Performance test results

tony_459 · July 2019

I configured my browser to always use my all-powerful NVIDIA graphics card, but my Windows performance manager suggests it isn't being used much at all. What I see while I wait for a feature to register in a model:

Image: https://us.v-cdn.net/5022071/uploads/editor/1q/lwakz4popmdf.png

Notice the NVIDIA row at the bottom left, which says only 7% of the GPU resources are being used. I have never seen it above 15% while using Onshape, even though it does spike a slight bit in more demanding tasks. This is very different from what I see when I run the Onshape performance test:

Image: https://us.v-cdn.net/5022071/uploads/editor/wf/ttoqa8obw1xy.png

I took this capture before the GPU resources peaked at over 90%. Not sure why the performance test would tap so much into my NVIDIA GPU when my modeling work doesn't. I've gone through all the browser configuring and driver updating there is to do. Anyway, just thought I'd share. Anyone else seeing the same discrepancy?

ilya_baran · July 2019

@tony_soares459
Performance is determined by many factors -- the GPU is responsible for only certain aspects. Please see my blog post on the subject: https://www.onshape.com/cad-blog/under-the-hood-how-onshapes-development-team-tackles-cad-performance especially the table in the post.

daniel_cook · July 2019

tony_soares459 said:

I configured my browser to always use my all-powerful NVIDIA graphics card, but my Windows performance manager suggests it isn't being used much at all. What I see while I wait for a feature to register in a model:

Notice the NVIDIA row at the bottom left, which says only 7% of the GPU resources are being used. I have never seen it above 15% while using Onshape, even though it does spike a slight bit in more demanding tasks. This is very different from what I see when I run the Onshape performance test:

I took this capture before the GPU resources peaked at over 90%. Not sure why the performance test would tap so much into my NVIDIA GPU when my modeling work doesn't. I've gone through all the browser configuring and driver updating there is to do. Anyway, just thought I'd share. Anyone else seeing the same discrepancy?

Usually a performance test is a sustained series of commands sent to test for maximum performance so it loads up the GPU. When you are actually modelling, it's a series of short bursts only when doing actual changes to rotation or panning / zooming of your model.

Windows performance monitor has a polling interval of a few seconds - your GPU is probably bursting to high usage but only for a half second at a time.

Get your most complex model, put lots of detail on the screen and pan around rapidly for 20-30 seconds then check your GPU usage....

Jake_Rosenfeld · July 2019

@tony_soares459 I can confirm what @daniel_cook is saying.

The performance tests works by drawing a bunch of geometry into a buffer. It starts with a small amount of geometry and scales up how much geometry is being drawn until the framerate drops below a certain threshold (i.e. until the GPU can no longer keep up with the amount of geometry we are asking it to draw). Then we report the results back to you. For that I would definitely expect a ramp up to a spike.

For normal usage, if you are not interacting with the model, the system would not being doing any redraws. So as Daniel says, maybe wiggle the model around for a while and watch that panel to see how it changes. If your graphics card is good enough, you may even just register blips until you open something really complicated and wiggle it around.

tony_459 · July 2019

Thank you both for the (very helpful) responses! Good to know a 7% GPU performance reading doesn't mean my hardware is just sitting idle. I still feel things could run faster, and I suspect that maybe thermal throttling is stunting my laptop performance at times, but at least things are running!

john_mcclary · July 2019

Also, on a laptop keep in mind, you will need to have the power pack plugged in too.
at least on mine, it won't go past 2ghz unless it's plugged in.

bill_daniels · July 2019

I noticed a loss in graphics performance last week. I ran the OS system performance check, got all green check marks but the speed test failed because "frame rate was below 60FPS". So I ran several on-line graphics benchmarks on my system (Nvidia 1050 Ti) and found I was faster than 2/3rds of the systems tested. I had suspected an Open GL upgrade I had applied was the problem but apparently not. No one else in the Linux world is complaining. So what's going on?

Henk_de_Vlaam · July 2019

@bill_daniels

I had the same problem, see https://forum.onshape.com/discussion/12065/no-results-from-performance-check#latest.

But today I got my results of a DELL G7 7790 with a NVIDIA GeForce RTX 2060.

owen_sparks · August 2019

bill_daniels said:

I noticed a loss in graphics performance last week. I ran the OS system performance check, got all green check marks but the speed test failed because "frame rate was below 60FPS".

I tried the "System check" on a couple of low end PC's in our company just to gather some numbers to justify some better kit. Both had previously given numbers around 20 Million Triangles but I couldn't remember the number exactly. Each now refuse to complete the test reporting the low frame rate.

My proper PC also reports 100M fewer triangles than it used to.

Owen S.

svenolov_nystrom · August 2019

I am currently working in Spain with a connection that goes down to 4-5 Gbs (it works) if it drops under 3Gbs it does not

I normally work with 500Gbs and it is no big difference It suprices me a bit

An AMD RX580 Ryzen 5 Ubuntu 19.04 (latest kernel has support for Rx580) on older linux kernel you must install the proprietor driver. The graphic performance is great ca 800 million triangles.

Linux provides a pleasant experience with Onshape. The space-mouse doesn't work but you can live with that

tony_459 · October 2020

On a new Razer Blade 15 with 10th gen Intel I7 CPU and NVIDIA GeForce RTX 2060 GPU, running on Firefox:

By comparison, my older Razer Blade 15 with 8th gen Intel I7 CPU and NVIDIA GeForce GTX1060 Max Q, also run on Firefox:

Image: https://us.v-cdn.net/5022071/uploads/editor/yt/svgedhla409e.png

Can't tell if the improvement is real since the test results can vary by more than that from one run to the next.

matthew_stacy · October 2020

I recently splurged on a new laptop and a shovel ... to bury the old HP. Discrete graphics increase the performance by more than an order of magnitude. There is significant variability though. Might that have something to do with the whims of an internet connection?

Charles_Ingalz · October 2021

M1 Max 32 GB OSX12.0, everything else says compatible on both chrome and safari. This is the chrome result, but performance numbers are nearly the same with Chrome or Safari.

Image: https://us.v-cdn.net/5022071/uploads/editor/j1/7jn73p9gl15j.png

nick_papageorge073 · October 2021

Charles_Ingalz said:

M1 Max 32 GB OSX12.0, everything else says compatible on both chrome and safari. This is the chrome result, but performance numbers are nearly the same with Chrome or Safari.

Which gpu did you order?

Charles_Ingalz · November 2021

It is the 32 core GPU. To compare, this is the M1 MBP with the 8-core GPU. The triangles per second seems to taper out at some point. The lines per second scales fairly linearly to 32 cores.

I was using the Dell Vostro 7500 with GTX 1650ti, but the improved graphics, 120hz, brighter display, and ability to use it with battery power make a considerable difference in everyday use.

Image: https://us.v-cdn.net/5022071/uploads/editor/5n/11hswfrfnc0c.png

gkuhns · November 2021

The Ryzen CPUs with integrated graphics seem to perform pretty well. It is important to equip them with fast RAM.

Below are test results of a Ryzen 3 5300G. I ran tests with both 2666MHz and 3200MHz ram in both Edge and Chrome. Chrome performed better so these results are from Chrome.

First, here are the results from the slower 2666MHz ram:

Image: https://us.v-cdn.net/5022071/uploads/editor/vo/1784sbxtyf35.png

And here are the results from the faster 3200MHz ram.

Image: https://us.v-cdn.net/5022071/uploads/editor/x0/eirp5tcrgufw.png

Because of low utilization, I am ignoring the difference in quantity of ram. The 2666 kit was two 4GB sticks for a total of 8GB. The 3200MHz kit is two 16GB sticks for a total of 32GB.

Performance test results

Comments

Howdy, Stranger!

Quick Links

Categories