Specs for fastest ToneX advanced learning times?

[vintagecharlie]

Hi everyone,

what would be the main specs of a PC for the fastest possible learning speeds? What are the bottlenecks and what are the main priorities to get fast speeds?
I assume the nVidia GPU is Nr. 1, but how big are the gains in speed with high end cards over something like an RTX2060 (my current card)?

Thanks in advance for any input on this!

[RiF_2205]

I am getting 19 mins training time in advanced mode on my 4 year old Intel i9-9900K, nVidia GTX 1650 super 4 GB (€200 low end card, but fairly recent design), 32 GB of RAM.
I heard of people getting as low as 15 mins. I think it was a RTX 3070. No matter the system, it did not get faster than 15 mins, afaik.
That said, my findings so far are as follows:
CPU: As long as it's a decent CPU, it does not seem to make much of an difference
GPU: nVidia is a must, but the range goes from 19 mins (€200 card) up to a minimum of 15 mins (€600 card). There is no info that I know of where an even faster card yields better results. At least with the current TONEX version
RAM: 32 GB is a must as the process needs a good 15 GB already

[vintagecharlie]

Thanks for this very valuable input mate! That put it very well into context and i see what to expect and that it makes not much sense to go overboard with a cutting edge system for this at the moment. I have a decent nVidia GPU - an RTX 2060, but the CPU is just an i5 something soomething (i think 6600K) and i just got 17-18 minutes on my first advanced capture.

[RiF_2205]

If you are already getting 17-18 mins (which is pretty close to the best times around, afaik), I would not shoot into the dark and invest in anything to get better times. At least not without knowing 100% for sure what performance gain to expect.

[outloaf]

I'm tempted to buy a 1650 in hops that it would decrease my times from 37 min for an advanced capture.. not sure how a rtx 3080ti and an i9 10850k with 32GB ram is getting such slow advanced times..

[RiF_2205]

I highly doubt that a lowly 1650 would outperform a 3080i. No way.
There must be something else influencing the training times.

[ProTools4]

capture process time vs GPU models' effect on time...
that is the question....

there are differnt versions of CUDA core architecture as far as i can research

it may be possible tonex capture "is optimized" for older GPUs?

but until IK tells us some info or some tonex user goes non linear and test a range of GPUs and makes an excel spread sheet we are stabbing in the dark.

My $1600 PC gaming machine with a 3060 is around 18m for advanced
My $4000+ Mac Rack Pro industry standard recording studio machine w/ rtx6600 is 5ish hours
*pittiful and unusable

My personal example shows raw CPU power and multicore CPUs mean nothing to this issue.

Would be nice to know what GPU IK used for software development.

[RiF_2205]

The RTX660 in your Mac is an AMD GPU, which are not supporting CUDA. For TONEX, you'll want an nVidia GPU.

[ProTools4]

Mac Radeon Gigabyte 3x fan RTX 6600 and I am aware it is AMD
this is a top tier gpu in a Mac Rack Pro that retails over $7K

Apple has not supported any and all nVidia GPUs for a decade now.
NVIDIA GeForce GTX 760 released in 2013 was the last supported NV GPU.
GTX760 is not compatible with modern monitors.

ToneX is claimed to Mac Compatible
Mac ToneX has a Capture button and Advanced button.
right there on the screen.....

So how do we get reasonable capture time for Mac users?

Fast Forward to 3:30

[ProTools4]

AMD GPU, which are not supporting CUDA. For TONEX, you'll want an nVidia GPU.

Please show me exactly were is states that AMD is NOT supported?

The last Mac supported NV card 2013.
Minimum suported Mac OS for ToneX is OS11 = release date 2020.
If an NVidia card is 100% required for ToneX capture how can any Mac be compatable?

Can any other Mac users weigh in here?
My Mac has Intel Xeon W... Not Intel i3 or M2

BTW from Google : The Intel Core i3 processors do not have CUDA cores,

System Requirements
Mac®
Intel: Intel® Core™ i3 (Core i5 recommended for Modeler), macOS® 10.15 (macOS® 11 Big Sur required for Modeler), 4 GB of RAM, 6 GB of hard drive space

Apple Silicon: Apple M1, macOS® 11, 4 GB of RAM, 6 GB of hard drive space. TONEX is a native Apple Silicon application.

Windows®
Intel: Intel® Core™ i3 or equivalent CPU (Core i5 recommended for Modeler), Windows® 10, 4 GB of RAM, 6 GB of hard drive space, NVIDIA GPU with Maxwell or newer architecture is recommended to reduce training times (eg. GeForce GT 750 or higher).

[Peter_IK]

Nobody said it was 100% required, let's keep the discourse civil. This will be the only warning about this.

Bottom line is Nvidia graphics cards are going to speed up the process. If you need specific benchmarks and users are not chiming in, you might want to ask the IK Support team as they may have tested on various machines and have fielded questions from various customers so they could probably give some advice.

[ProTools4]

Nobody said it was 100% required,

I quoted above where Riff infers AMD is not supported.

let's keep the discourse civil. This will be the only warning about this.

Comments withheld via TOS

you might want to ask the IK Support team as they may have tested on various machines and have fielded questions from various customers so they could probably give some advice.

I contacted support a while ago with no reply, But this has less to do with me.
And more to do with the elephant in the room.

For the good of the Mac user base...
How to speed up Mac capture to a relevante capture time when Mac does not support Nvidia?

Peter can we get an olive branch on this topic?

[Peter_IK]

Riff stated AMD GPUs do not support CUDA and the Nvidia will perform better, as I've read it. It seems to be the case so I don't think that's worth arguing plus the Specs page that was referenced directly in the thread seems clear that it is a recommendation.

I don't understand what olive branch is needed here, unfortunately if Macs aren't supporting Nvidia GPUs then captures will take longer on Macs.