Optimal CAD/CAM Workstation specs and configuration for Mastercam

[jpatry]

As we all know, there is a difference between minimum specs, recommended specs, and optimal specs.

As I get more into 3D heavy toolpaths and rest mill computation, I notice huge computation times, and while watching my resource usage my CPU (Xeon E-2286M) cores are nowhere near maxed out, my GPU (Quadro RTX 5000) is basically idling, disk access is idle (Samsung 970 Pro), RAM (128GB DDR4-2666) usage is steady at around 10 to 20GB.

So what is bottlenecking mastercam, or should I say, what does mastercam really want?

[Simon Kausch]

While calculating toolpaths and stock models your CPU should be maxed out. I see your CPU is a mobile version, are there any power saving settings that are activated maybe?

If you want max performance you will need a Intel i7 or i9 and not a Xeon.

[jpatry]

On 8/14/2023 at 1:03 AM, Simon Kausch said:

you will need a Intel i7 or i9 and not a Xeon

That sounds almost preposterous, as all top end workstations are Xeon only,

 

But on  the initial subject, today I found out about the multithreading manager settings, and that appears to be where most of my bottleneck was.

Mastercam by default was only using 4 threads

[Johnward Holmeson]

16 minutes ago, jpatry said:

That sounds almost preposterous, as all top end workstations are Xeon only,

 

But on  the initial subject, today I found out about the multithreading manager settings, and that appears to be where most of my bottleneck was.

Mastercam by default was only using 4 threads

He's not wrong.  Mastercam benefits from speed (frequency) of the processor not core/thread count.  

[#Rekd™]

21 minutes ago, jpatry said:

That sounds almost preposterous, as all top end workstations are Xeon only,

Look for CPU with single thread performance, these are the ones that give the best numbers on the Benchmark test. Xeon's have never performed well in Mastercam going by past reviews. 

https://www.cpubenchmark.net/singleThread.html

 

 

[gcode]

I used to buy high end Xeon work stations from the refurb store at Dell.

This website has at least 3 benchmark threads going back about 15 years.

My high $$$ Xeon workstations were getting smoked by basic $1k i7 PC's

Mastercam, and almost all other Cad/Cam software likes clock speed, the faster the better.

Spending big $$$ on Xeon workstations is a waste of money.

Get a fast i9, 64g of ram, a solid state hard drive and the biggest Quadro video card you can afford and you are good to go.

For the last 10 years me and my employer have been buying overclocked liquid cooled Boxx Apex3 machines.

They were expensive but worth it. We've bought at least 10 of them over the years and they are 

all still in service and we've never had a problem with any of them.

Since covid hit though, Boxx's pricing structure has changed from expensive to f'ng absurd

and they played silly games the last time I called for a quote,

Since then we've been buying from Xi Computers in California.

We've purchased 2 from them and they have been solid machines at 60% the price of a Boxx.

I'm in the process of buying a new machine to replace my 3 year old Boxx at home.

 

 

 

 

[crazy^millman]

59 minutes ago, jpatry said:

That sounds almost preposterous, as all top end workstations are Xeon only,

Xeon's are meant for other things that CAD/CAM doesn't utilize the same way for. It has been proven time and time again that a i9 is much better than a Xeon.

[Aaron Eberhard]

24 minutes ago, crazy^millman said:

Xeon's are meant for other things that CAD/CAM doesn't utilize the same way for. It has been proven time and time again that a i9 is much better than a Xeon.

Correct.  If you're going to run a server (website or database), Xeons are the way to go!

[Corey Hampshire]

Any feedback on how much performance difference there is between the same generations of i7 and i9? I run i7's and was curious as to how much is gained going to an i9.

[Jobnt]

1 hour ago, jpatry said:

That sounds almost preposterous, as all top end workstations are Xeon only,

...

Oh xxxx. I'm back, sorry. I was LOLing so hard I had to let go of my desk to hold my belly (cuz it hurt from so much LOLing) so I fell over backwards and hit my head but I'm ok now!

Seriously though, not all software is the same. 

[#Rekd™]

17 minutes ago, Corey Hampshire said:

Any feedback on how much performance difference there is between the same generations of i7 and i9? I run i7's and was curious as to how much is gained going to an i9.

Look at the Bench mark link I posted above for the CPU's. The newer 13 generation CPU's  are producing better times on the Mastercam Benchmark tests as well. 

[Aaron Eberhard]

3 hours ago, Corey Hampshire said:

Any feedback on how much performance difference there is between the same generations of i7 and i9? I run i7's and was curious as to how much is gained going to an i9.

In general, the i9s are going to have a higher "boost" Ghz as well as potentially more cores.  The higher boost speeds seem to be resulting in faster processing.   In practice, like all things Intel and modern computing, it really depends on exactly the processors you're comparing, as a lot of the i7s and i9s are either equivalent or worse.   

As a note, the i7s generally have a lower power consumption (a result of the lower base & boost Ghz), so it's often a good trade off if you need to be mobile more, but in practice, you really want the i9, you're not programming long away from a power source!

Unfortunately, I don't know that anyone has done any back-to-back comparisons, say, take a comparable i7, then swap in an i9, but last I checked, the i9s were handily beating the i7s in the benchmark threads.

[Corey Hampshire]

Thanks Aaron, and #Rekd. That's the info I was after.

[Simon Kausch]

In benchmarks the i9 is 3-4% faster. The boost clock rate is 5.4ghz vs 5.8ghz.

https://cpu.userbenchmark.com/Compare/Intel-Core-i9-13900K-vs-Intel-Core-i7-13700K/4129vs4137

[jpatry]

On 8/21/2023 at 7:48 AM, Johnward Holmeson said:

Mastercam benefits from speed (frequency) of the processor not core/thread count. 

Okay, that makes sense.

So mastercam wants all the clock speed you can give it.

 

I had a sneaking suspicion that it wasn't really engineered to heavily multithread to the true extent of such hardware.

[byte]

37 minutes ago, jpatry said:

Okay, that makes sense.

So mastercam wants all the clock speed you can give it.

 

I had a sneaking suspicion that it wasn't really engineered to heavily multithread to the true extent of such hardware.

That's not really true, where multi-threading comes into play is with multithreaded operations.

If you regen a boatload of multi-threaded operations, you will see the cpu spike to 100%.

Its important to note that multi-threading needs to be set to on in the configuration.

[byte]

Since core operations such as Area Mill, Opti-Rough, & Unified Multi-Axis heavily employ multi-threaded & regeneration times tend to be the largest bottleneck in the software, I would say nothing could be further from the truth, although the single threaded graphics pipeline and file system events can cause bottlenecks, so a balance of many cores & many threads as well as high clock speed and a strong GPU is ideal.

Of course a lot of it has to do with how your programs are built as well, I find more operations targeting less features is typically faster, although Mastercam has been working hard to fix that, I had a case a few years ago where it was infinitely faster to use one operation per pocket in area mill, and that was no longer the case as of 2023 I believe 

[jpatry]

51 minutes ago, byte said:

Since core operations such as Area Mill, Opti-Rough, & Unified Multi-Axis heavily employ multi-threaded & regeneration times tend to be the largest bottleneck in the software, I would say nothing could be further from the truth, although the single threaded graphics pipeline and file system events can cause bottlenecks, so a balance of many cores & many threads as well as high clock speed and a strong GPU is ideal.

Of course a lot of it has to do with how your programs are built as well, I find more operations targeting less features is typically faster, although Mastercam has been working hard to fix that, I had a case a few years ago where it was infinitely faster to use one operation per pocket in area mill, and that was no longer the case as of 2023 I believe 

Yes, I mentioned that I found out about that a few days ago, and it made a massive difference when computing dynamic rest mill operations.

Out of the box, mastercam seems to be set at 4 threads only.

[byte]

46 minutes ago, jpatry said:

Yes, I mentioned that I found out about that a few days ago, and it made a massive difference when computing dynamic rest mill operations.

Out of the box, mastercam seems to be set at 4 threads only.

Well, multi-threading works best with a different number of threads based on your application, too many threads will cause lag as well because there is a performance hit for changing contexts,

I don't generally farm work out to more than 4 threads, I find splitting work up into smaller separate applications works better than using many threads from one application. That's what I see the big guys doing too, a lot of small applications working together..

[byte]

I ran 2 identical builds for a customer today of a chook, there is a lot of regeneration 2024 was about 2x faster

Than 2023

[Jake L]

Gonna hijack this thread. I'm getting back into a large project from a few months ago and I forgot how much it sucks to wait for 20 stock models to regen when each one takes 30-60 seconds. I enjoy whistling but my lips get tired after that long.

Here's where I'm at:

Dell Precision 3571 laptop

CPU – Intel i7-12800 2.4GHz

RAM – 32GB

GPU – NVIDIA T600

The part - picture a mashup of an aerospace part and a mold 3 x 9 x 24".

The files - 300-600MB with 150-300 operations (10-15% of operations are stock models). All stored locally. I am connected to a company network but the only thing          Mastercam pulls from the network is tool holder libraries when I select a new holder.

stock models - each uses the previous stock model as stock and on average gets 1-10 operations each. All tolerances set to .001.

Here's my main issue, nothing seems to be maxed out when regenerating these stock models. I have multithreading set to use 10 out of my 14 threads and priority set to high. Watching task manager, CPU almost never goes over 35% (sits around 25-30% and spikes to 35%) and Memory stays at 50%. In configuration > Toolpaths > Memory buffering I set "% of physical memory" from default 50 to 85 and saw no difference. It seems something is still limiting Mastercam to only using 50% of my memory.

My question is, what am I missing here? Obviously something else is limiting performance but what?

Side note, when Mastercam is open and idle, my memory is at 45-50%. Not sure if that's normal or something else is screwed up on my end. This is the case for any file, not just these larger ones.

Thanks in advance for any help. MC2023

 

[crazy^millman]

10 minutes ago, Jake L said:

Gonna hijack this thread. I'm getting back into a large project from a few months ago and I forgot how much it sucks to wait for 20 stock models to regen when each one takes 30-60 seconds. I enjoy whistling but my lips get tired after that long.

Here's where I'm at:

Dell Precision 3571 laptop

CPU – Intel i7-12800 2.4GHz

RAM – 32GB

GPU – NVIDIA T600

The part - picture a mashup of an aerospace part and a mold 3 x 9 x 24".

The files - 300-600MB with 150-300 operations (10-15% of operations are stock models). All stored locally. I am connected to a company network but the only thing          Mastercam pulls from the network is tool holder libraries when I select a new holder.

stock models - each uses the previous stock model as stock and on average gets 1-10 operations each. All tolerances set to .001.

Here's my main issue, nothing seems to be maxed out when regenerating these stock models. I have multithreading set to use 10 out of my 14 threads and priority set to high. Watching task manager, CPU almost never goes over 35% (sits around 25-30% and spikes to 35%) and Memory stays at 50%. In configuration > Toolpaths > Memory buffering I set "% of physical memory" from default 50 to 85 and saw no difference. It seems something is still limiting Mastercam to only using 50% of my memory.

My question is, what am I missing here? Obviously something else is limiting performance but what?

Side note, when Mastercam is open and idle, my memory is at 45-50%. Not sure if that's normal or something else is screwed up on my end. This is the case for any file, not just these larger ones.

Thanks in advance for any help. MC2023

 

Why are the stock models tolerance .001? Try .005 or even .01 as a tolerance. People who use stock models for verification are wasting time and effort for no good reason in my humble opinion. Why over burden the process for something that is meant to help speed up the process. I run 128GB on a i9 with RTX5000 card and I cannot get away with .001 tolerance on large files. Sorry you have a cracker Jack computer trying to do high end work. Tell management to quit being $1 million foolish and penny wise and get a real professional system.

[Jake L]

12 minutes ago, crazy^millman said:

Why are the stock models tolerance .001? Try .005 or even .01 as a tolerance. People who use stock models for verification are wasting time and effort for no good reason in my humble opinion. Why over burden the process for something that is meant to help speed up the process. I run 128GB on a i9 with RTX5000 card and I cannot get away with .001 tolerance on large files. Sorry you have a cracker Jack computer trying to do high end work. Tell management to quit being $1 million foolish and penny wise and get a real professional system.

Thanks for the response! Management recently said they were willing to spend some money on new computers but wanted to figure out a good "baseline" computer so all the programmers (~10 of us) can have the same setup. They started by using the benchmark file to test different GPU's..... 

They're current stance is "why get new computers if we aren't leveraging 100% of what we have?" So I need to figure out how to answer that question now.

You hit the nail on the head with "stock models for verification" that is what I do. What're alternative methods? Verify compare? Always looking to program faster. I usually stick around .001 because I read (probably on this forum) that stock model tolerance should be around 30% of stock being left on. I rough to +.005 stock so .001 seemed suitable.

Original question still stands, I'd like to see my CPU and Memory at 80-90% (or more) when crunching stock models, how do I get there?

[crazy^millman]

With that kind of logic have the put in Cell systems to leverage 100% up time on the machines? Do they have redundant pallets and use parent child tools on the machine? Let me guess they show up to the INDY care race in a 1962 VW bug and wonder why they don't even qualify for the race. Not leveraging what you have means they lack a true understanding of what computers do and how they work. Does every machine in your shop running 100% spindle load 100% of the time? You are falling short if you are not with this stupid logic. If that tap is not using 100% spindle load every time you are tapping the you have failed as machinist in their eyes. Do they drive their cars at max RPM through all gears and make sure they are driving at 200 mph every where they go? No way just an excuse to penny pinch due to their lack of understanding and making up ways to do it.

This right here reminds me why I am not a good employee. I went out and bought my own system at the last 5 machine shops I worked in. I showed the owners I was willing to put my money where my mouth was. Think any of the gave me a bonus for making them more money than they were ever making? No way no how, but I know what is the right way to do my job and if needed a tool I bought it. I still have over $40k of Precision hand tools. I sold $40k of tools those last 5 shops because I had better tools than they did. Takes money to make money and sit down and figure out how much time each programmer is wasting in a day waiting. Then multiply it by 5. That is how much profit the company is losing not getting better systems. My answer to them would be the following as I walked out the door. You want to throw away money fine, but don't come complain to me when we have to close the doors because you kept making lame excuses. Will it be night and day different now, but better systems for the programmers is money in the bank plain and simple.

[Jake L]

2 minutes ago, crazy^millman said:

With that kind of logic have the put in Cell systems to leverage 100% up time on the machines? Do they have redundant pallets and use parent child tools on the machine? Let me guess they show up to the INDY care race in a 1962 VW bug and wonder why they don't even qualify for the race. Not leveraging what you have means they lack a true understanding of what computers do and how they work. Does every machine in your shop running 100% spindle load 100% of the time? You are falling short if you are not with this stupid logic. If that tap is not using 100% spindle load every time you are tapping the you have failed as machinist in their eyes. Do they drive their cars at max RPM through all gears and make sure they are driving at 200 mph every where they go? No way just an excuse to penny pinch due to their lack of understanding and making up ways to do it.

This right here reminds me why I am not a good employee. I went out and bought my own system at the last 5 machine shops I worked in. I showed the owners I was willing to put my money where my mouth was. Think any of the gave me a bonus for making them more money than they were ever making? No way no how, but I know what is the right way to do my job and if needed a tool I bought it. I still have over $40k of Precision hand tools. I sold $40k of tools those last 5 shops because I had better tools than they did. Takes money to make money and sit down and figure out how much time each programmer is wasting in a day waiting. Then multiply it by 5. That is how much profit the company is losing not getting better systems. My answer to them would be the following as I walked out the door. You want to throw away money fine, but don't come complain to me when we have to close the doors because you kept making lame excuses. Will it be night and day different now, but better systems for the programmers is money in the bank plain and simple.

I loved reading every word of this response, thank you for this. 

What's a good load range for a computer? above or below 50%? We run some of our machines at 90% load consistently, but you won't see me running a car at 90% of it's RPM.