and crippling athlons for a discipline that should be dominated by athlons ... lets face it, athlons excel at 3d rendering. Someone needs to bring this to their attention, I'm amazed they haven't stepped back and started scratching their heads over it on their own.

~ Eternal Seven

MadCat · Jan 8, 2002

"... I wouldn't be at all surprised if Lightwave is demanding a "GenuineIntel" identification in order to utilize SSE. We'll see ... ..."

The problem seems wide spread. I had to patch one of my games to utilize SSE on my AMD system. In the executable, I patched the string "GenuineIntel" with "AuthenticAMD" and the multi-light source algorithm kicked in. The game uses SSE exclusively for low level night time multi light source calculations and it was rather obvious that SSE was not employed.

MadCat · Jan 8, 2002

"... What does a Celeron 1.2ghz have to excel here? SSE-2 optimization? No. Celly doesn't have a clue of what SSE-2 is about. SSE? No. XP has SSE, but still been stomped. Higher FSB? No. Celly still runs at the meager 100mhz FSB. Higher Clockrate? No. Bigger L2 cache? No. Hardware Prefetch? No. XP also has it. ..."

The latest generation of Celerons have SSE capability built-in.

Copenhagen · Jan 8, 2002

OK, here's an example of extreme performance due to the larger 2'nd level cache of the Northwood. Compare against the old Willamette:

<A HREF="http://www.aceshardware.com/read.jsp?id=45000283" target="_new">http://www.aceshardware.com/read.jsp?id=45000283</A>

/Copenhagen - P4 Willamette 1700MHz@2109 MHz, Vcore 1.75V@2.20V on Abit TH7II-RAID.

FatBurger · Jan 8, 2002

His point was that it is not an explicit advantage for the Celery, because they both have it. He is merely trying to point out that the Celery does not have any advantages (only disadvantages), but still triumphs.

Quarter Pounder Inside

eden · Jan 8, 2002

Yeah I jumped off my seat when I saw that!
I mean WTF? A Celeron betta than both new processors?
Lol something is clearly not in the picture. E-mail those guys or tell THG to switch rendering programs because I too find this odd, especially since Anandtech used Maya and had incredible scores with AXP against NW 2.2GHZ.

--
The other day I heard an explosion from the other side of town.... It was a 486 booting up...

MadCat · Jan 8, 2002

@Fatburger - You're right, I missed that meaning.

------------

At least two of my games are affected. What I've found so far:

Game #1: "Earth 2150" will not use SSE on Athlon systems. This fact was evident in the graphical quality of nigthtime scenes. I could not figure out how to patch this. The string "GenuineIntel" was not present in the binary.

Game #2: "The Moon Project" will not use SSE on Athlon systems. This fact was evident in the graphical quality of nigthtime scenes. I was able to patch the binary to force it to use SSE. I substituted "AuthenticAMD" for "GenuineIntel" to force the program to use SSE.

Game #3: "World War 3: Black Gold" may or may not use SSE (I can't tell, but the SSE capability is registered for this game in the same matter as the above two). The string "GenuineIntel" was found in this binary also.

All three game engines (they are 3D RTS games) are closely related to one another so it could be a fluke to one game developer.

Hmmmm .... Wondering if Quake 3 employs SSE optimizations ...

FatBurger · Jan 8, 2002

I believe Quake 3 does have SSE, but I'm not certain. Thanks for the info on the other three games.

Quarter Pounder Inside

Guest · Jan 9, 2002

Games may need SSE patch, but the final rendering of a 3d application does NOT. In Games we use lower precision calculations, but in a 3d application, the final rendering uses 80-bit FP. 80-bit excedes what SSE or SSE-2 (up to 2x64-bit FP SIMD) can possibly handle. (I also admit that SSE and SSE-2 can be used to optimize viewport visualization performance, but not the final rendering.)

Current so-called "P4-oriented" optimizations of decent 3d applcations like Maya and 3ds max actaully achieve improved performance through:
1) better coding; and
2) utilization of hardware prefetch.

From 1), old PIII/Celeron and old K7 (Athlon, Athlon TB and Duron) will benefit. From both 1) and 2), P4, PIII/Celeron-Tualatin, Athlon XP and Duron-Morgan will benefit.

Newtek's Lightwave 7.0b should have used both 1) and 2) to achieve a performance gain that much. However, only P4 and Tualatin see the gain. This is really strange. The AMD family even didn't benefit from 1). And I am sure if we test old PIII's and Celeron's, they will see better scores because of 1).

This is not ethical optimization. This is brand-name DISCRIMINATION.

-----------------------------
Some are ignorantly happy,
While some, happily ignorant.Edited by frankel on 01/09/02 06:29 AM.

Guest · Jan 9, 2002

Quake 3 does not have SSE or 3DNow!. John Carmack confirmed that once, but I don't have the link right now.

Quake 3 is very sensitive to FSB speed to a certain extent. A DDR266 FSB is certainly enough for Quake 3. A quad-pumped 400mhz FSB is overkill. We can tell this from the fact that clock by clock, Athlon XP is already much faster than P4 in Quake 3.

Athlon TB does not perform as well as XP does, because XP has hardware prefetch to make full use of the DDR266 FSB.

-----------------------------
Some are ignorantly happy,
While some, happily ignorant.

MadCat · Jan 9, 2002

Did find the strings "GenuineIntel" and "AuthenticAMD" in the "Getinfo.dll" file for "Return to Castle Wolfenstein". I'm going to try to obtain a benchmark for this game so that I can measure the differences in framerate when I patch "GenuineIntel" with "AuthenticAMD". I've heard that this game is based on the Quake III engine.

eden · Jan 9, 2002

Yes RTCW is Q3 based. Notice how the textures are quite similar but updated, and so as Medal of Honor AA uses the Q3 engine. Weird how they all seem to be in trend with using this old engine!
However the oddest is that in Q3 scores for Northwood are skyhigh, but in RTCW scores, the AXP rams it all, even though they both are Q3-engine-based!
But yeah I guess we all thought P4 is a gaming machine for Q3 i.e. but this is blindly cuz of high MHZ! If we took a look at the AXP at such speeds, it would definitly paint a new picture on gaming...

--
The other day I heard an explosion from the other side of town.... It was a 486 booting up...

MadCat · Jan 9, 2002

It looks like the Athlon 2000+ and the P4 2.0A are roughly equivalent in the <A HREF="http://www.anandtech.com/cpu/showdoc.html?i=1574&p=12" target="_new">Q3 and RTCW benchmarks</A> on Anandtech. But the Athlon 2000+ rams the P4 2.2GHz in Serious Sam test (my favorite FPS). Sorry for going off in different directions in this thread.

I going to try to benchmark the differences with RTCW and a patched copy of RTCW. May uncover something interesting!

Guest · Jan 9, 2002

It's not a surprise to me at all. All in all, Athlon is faster in RTCW AND Quake 3.

If all the test we do with Q3 is only Demo001 in version 1.17, then P4 is as goog as Athlon.

If we test them in other demos like NV15, which is truely heavy and demanding, Athlon is the king. If we test in later Q3 versions like TeamArena 1.29, even Athlon TB 1.4ghz is somehow faster than P4 1.6ghz. Athlon XP 1600+ @ 1.4ghz is faster than P4 1.8ghz!!! (<A HREF="http://www.anandtech.com/showdoc.html?i=1543&p=10" target="_new">link</A>)

To me, Athlon has long been the king of the "real world" and "LIVE" Quake 3. Bots can be smarter with an Athlon than with a P4, thanks to more computing power spared for gaming AI...

Just take RTCW demos as Q3 demos that don't favor P4, you will understand that P4 is stomped in RTCW.

I would like to quote John Carmack again that Q3 does NOT have SSE. RTCW may have, I am not sure. However, the SSE of RTCW should not be at the game engine level.

-----------------------------
Some are ignorantly happy,
While some, happily ignorant.

Matisaro · Jan 9, 2002

The latest generation of Celerons have SSE capability built-in.

Madcat, I think he meant since both cpus have sse, it should not be the explination for why one is so much faster.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!

Matisaro · Jan 9, 2002

A few months ago, I poiinted out that the makers of lightwave are sponsored by intel.

Does anyone have any more information on that?

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!

Guest · Jan 9, 2002

A few months ago, I poiinted out that the makers of lightwave are sponsored by intel.

Does anyone have any more information on that?

According to <A HREF="http://maccentral.macworld.com/news/0110/22.osx.php" target="_new">here</A>, Newtek admits that "...we sold more copies of LightWave 3D for the Macintosh than for Windows."

Is this good news for a software company? No. Windows PC is a much larger market than Mac market. When you turn out selling more to Mac's than to PC's, it actually means that your are losing ground to the competition in the PC market tremendously.

3ds max is becoming ever-more popular and sells like hotdogs. Cinema 4D is becoming more and more competitive. While Maya and Softimage tightly hold the high-end market. Lightwave's PC market share shrinked to such a level that its Mac version can overtake its PC version in sales.

Furthermore, Maya landed on the Mac market as well recently and this event is so much welcome by Steve Jobs (also owner of Pixar) and the whole Apple community.

Newtek and Lightwave is in real big trouble. And this is why it should not be too surprising that a once-well-respected company will take some extreme approach trying to guard its ever-shrinking market share. So they decided to stick with Intel the "market leader" and jumped down to the P4 bandwagon.

Ironically, no matter how they tried to optimize for P4, they see even better performance on the CPU's from AMD. Intel was not happy about that. Not like Autodesk or Alias|Wavefront, Newtek gave in to Intel. Then came the strange skyline of the Lightwave 7.0b performance, where P4 and its siblings are on the top top, while all those without an Intel label are caged down deep in the dungeon.

That was a day of INFAMY.

Just my guess.

-----------------------------
Some are ignorantly happy,
While some, happily ignorant.

FatBurger · Jan 9, 2002

Steve Jobs owns Pixar? I didn't know that.

Quarter Pounder Inside

LoveGuRu · Jan 9, 2002

the IQ test that was taken by both Intel and AMD gave an odd score, as Intel scored much higher, after checking it out i found the string "Genuineasshole" in the test.dll file, this is becoming a pattern.

i hope this backfires at companies and there would be some understanding about this "selected optimisation" crap.
i want same results as ATI was forced to make.
no mercy!


*******
*K.I.S.S*
*(k)eep (I)t (S)imple (S)tupid*
*******

eternalseven · Jan 9, 2002

Maya and Softimage do hold on to the high-end market, and ultimately the benchmarks posted on THG are for the high-end market as well (in terms of hardware for everyday users.) It is very poor choice for THG to include benchmarks as Lightwave 7b as in the most recent CPU Guide Column. I wasn't aware of all the specifics but I had an idea and wanted to point it out to you guys because I figured something strange was going on behind those results. Lets hope Tom (or others) notice this criticism here.

Once again, to run benchmarks using Maya is trivial, the software keeps statistics on it's rendering process any time you render and those stats are very easy to extract. They may have lisencing problems with Maya however, and that could explain why they aren't running those benchmarks. Unless they get some special agreement from Aliaswavefront they shouldn't waste their money (something like 20 thousand dollars these days.) for a regular lisence.

They need to use raytracing, the floating point unit of the Athlon will not shine unless it's a raytraced image/animation they render. And a test such as this will show exactly how much more work an Athlon does in comparison to a Pentium per clock cycle.

One last thing, the graphical business out there is huge with Macs. Big software companies for 3d software have ungodly vested interest in catering to Mac platforms because they are so widely used in the professional fields. Recently Maya went to Mac for the same reason too. Of course none of us really care to see that on THG's benchmarks because only a moron would buy a Mac for home.

~ Eternal Seven

pjotr · Jan 9, 2002

"What does a Celeron 1.2ghz have to excel here? SSE-2 optimization? No. Celly doesn't have a clue of what SSE-2 is about."

Neither does Athlon XP.

"SSE? No. XP has SSE, but still been stomped."

So does Celeron 1200.

Higher FSB? No. Celly still runs at the meager 100mhz FSB. Higher Clockrate? No. Bigger L2 cache? No."

Celeron 1200 has 256 kB L2 cache, just like Athlon XP. Celeron L2 cache is however much faster with both lower latency and higher bandwidth.

"Hardware Prefetch? No. XP also has it."

So does Celeron 1200.

Remember that Celeron 1200 is really a Pentium III Tualatin 0.13 um CPU. Sometimes P3 beat Athlon, we all know that and the Tualatin P3 has some improvements like hardware prefetch, something lacking from Athlon 1400 but not Athlon XPs.

---

Another issue you should all consider when you read THG Lightwave 7.0b tests is that THG only publish one specific rendering test. There are other tests, often used more in the real world, that displays Athlon XP as a clear winner. Check http://www.extremetech.com/article/0,3396,apn=6&s=1005&a=20703&app=4&ap=5,00.asp

"Finally, even the Northwood 2.2GHz processor can't keep up with the AMD CPU using Lightwave's volumetric rendering capabilities, as illustrated by the Lightwave Nebulae benchmark."

AMD vs. Intel 3d Rendering Benchmarks

Distinguished

Guest

Guest

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Illustrious

Champion

Distinguished

Illustrious

Guest

Guest

Guest

Guest

Distinguished

Champion

Distinguished

Guest

Guest

Splendid

Splendid

Guest

Guest

Illustrious

Distinguished

Distinguished

Distinguished

Share this page