CPU Render Benchmark

[NancyGormezano]

More dilemma for me (all 64bit):

 

Shelton: 2:32 for i7-960 (3.2GHz) ver 16b

Bobbesch: 2:44 for i7 2600 (3.4GHz) ver 16b

Illidrake: 4:30 for amd athlon 250 (3.0 ghz) ver 16 rc02

 

Illidrake: I noticed a speed up in rendering when I went from A:M RC01 (32 bit) to ver16b (7:19 versus 6:01). You might also see a speed up in 16b for the 64 bit version? Would be worth a try!

[Paul Forwood]

AMD Athlon II X2P340

ATI Mobility Radeon HD 4250

4 GB DDR3 RAM

 

A:M 16.0b (32 bit) 8:47

[Fuchur]

AMD Phenom II X6 1090T at 3,4 GHz, using 1 of 6 cores, 8 GB DDR3-1333MHz RAM, ATI HD 4870 1GB (so I dont think that that contributes to the rendering till now.

> 3:35 min while importing an obj with Troer and surfing... (so I dont think that that really took some time from my rendering).

 

Compared to that:

> 2:32 with 1/4-core-system.

 

3:35 => 215 seconds.

2:32 => 152 seconds.

 

> calculated times per frame with 6 cores / 4 cores rendering:

215 / 6 = 35,83 seconds per frame

152 / 4 = 38 seconds per frame

 

It all depends on what you want to do... on 1-core-operations, Intel i7s (4 cores but the 980x) are faster. But they are although more expensive.

On a per frame rate, AMD x6 are (really slightly) faster. (not to talk about Bulldozer-Systems with 8 cores... these, even so they are even less fast on 1-core-base would be even faster on multi-core-base)

 

So for rendering-purposes with A:M, AMDs Flagships are the better choice... for working in A:M, i7 960 (for instance) are the better choice.

But both choices should work very well with A:M and none of them is a bad choice neigther.

 

See you

*Fuchur*

[jason1025]

Thanks Furchur

 

I was curiouse how the AMD X6 1090T at 3.4GHZ would do with the AM Bench Test. Intel will always be faster but the price is not always worth it. I wish they made Mac pros with dual AMD 1090T's. If they did that they could probably drop the price $500.00-$1000.00 on the mac pros.

 

Apple / intel wants to charge a wapping $1200.00 more simply for this.

[jason1025]

Furchur

 

 

What do you think about the 8core Bull dozer vs the X6 ????

 

I was customizing on IBUYPOWER.com

 

and I can basically purchase the same system for the same price only difference is these two procs. My thinking is that I should go with the 8core even though it is slower than the 6 core. I think this because my biggest complaint with AM is render time. 8 cores will render more frames per hour than 6cores even if the six coures are a little faster core for core.

[Fuchur]

Furchur

 

 

What do you think about the 8core Bull dozer vs the X6 ????

 

I was customizing on IBUYPOWER.com

 

and I can basically purchase the same system for the same price only difference is these two procs. My thinking is that I should go with the 8core even though it is slower than the 6 core. I think this because my biggest complaint with AM is render time. 8 cores will render more frames per hour than 6cores even if the six coures are a little faster core for core.

 

Hi Jason,

 

since you are going for rendertime, I would recommend you to get the 8-core. I would go for the 8150 and at least 16 GB of RAM so. (2 GB of RAM per core should at least be)

In general I tend to buy RAM at 1333 MHz, because for most operations you dont notice 1600 MHz or more...

 

I dont have a 8-core AMD-CPU so, but it sounds promising for A:M netrenderer and you get a nice SSD-drive for the price-difference.

For not singlecore-operations (if your software can't use several cores) I would get an AMD x6 because it is less expensive or if money isnt the problem an i7 2600.

The biggest question for me is, when to get it. Maybe if you wait a month you get some pricesavings on the new CPUs, so I am not sure about that... just a guess.

 

See you

*Fuchur*

[jason1025]

It looks like Cyber power is the victor when an apples to apples comparison is made.

 

Still less than $1000.00 for all this wow.

 

 

[Vertexspline]

There are few joys in the world like the talk of a brand new computer. Smiles. (especially this many cores a churning).

[Fuchur]

With a watercooling device! If you are a little into overclocking: That will get that cores up in the sky too .

[yoda64]

Got the pieces and build today my new toy (aka developing machine :-)), and offcourse can't resist to compare with my old one

 

Intel Core i7-2600k (3,4 GHz)

32GB memory

Win7 Pro 64bit

 

A:M 16.0b 2:37

A:M 17a2 2:16

 

the old one

Intel® Core™2 Quad CPU Q9550 @ 2.83GHz

8GB memory

Win7 Pro 64bit

 

A:M 16.0b 4:25

A:M 17a2 4:04

 

Now the weekend is reserved , for installation and getting back a working machine ...

[jason1025]

Got the pieces and build today my new toy (aka developing machine :-)), and offcourse can't resist to compare with my old one

 

Intel Core i7-2600k (3,4 GHz)

32GB memory

Win7 Pro 64bit

 

A:M 16.0b 2:37

A:M 17a2 2:16

 

the old one

Intel® Core™2 Quad CPU Q9550 @ 2.83GHz

8GB memory

Win7 Pro 64bit

 

A:M 16.0b 4:25

A:M 17a2 4:04

 

Now the weekend is reserved , for installation and getting back a working machine ...

 

 

2:16

 

Yoda you now have the record for the fastest AM benchtest. But its no fair because you got to use AM 17

[robcat2075]

Smokin'!

 

How long before someone gets it down to a minute?

[Fuchur]

Got the pieces and build today my new toy (aka developing machine :-)), and offcourse can't resist to compare with my old one

 

Intel Core i7-2600k (3,4 GHz)

32GB memory

Win7 Pro 64bit

 

A:M 16.0b 2:37

A:M 17a2 2:16

 

the old one

Intel® Core™2 Quad CPU Q9550 @ 2.83GHz

8GB memory

Win7 Pro 64bit

 

A:M 16.0b 4:25

A:M 17a2 4:04

 

Now the weekend is reserved , for installation and getting back a working machine ...

 

Wow you've got a massive amount of RAM there .

Have fun with the powerful toy . Sounds like a hell of a machine...

 

See u

*Fuchur*

[yoda64]

Wow you've got a massive amount of RAM there .

 

I hope I can run now "Whole Program Optimization" build , on my old machine I got after 1 hour running a "Out of memory" error :-(

[mouseman]

Wow you've got a massive amount of RAM there .

I hope I can run now "Whole Program Optimization" build , on my old machine I got after 1 hour running a "Out of memory" error :-(

Changes like this really do change everything! I hope you enjoy the new setup!

[jason1025]

AM17Alpha2 is 14% faster on my System Than AM16b and 58% faster than AM15j+

[robcat2075]

AM17Alpha2 is 14% faster on my System Than AM16b and 58% faster than AM15j+

 

Good news, it's actually better than that.

 

With V15j+ you were getting about 6 frames per hour, with V17b you are getting about 14.5 frames per hour

 

2.4 x as many frames or... 140% faster!

[jason1025]

I am bad at math. How do you calculate this. What is the formula?

 

I did the cross multiply and divide thing. Am 15 was 595 second Am 16 was 289 and Am 17 was 249.

 

On a separate note I tested the render lock and got these results. I think its important because we often use this to see what we are getting.

 

AM 15 57-105 second

AM 16 26-34 seconds

AM 17 24-28 seconds

 

I used a stop watch but got various results. The only constant was that AM 17 was faster than the rest.

[mouseman]

I am bad at math. How do you calculate this. What is the formula?

One can make the numbers say almost anything. Statistics are evil.

[jason1025]

but I heard statistics don't lie?

 

Like 2 out of 3 doctors prefer camels.

 

[robcat2075]

It's a problem of "units" and the difference between frames/minute and minutes/frame.

 

"speed" is generally thought of in units/time like miles/hour and not hours/mile.

 

So if Car A is going 100 miles per hour and Car B is going 101 miles per hour, then Car B is going 101% of the speed of Car A and we can say Car B is gong 1% faster than Car A.

 

If Car A is going 100 miles per hour and Car B is going 200 miles per hour miles per hour, then Car B is going 200% of the speed of Car A and we say Car B is going 100% faster than Car A. Somewhat confusingly, 100% faster means twice the speed.

 

Is Car A going 100% slower than Car B? No, that would be motionless. We might say Car A is going 50% slower since it is moving half the speed of Car B, but "50% slower" isn't as intuitively meaningful as "X percent faster" is.

 

 

But if A:M v15 did 6 frames per minute and V17 does 14.5 frames per minute we can accurately say that V17, which produces frames at 240% of the speed of v15, is 140% faster than v15.

[jason1025]

Thank you my friend.

[robcat2075]

I cranked my 2.4GHz Q6600 to 3.0GHz and got

 

V15 9:53 or 6.07 frames per hour

V16 4:45 or 12.63 frames per hour

V17 4:05 or 14.69 frames per hour

 

 

so... V16 is slightly more than twice as fast as v15, it is about 108% faster.

 

V17 is quite a bit more than twice as fast as v15, it is 142% faster.

 

 

That is on this benchmark. I've had other scenes where the improvement wasn't quite so much.

[jason1025]

Not a bad deal.

 

at first glance it may seem slightly more expensive than the previous post I made above but with this coupon code SPRING0410 for basically the same price as before you get a 700w power supply up from 500, you get a 60GB OCZ agility 3 6GBs SSD up from a 500GB sata 6 HDD and you get 1600 ram up from 1333

 

 

I tried dropping the graphics card and downgrading the power supply but then you are not eligible for the coupon code. So you might as well take the free upgrades.

 

I talked to a customer service rep and he told me to wait until Friday. So im guessing ill be able to shave $100 off this price. I might grab 2 of these to add them to my render Q

 

 

Or I could do 2 of these which gets me 12 cores and 18gs of ram vs 8 cores and 16gs of ram.

[jason1025]

I purchased 2 yes 2 8core systems 3.6GHZ with 16GB of 1600mhz ram each for a total of $1286.30

 

I paid $643.15 per computer.

 

What a killer deal.

 

I wonder how these will handle the benchtest.

[jason1025]

My new system broke the 4 minute barrier.

 

This is not exactly apples to apples because I was using Netrender, however this is the fastest time on record for the AM bench test. I wonder if net render is faster at rendering because it has less overhead?

[John Bigboote]

2 Minutes EVEN. This new Boxx is pretty darned slick!

 

 

version of A:M (see Help>About A:M) ___ V17alpha2 64bit

render time min:sec ___ 02:00

CPU Brand and model ___ 3D BOXX W4920

Actual CPU speed in GHz ___ Intel i7-3960X CPU @ 3.30Ghz

how many cores A:M is using ___1(?)

RAM___ 16.0 GB

OS ___Win7Pro

[robcat2075]

render time min:sec ___ 02:00

 

 

WOW!

[John Bigboote]

2:13 in V 16.0B

 

2:08 in V17 NetRender

 

37:04 in V17 at 1280 X 720 resolution.

[thejobe]

took my work machine and put it thru the test... i think i need a new work machine. i guess its showing its age.

tonight ill send it to the render box and see how that beast does

final time was 6:18 (mins:secs)

this machine is kind of old its only a AMD 64 6000+ running a 32 bit windows (wow i screwed that up)

8GB of ram 3GB usable (wait what? have to look into that)

so i think its time for an upgrade myself

[robcat2075]

this machine is kind of old its only a AMD 64 6000+ running a 32 bit windows (wow i screwed that up)

8GB of ram 3GB usable (wait what? have to look into that)

so i think its time for an upgrade myself

 

If you put 64 bit windows on it it would be able to use all 8GB

[thejobe]

I just sent the project file to the render box. Should get results in a bit. If its too fast ill do a high res render

 

Edit:

i just got the results back

im going to have to do a bigger render because it was too fast for it to make any sense.

final time on the render box on low res was 0:23 (mins:secs)

so im going to set it up to do the render at 1080p and see how long that takes.

[Fuchur]

I just sent the project file to the render box. Should get results in a bit. If its too fast ill do a high res render

 

Edit:

i just got the results back

im going to have to do a bigger render because it was too fast for it to make any sense.

final time on the render box on low res was 0:23 (mins:secs)

so im going to set it up to do the render at 1080p and see how long that takes.

 

What is your setup on the render box?

A:M is only using one single core for rendering one frame so I can't see how you get that much faster than a Intel i7-3960X CPU @ 3.30Ghz (which is really one of the fastest desktop cpus available)?

about 6 times faster than that machine? What is the processor you are using there?

 

Sounds crazy to me or are you on liquid-cooling or better and overclocked the hell out of that?

 

See you

*Fuchur*

[thejobe]

to save time ill post what i posted in another forum about the box.

its an IBM mini, not your normal computer. someone was throwing away who deals in very large corporate type computing (banks and etc.)

the machine itself usually runs for about $5000-9000 but the software to run on it starts in the range of $50,000-150,000.

so i took the machine as is sans software. been working to get it running a unix based version but its not seeing all components yet.

each core is rated at 800mhz. it can handle up to 256 processes at one instance.

i do have access to a larger machine (that one is amazing. has 600 blades and each blade is a server. each server has 16 cores.)

and im working with the guy who runs it to help me get mine running. from what hes saying my machine can handle up to 8 versions of windows XP running at the same time

each version having 4 cores dedicated.

 

I have it semi working but can only handle so much at one time.

this was a good test for it. i was using about 50% of its full potential.

as far as the actual on screen setup this is what i got.

there are 36 versions of the same Windows running at the same time. the machine knows its 36 but it sees it as one (dont ask me how! i dont know as my friend hes the expert in this stuff) basicly when i send it out to render on that machine it is processing the prj as 1 item but it is spliting up the process into 36 parts. once each part is completed it puts it back together. got to give props to my friend for figuring this one out.

[John Bigboote]

the machine knows its 36 but it sees it as one (dont ask me how! i dont know as my friend hes the expert in this stuff) basicly when i send it out to render on that machine it is processing the prj as 1 item but it is spliting up the process into 36 parts. once each part is completed it puts it back together.

 

 

I hope this works for you, it would be AMAZING! However, I think I remember Martin talking about this, and how 3D image processing cannot be split-up that way. We'll see though! I look forward to your number(s)!

[robcat2075]

When you do the Benchmark on v16 or 17 be sure to use the associated .pre file to make sure you are using all the same render settings.

[thejobe]

well i sent the file last night with instructions to do 1080p render.

my friend got up this morning telling me that the machine has crashed and has to be redone.

he was saying that the process split was not from the A:M software. i dont understand his tech speak sometimes

but he mention something about software to hardware interrupts with a software instruction set relay system. if that makes

any sense to anyone.

 

once he gets it up and running again im going to give it another try. this has been an on and off thing for the last year now so didnt expect it to

work this time. would of been cool though. it handled the small image ok though.

[John Bigboote]

So the render only took 23 seconds, but it was too fast to make any sense? Can we see the image? 23 seconds is fabulous!

[HomeSlice]

Next time you get a successful render, I wanna see a Screenshot of the Render Panel!

[Rodney]

Does anyone know if A:M temporarily stores the render data we see in the 'Status' and 'Options' panels to a file (or perhaps if this is a debugging option that can be turned on?). At this point it is more of a curiousity for me but since that is very useful data displayed in those two panels I'd like to save it as data versus a screen captured image.

 

Anyone know if/where the rendering data is saved?

[robcat2075]

Anyone know if/where the rendering data is saved?

 

there is no log that i know of for that rendering time info

[MMZ_TimeLord]

 

version of A:M - V16.0a - 64 bit - Web

render time - 4:28

CPU Brand and model - AMD Phenom II X6 1055T

Actual CPU speed in GHz - 2.8 Ghz

how many cores A:M is using -1 Core (CPU load was averaging only 28%-32% over all six cores)

RAM - 16 GB

OS - Windows 7 64 bit w/Service Pack 1 + all updates (Current)

 

 

version of A:M - V16.0b - 64 bit - Web

render time - 4:29

CPU Brand and model - AMD Phenom II X6 1055T

Actual CPU speed in GHz - 2.8 Ghz

how many cores A:M is using -1 Core (CPU load was averaging only 28%-32% over all six cores)

RAM - 16 GB

OS - Windows 7 64 bit w/Service Pack 1 + all updates (Current)

 

Where do I find out how many CPU cores are being utilized? I don't see the option anymore under "Tools->Options" in any of the tabs. - Answered

[Fuchur]

A:M only uses one core for renderings in A:M itself.

You can only use more than one by opening A:M several times or by using Netrenderer with several netmessengers.

 

In any case each core will only work on one frame at a time. > a frame is only as fast rendered as one core would do, but you can (in your case) render 12 of them at the same time.

(if you have the licence to allow that... there should be an option to add rendernodes to your licence very soon, if Jason finds the time to test the new php-formular on the website. my first developement-tests were successful)

 

See you

*Fuchur*

[Wildsided]

Version 16a

render time min:sec - 04:14

CPU Brand and model - Intel Corei5

Actual CPU speed in GHz - 3.0

how many cores A:M is using: Pass

OS Windows 7 64bit

[MMZ_TimeLord]

Version 16a

render time min:sec - 04:14

CPU Brand and model - Intel Corei5

Actual CPU speed in GHz - 3.0

how many cores A:M is using: Pass

OS Windows 7 64bit

As far as I know A:M can only use one thread/renderer per frame. A:M itself or NetRender can NOT break down a single frame into multiple parts as far as I know.

[JohnWayne]

Happened to break the 4 minute mark with my little rig using the 15 beta and all 4 Cores with a little overclocking 4:39 without.

 

Athlon II X4 Propus 2.6ghz - OC 3.0Ghz

4GB Ram

Windows 7 64-bit

AM2007 15Beta 32-bit

 

I would imagine if some of you are breaking the 3 minute mark with some of the newer CPUs in a Single Core setup, I wonder how they would fair when you could still set the threads. I know a few of you must have the TWO AM2007 Disc as well. Nancy's Shelton Computer and Fuchar's 6 Core should not only render the teapots but probably make the tea as well.

 

Peace,

 

JW

[JohnWayne]

I didn't have time to do the math and the figuring last night, but TimeLord's Phenom II X6 using I assume 1/6 processing on V16 is getting 4:29 28% to 33% usage he said.

 

It uses similar technology to my Athlon II X4 (as the FPU - Math Coprocessor - Rendering bits are almost identical in both chips when Ghz speeds are equal). and I had to leverage all 4 Cores 95% usage on 15v to get close to the same time.

 

Seems safe to say that from v15 to v16, render times may not have necessarily when down dramatically, but the Renderer is becoming more and more efficient.

That's something to thank the Hash team for.

 

I would be interested in seeing if anyone has tests based on the new AMD FX (aka Bulldozer) CPU, from what I understand they share a FPU with 2 cores, so technically they only have 4FPUs in whats considered an 8 Core CPU.

 

Price Performance ratios for AM seem to be -- Phenom II X6 possibly then AMD FX then i7 2600K if you have a little extra, then again the extra could always be used for more RAM.

 

Peace,

 

JW

[Fuchur]

I didn't have time to do the math and the figuring last night, but TimeLord's Phenom II X6 using I assume 1/6 processing on V16 is getting 4:29 28% to 33% usage he said.

 

It uses similar technology to my Athlon II X4 (as the FPU - Math Coprocessor - Rendering bits are almost identical in both chips when Ghz speeds are equal). and I had to leverage all 4 Cores 95% usage on 15v to get close to the same time.

 

Seems safe to say that from v15 to v16, render times may not have necessarily when down dramatically, but the Renderer is becoming more and more efficient.

That's something to thank the Hash team for.

 

I would be interested in seeing if anyone has tests based on the new AMD FX (aka Bulldozer) CPU, from what I understand they share a FPU with 2 cores, so technically they only have 4FPUs in whats considered an 8 Core CPU.

 

Price Performance ratios for AM seem to be -- Phenom II X6 possibly then AMD FX then i7 2600K if you have a little extra, then again the extra could always be used for more RAM.

 

Peace,

 

JW

 

Did you render with Netrender or with Animation:Master open four times?

 

See you

*Fuchur*

[JohnWayne]

I didn't have time to do the math and the figuring last night, but TimeLord's Phenom II X6 using I assume 1/6 processing on V16 is getting 4:29 28% to 33% usage he said.

 

It uses similar technology to my Athlon II X4 (as the FPU - Math Coprocessor - Rendering bits are almost identical in both chips when Ghz speeds are equal). and I had to leverage all 4 Cores 95% usage on 15v to get close to the same time.

 

Seems safe to say that from v15 to v16, render times may not have necessarily when down dramatically, but the Renderer is becoming more and more efficient.

That's something to thank the Hash team for.

 

I would be interested in seeing if anyone has tests based on the new AMD FX (aka Bulldozer) CPU, from what I understand they share a FPU with 2 cores, so technically they only have 4FPUs in whats considered an 8 Core CPU.

 

Price Performance ratios for AM seem to be -- Phenom II X6 possibly then AMD FX then i7 2600K if you have a little extra, then again the extra could always be used for more RAM.

 

Peace,

 

JW

 

Did you render with Netrender or with Animation:Master open four times?

 

See you

*Fuchur*

Sorry for the delayed reply Fuchur,

 

I did the render with a single instance of AM with the Threads setting to 4 on v15beta which works with the TWO CD labeled HASH2007. This setting exists in 14v through the v15 betas - I do believe they removed it some time ago during the v15 cycle. I know I have to upgrade...

 

John

[MMZ_TimeLord]

Well, I re-rendered this evening for giggles.

 

On A:M V16.0b (64 bit) - Render time 04:18.

 

I had a browser open and a bit under 3.4 GB of RAM used. CPU Usage was 18-20% during the course of the render.

 

Here are the screenshots of the render in A:M and of my CPU meter during the render.

 

.

 

On NetRender (64 bit) with a single node - Render time 04:30

 

Same browser open and same pages. A bit under 3.5GB of RAM used. CPU Usage was 18-20% during the course of the render.

 

Here are the screenshots of the completed NetRender and the CPU meter during the render.

 

.

 

 

So the rule seems to be - One frame = One core's worth of Processing with A:M. BUT even with NetRender it was distributed. I had not forced an affinity on each node for a specific core.

 

For fun, I ran the same test with frame 0-11 rendered with NetRender (64 bit) and four nodes (64 bit) (my limit).

This should give a better average render number per frame. Total Time = 15:06 ... Average time per frame = (15:06 *60 / 12 /60) = 01:15.5

 

Same browser open and same pages. RAM usage went up to a bit under 3.6GB. CPU Usage jumped to a 68%-70% level during the course of the render.

 

Here are the screenshots of the completed NetRender and the CPU meter during the render.

 

.

 

Granted this was with a four node limit. Imagine the possibilities with, say, twenty nodes across four or five machines.