Steps & Settings for use of Whisper in Subtitle Edit ?

[Seeker47]

I hadn't found any good guides to follow on this so far, so I've just been stumbling my way through the menus, trying to queue things up properly to get a satisfactory result. Many failures at this along the way, using SE 3.6. Kept going in circles, without anything to show for it. The setup and attempt at getting some processing going clearly was not right and not complete. But I downloaded v. 4.0 to give this a further try. Apparently there have been significant improvements from that last version to this. I'm hoping that I've managed to stumble into that first decent result, will know when I try to watch the movie with the generated .SRT file. (A visual scan of the file looks fairly promising.)

I was using the new "Faster Whisper" module. It took around 8 hours to process a 90 minute film, on a Win 10 x64 with i5 and 16 GB of Ram. The computer was not doing anything else for most of that time. It was kind of laughable when the processing said that it was up to minute 51, with "a few more seconds to go." A minor point -- I can live with that, so long as it works.

One Settings thing that eluded me: I chose the "Large Model" (2.xx GB), without any real understanding of what that actually meant, but the .MKV file being processed was over 2GB in size, so I figured that I might as well choose the max option. If someone knows of a simplified explanation for all this setup, please point me in that direction. Also, when might one be better off going with the Vosk option instead ? (I happen to have the standalone Deepl as well, but have not tried using it yet. Not sure why they have Google Translate included in the Setup screens. I thought it had major text quantity limitations being enforced.)

[Subtitles]

You didn't mention if you have a GPU or not.
If you are relying on the CPU to do the job, then i5 will take a very long time. Using a GPU and letting Whisper use it instead of the CPU will speed things up.
Using the large model is not going to work unless you have a GPU with at least 12GB VRAM.
Models in Whisper controls the accuracy of the transcription
Please visit this website to get bbasic information
https://github.com/openai/whisper
In the Available models and languages you can see a list of the models and the VRAM needed to work for each model
I don't use Subtitle Edit it is just a nice GUI but use Whisper AI Command Line (uses Python) is much more reliable and flexible for my needs and of course enabled the GPU to do the transcription.
I would advise to stick to Whisper and NOT VOSK. Download all the other models starting with tiny and test each model. The accuracy isn't that great but gets better with medium (needs 8GB VRAM).
I am attaching the lyrics for a song created with different models. You can see the differences.
I hope that clarifies few things for you.

[VoodooFX]

Using the large model is not going to work unless you have a GPU with at least 12GB VRAM.
https://github.com/openai/whisper

Why use Whisper OpenAI when Faster-Whisper is much faster & better, and you can run large model on 4GB VRAM.

[Subtitles]

Using the large model is not going to work unless you have a GPU with at least 12GB VRAM.
https://github.com/openai/whisper

Why use Whisper OpenAI when Faster-Whisper is much faster & better, and you can run large model on 4GB VRAM.

I have been using Whisper AI for almost a year now and got used to it I guess. Not really interested in anything else ATM.
But it is good to know about the advantage of using the model large on 4GB VRAM. Might give it a try in the future.

[VoodooFX]

I have been using Whisper AI for almost a year now and got used to it I guess. Not really interested in anything else ATM.
But it is good to know about the advantage of using the model large on 4GB VRAM. Might give it a try in the future.

Standalone Faster-Whisper has same named args, you would need just to change exe name and have faster and more accurate transcriptions without any effort.

[VoodooFX]

I was using the new "Faster Whisper" module. It took around 8 hours to process a 90 minute film, on a Win 10 x64 with i5 and 16 GB of Ram.

What model is your CPU? Test if "--compute_type=float32" is faster, for me it's twice faster than default "int8_float32".

It was kind of laughable when the processing said that it was up to minute 51, with "a few more seconds to go."

Progress in SE is not implemented. To see progress stats you need to use it in console without SE.

One Settings thing that eluded me: I chose the "Large Model" (2.xx GB), without any real understanding of what that actually meant, but the .MKV file being processed was over 2GB in size

Bigger model size - more accurate and slower it's. Accuracy difference between medium and large is pretty small.
Then you can try "--beam_size=1", it should be faster with insignificant hit on accuracy.

[Seeker47]

Using the large model is not going to work unless you have a GPU with at least 12GB VRAM.
https://github.com/openai/whisper

Why use Whisper OpenAI when Faster-Whisper is much faster & better, and you can run large model on 4GB VRAM.

Thanks to all for your replies. Some further notes re my original post, above. As it happens, I did indeed use Faster Whisper. I don't know whether the rig I'm using includes a GPU, would have to check System Information on that. BUT, it worked. I watched that documentary (which I had downloaded several years ago) last evening. I don't understand German, and so cannot entirely vouch for the outcome, but nearly all of it seemed to be there, and Whisper certainly caught at least the gist of it.

The audio track was clean and well recorded, so that must have helped. I'd say the EN subs synch was pretty good. Rapid speech by some individuals at various times was coped with. (Even when you're a native speaker of a language, that can pose a problem for the listener to follow.) Whisper translated the lyrics of some songs on the soundtrack (again, they were being clearly rendered in the audio), but not others which were heard equally well. I'd been advised that background sounds could interfere with the results, but that was not a factor in this case.

One thing which was kind of annoying and which I would mention to the developer(s) -- if I knew where to post it (Github ?) -- was that once Whisper starts processing you apparently cannot minimize SE. Since I didn't want to leave up a static SE screen blocking out everything else, I had to turn off the monitor. Not that big a deal though if I'm leaving anyway, to tend to other things while Whisper works its magic.

[Seeker47]

[QUOTE=VoodooFX;2705290]

One Settings thing that eluded me: I chose the "Large Model" (2.xx GB), without any real understanding of what that actually meant, but the .MKV file being processed was over 2GB in size

Bigger model size - more accurate and slower it's. Accuracy difference between medium and large is pretty small.
Then you can try "--beam_size=1", it should be faster with insignificant hit on accuracy.

I think I would opt in favor of accuracy, in just about any case that I could imagine.

[Subtitles]

Thanks for the update.
Yes it is very important to have a GPU. A 4GB VRAM can make a big difference with the accuracy and speed of transcription.
I usually let Whisper AI does transcription instead of tranalation in any language. This way I can check if the audio matches the text.
Then I use Subtitle Edit to translate to English or any other reliable translator.
If the clip isn't long, then I might try to get it to translate directly by running it again, and compare the results with other translators. From my experience, Whisper isn't that accurate with translation, but that might be just my case.

[Seeker47]

You didn't mention if you have a GPU or not.

I don't see one explicitly mentioned under Device Manager. Presumably it would be listed there under 'Processors' ?

"Intel Core i5-3550 @ 3.30 Ghz." Under display adapters it shows "Intel HD Graphics", which I take to be integrated graphics. I'd have to open it up to be sure, but I don't believe this one has a separate graphics card, nor did I see any in Device Mgr. I have at least one i7 system on hand here also -- which is sorely in need of a lot of software updates -- but that should be interesting later on for processing time comparisons. Not sure about whether that one had a GPU either.

[VoodooFX]

I don't know whether the rig I'm using includes a GPU.

Standalone Faster-Whisper [aka "Purfview's"] checks for GPU (with CUDA) and runs on GPU automatically otherwise it runs on CPU, if it chose to run on CPU then you don't have GPU with CUDA.

One thing which was kind of annoying and which I would mention to the developer(s) -- if I knew where to post it (Github ?) -- was that once Whisper starts processing you apparently cannot minimize SE.

SE issues you can report there -> https://github.com/SubtitleEdit/subtitleedit/issues

I think I would opt in favor of accuracy, in just about any case that I could imagine.

For German probably that would be ~0.1% more accurate when it's twice slower.
Btw, sometimes medium model can be more accurate than large.

[Seeker47]

.
I usually let Whisper AI does transcription instead of tranalation in any language. This way I can check if the audio matches the text.
Then I use Subtitle Edit to translate to English or any other reliable translator.
If the clip isn't long, then I might try to get it to translate directly by running it again, and compare the results with other translators. From my experience, Whisper isn't that accurate with translation, but that might be just my case.

A poll for those who have been using this for awhile: which languages seem to fare best and which the worst, if the desired end result is an EN .SRT file ? (I can make some wild guesses . . . . )

My next few trials will likely be for French language. I've collected a few for which no subs have ever been available, the next one on a film from 1974. If something never came out on VHS or DVD and was not much seen outside of the home country, that must account for a lot of these.

[VoodooFX]

Word error rate with large-v2 model:

[Subtitles]

Looks like you don't have a GPU.
A 90 minute video (audio) takes me about 30-40 minutes using a GPU of 8GB VRAM.
You don't need to open the PC. I use BelArc Advisor to give me a list of all the hardware and software on my computers. It even shows the licenses for each software including the Operating System. The free version is good and doesn't expire (until they become wiser and start asking for payment).
https://www.belarc.com/products/belarc-advisor

[Subtitles]

.
I usually let Whisper AI does transcription instead of tranalation in any language. This way I can check if the audio matches the text.
Then I use Subtitle Edit to translate to English or any other reliable translator.
If the clip isn't long, then I might try to get it to translate directly by running it again, and compare the results with other translators. From my experience, Whisper isn't that accurate with translation, but that might be just my case.

A poll for those who have been using this for awhile: which languages seem to fare best and which the worst, if the desired end result is an EN .SRT file ? (I can make some wild guesses . . . . )

My next few trials will likely be for French language. I've collected a few for which no subs have ever been available, the next one on a film from 1974. If something never came out on VHS or DVD and was not much seen outside of the home country, that must account for a lot of these.

I have used Whisper to transcribe several languages, even long operas of about 3 hours in German, Italian, French and Japanese movies. It is difficult to give you an answer but as you guess (just guessing) the English is the most accurate because the developers have put more efforts than other languages.
All in all, Whisper is a blessing to this world and it is all free. Unfortunately it has put thousands of transcribers and translators out of business in just one year. Hope they find something else to make them earn a living.

[VoodooFX]

...the English is the most accurate because the developers have put more efforts than other languages.

From the chart it's obvious that Spanish is most accurate.

[Subtitles]

...the English is the most accurate because the developers have put more efforts than other languages.

From the chart it's obvious that Spanish is most accurate.

Oh well. I am surprised to be honest.
Do you have any idea why?

[Seeker47]

Word error rate with large-v2 model:

Thanks for that chart. My guess featured those languages that are written Right to Left, or Up & Down (e.g., Kanji & Chinese), but should also have encompassed relative rarity, in terms of how many speakers vs. world population, and a country's relative "reach" for films / tv made in that language. (Yes, the written form of the language is probably not a logical indicator, but just what sprang to my mind.) Surprised that Japanese lang. did as well as it did, and that Chinese came in around the middle of the list.

[VoodooFX]

Do you have any idea why?

Here some answers:

I think it's helping that the orthography of Spanish uses more direct phonetical transcriptions compared to English, despite having a smaller amount of training data.

We also observed that many of the largest outliers in terms of worse-than-expected performance according to this trend are languages that have unique scripts and are more distantly related to the Indo-European languages making up the majority of the training datasets, such as Hebrew (HE), Telugu (TE), Chinese (ZH), and Korean (KO). These differences could be due to a lack of transfer due to linguistic distance, our byte-level BPE tokenizer being a poor match for these languages, or variations in data quality.

[Seeker47]

Looks like you don't have a GPU.
A 90 minute video (audio) takes me about 30-40 minutes using a GPU of 8GB VRAM.
You don't need to open the PC. I use BelArc Advisor to give me a list of all the hardware and software on my computers. It even shows the licenses for each software including the Operating System. The free version is good and doesn't expire (until they become wiser and start asking for payment).
https://www.belarc.com/products/belarc-advisor

Am I correct in a couple assumptions: 1) That there are some CPUs that include a built-in GPU onboard. & 2) That if the computer has an added graphics card, this must have its own GPU ? But having 8 GB of RAM on the video card sounds pretty high end from what I recall. So it will be much more expensive and generate a lot more heat. (?) If anyone has strong recommendations for such video cards so equipped -- that manage to evade those last two points -- I'd like to hear about it. The last time I paid any real attention to this subject, I think lordsmurf was opining that there were (by that point) commonly some CPUs with a TI (?) 1050 onboard that would deliver all the video performance that the vast majority of us -- running various video apps -- were likely to need. Hope I'm not misremembering or mis-paraphrasing him there.

Oh, almost forgot to add: No "thick" video cards that must take up TWO expansion slots. None of my present computers will accommodate those.

[Seeker47]

[QUOTE=Subtitles;2705339][QUOTE=Seeker47;2705336]

.

All in all, Whisper is a blessing to this world and it is all free. Unfortunately it has put thousands of transcribers and translators out of business in just one year. Hope they find something else to make them earn a living.

AI and other computerized processes are going to put a large swath of former job descriptions into the dumpster. We still don't have fully autonomous, fully capable in all real world situations, driverless vehicles -- as seen in the 2006 Schwartzenegger movie"The Sixth Day" -- but it is coming, eventually. When it does, say goodbye to taxis, Uber, Lyft, and maybe even trucking jobs. This hasn't even begun to be thought through by politicos around the world. People need to make a living. What overriding structure can there be that's viable, for future societies ? Will majority portions of the population have to be on some version of the dole ? And where can the financing for that come from ? Questions the next generation may have to answer.

[Subtitles]

Looks like you don't have a GPU.
A 90 minute video (audio) takes me about 30-40 minutes using a GPU of 8GB VRAM.
You don't need to open the PC. I use BelArc Advisor to give me a list of all the hardware and software on my computers. It even shows the licenses for each software including the Operating System. The free version is good and doesn't expire (until they become wiser and start asking for payment).
https://www.belarc.com/products/belarc-advisor

Am I correct in a couple assumptions: 1) That there are some CPUs that include a built-in GPU onboard. & 2) That if the computer has an added graphics card, this must have its own GPU ? But having 8 GB of RAM on the video card sounds pretty high end from what I recall. So it will be much more expensive and generate a lot more heat. (?) If anyone has strong recommendations for such video cards so equipped -- that manage to evade those last two points -- I'd like to hear about it. The last time I paid any real attention to this subject, I think lordsmurf was opining that there were (by that point) commonly some CPUs with a TI (?) 1050 onboard that would deliver all the video performance that the vast majority of us -- running various video apps -- were likely to need. Hope I'm not misremembering or mis-paraphrasing him there.

Oh, almost forgot to add: No "thick" video cards that must take up TWO expansion slots. None of my present computers will accommodate those.

Last year my PC had 2GB VRAM GPU Asus nvidia GEFORCE GT 1030 which only needs one expansion slot. Running Whisper AI wasn't enough to use the model medium so I replaced it with Asus GEFORCE RTX 3050 with 8GB VRAM but needed two expansion slots which I had.
So if you have at least one free expansion slot, then perhaps you can add 2GB GPU, but check your PSU rating first.

Your on board graphic card is obviously helping with the transcription and there must be a way to measure how much VRAM it has.
If you can't put an external GPU in your computer, then assuming you have a Google account and a Google Drive, perhaps you can try to use Colaboratory Jupyter Notebook.
I haven't used it myself because my Google Drive doesn't work, but you can find a lot of information with google search for Colaboratory Jupyter Notebook or Colaboratory Jupyter Notebook GPU. It is supposed to have a free GPU but they don't specify the size.
There is a paid option as well if you need more memory.

[Subtitles]

Whisper AI doesn't handle multilanguage speakers well. For example a movie with an American actor and a French actress you have to choose only one language and the transcription will go well for the English speaker but when the French actress speaks in French, the transcription will not be accurate. Same with language learning courses like Berlitz that has mixed languages you have to run Whisper twice for each language and then do so tedious editting to get everything correct.

[Seeker47]

Am I correct in a couple assumptions: 1) That there are some CPUs that include a built-in GPU onboard. & 2) That if the computer has an added graphics card, this must have its own GPU ? But having 8 GB of RAM on the video card sounds pretty high end from what I recall. So it will be much more expensive and generate a lot more heat. (?) If anyone has strong recommendations for such video cards so equipped -- that manage to evade those last two points -- I'd like to hear about it. The last time I paid any real attention to this subject, I think lordsmurf was opining that there were (by that point) commonly some CPUs with a TI (?) 1050 onboard that would deliver all the video performance that the vast majority of us -- running various video apps -- were likely to need. Hope I'm not misremembering or mis-paraphrasing him there.

Oh, almost forgot to add: No "thick" video cards that must take up TWO expansion slots. None of my present computers will accommodate those.

Last year my PC had 2GB VRAM GPU Asus nvidia GEFORCE GT 1030 which only needs one expansion slot. Running Whisper AI wasn't enough to use the model medium so I replaced it with Asus GEFORCE RTX 3050 with 8GB VRAM but needed two expansion slots which I had.
So if you have at least one free expansion slot, then perhaps you can add 2GB GPU, but check your PSU rating first.
[/QUOTE]

PSU rating ?? Whazzat ?

And what about the graphics chip onboard a CPU, from post #20 ?

[Subtitles]

Am I correct in a couple assumptions: 1) That there are some CPUs that include a built-in GPU onboard. & 2) That if the computer has an added graphics card, this must have its own GPU ? But having 8 GB of RAM on the video card sounds pretty high end from what I recall. So it will be much more expensive and generate a lot more heat. (?) If anyone has strong recommendations for such video cards so equipped -- that manage to evade those last two points -- I'd like to hear about it. The last time I paid any real attention to this subject, I think lordsmurf was opining that there were (by that point) commonly some CPUs with a TI (?) 1050 onboard that would deliver all the video performance that the vast majority of us -- running various video apps -- were likely to need. Hope I'm not misremembering or mis-paraphrasing him there.

Oh, almost forgot to add: No "thick" video cards that must take up TWO expansion slots. None of my present computers will accommodate those.

Last year my PC had 2GB VRAM GPU Asus nvidia GEFORCE GT 1030 which only needs one expansion slot. Running Whisper AI wasn't enough to use the model medium so I replaced it with Asus GEFORCE RTX 3050 with 8GB VRAM but needed two expansion slots which I had.
So if you have at least one free expansion slot, then perhaps you can add 2GB GPU, but check your PSU rating first.

PSU rating ?? Whazzat ?

And what about the graphics chip onboard a CPU, from post #20 ?[/QUOTE]

What is the power supply rating of your computer? (PSU=Power Supply Unit) and if the GPU power rating is within the PSU capability.
No idea about graphic chip onboard CPU. Changing the CPU is going to cost you a lot more than a standalone GPU and I doubt if it will fit in your present motherboard.

[Seeker47]

What is the power supply rating of your computer? (PSU=Power Supply Unit) and if the GPU power rating is within the PSU capability.
No idea about graphic chip onboard CPU. Changing the CPU is going to cost you a lot more than a standalone GPU and I doubt if it will fit in your present motherboard.

Belarc reports nothing in regard to PSU. I popped the hood and saw that I have a 300W power supply. (80 Plus Bronze Certified, if that means anything.) How that equates to your 'PSU value' I just don't know. I believe there might be optional 450 or 500 Watt power supplies that this computer (and its primary sibling) could optionally accept. Because these are Shuttle XPCs, parts such as these are proprietary, and not standard off-the-shelf. So there are certain limitations. The cost of switching to a different -- possibly hybrid -- CPU is less of a barrier to me than having it be on the approved 'compatible list' for the computer. I'll also look into the possible options for graphics cards. If one wouldn't make a significant difference in shortening the file processing time, it wouldn't be worth the bother.

[loninappleton]

[QUOTE=Seeker47;2705371][QUOTE=Subtitles;2705339]

.

All in all, Whisper is a blessing to this world and it is all free. Unfortunately it has put thousands of transcribers and translators out of business in just one year. Hope they find something else to make them earn a living.

AI and other computerized processes are going to put a large swath of former job descriptions into the dumpster. We still don't have fully autonomous, fully capable in all real world situations, driverless vehicles -- as seen in the 2006 Schwartzenegger movie"The Sixth Day" -- but it is coming, eventually. When it does, say goodbye to taxis, Uber, Lyft, and maybe even trucking jobs. This hasn't even begun to be thought through by politicos around the world. People need to make a living. What overriding structure can there be that's viable, for future societies ? Will majority portions of the population have to be on some version of the dole ? And where can the financing for that come from ? Questions the next generation may have to answer.

I shook my head at mention of putting translators and transcribers out of work. I shook my in the negative. From what I've seen of the various programs here on videohelp and a couple others, the full job of subtitles and HI descriptions accurately synched to the video happenings is pretty far off. An AI doesn't hear a doorbell ring for a mundane example. Just doing a few subs for enjoyement I've learned that real subtitling is closer to -- how to say -- composition, catching the speech rhythms etc.

[Seeker47]

[QUOTE=loninappleton;2705832][QUOTE=Seeker47;2705371]

.

All in all, Whisper is a blessing to this world and it is all free. Unfortunately it has put thousands of transcribers and translators out of business in just one year. Hope they find something else to make them earn a living.

AI and other computerized processes are going to put a large swath of former job descriptions into the dumpster. We still don't have fully autonomous, fully capable in all real world situations, driverless vehicles -- as seen in the 2006 Schwartzenegger movie"The Sixth Day" -- but it is coming, eventually. When it does, say goodbye to taxis, Uber, Lyft, and maybe even trucking jobs. This hasn't even begun to be thought through by politicos around the world. People need to make a living. What overriding structure can there be that's viable, for future societies ? Will majority portions of the population have to be on some version of the dole ? And where can the financing for that come from ? Questions the next generation may have to answer.

I shook my head at mention of putting translators and transcribers out of work. I shook my in the negative. From what I've seen of the various programs here on videohelp and a couple others, the full job of subtitles and HI descriptions accurately synched to the video happenings is pretty far off. An AI doesn't hear a doorbell ring for a mundane example. Just doing a few subs for enjoyement I've learned that real subtitling is closer to -- how to say -- composition, catching the speech rhythms etc.

I get what you are saying, and 'Yes', it's clearly better to have this done the way that -- say -- Criterion or Severin would do it. But on a practical level, I think this runs into the old "Don't let the Perfect be the enemy of the Good."

[Subtitles]

What is the power supply rating of your computer? (PSU=Power Supply Unit) and if the GPU power rating is within the PSU capability.
No idea about graphic chip onboard CPU. Changing the CPU is going to cost you a lot more than a standalone GPU and I doubt if it will fit in your present motherboard.

Belarc reports nothing in regard to PSU. I popped the hood and saw that I have a 300W power supply. (80 Plus Bronze Certified, if that means anything.) How that equates to your 'PSU value' I just don't know. I believe there might be optional 450 or 500 Watt power supplies that this computer (and its primary sibling) could optionally accept. Because these are Shuttle XPCs, parts such as these are proprietary, and not standard off-the-shelf. So there are certain limitations. The cost of switching to a different -- possibly hybrid -- CPU is less of a barrier to me than having it be on the approved 'compatible list' for the computer. I'll also look into the possible options for graphics cards. If one wouldn't make a significant difference in shortening the file processing time, it wouldn't be worth the bother.

PSU with 300 Watts is what I had in my Windows XT PC. I think adding a GPU might be asking a bit too much. Replacing a PSU just for this computer is not worth it.
A GPU with 2GB VRAM power consumption is about 30 Watts. It is up to you to decide, but if it took you 8 hours to transcribe a 90 minutes video then this is simpy not the way to go.

[Seeker47]

What is the power supply rating of your computer? (PSU=Power Supply Unit) and if the GPU power rating is within the PSU capability.
No idea about graphic chip onboard CPU. Changing the CPU is going to cost you a lot more than a standalone GPU and I doubt if it will fit in your present motherboard.

Belarc reports nothing in regard to PSU. I popped the hood and saw that I have a 300W power supply. (80 Plus Bronze Certified, if that means anything.) How that equates to your 'PSU value' I just don't know. I believe there might be optional 450 or 500 Watt power supplies that this computer (and its primary sibling) could optionally accept. Because these are Shuttle XPCs, parts such as these are proprietary, and not standard off-the-shelf. So there are certain limitations. The cost of switching to a different -- possibly hybrid -- CPU is less of a barrier to me than having it be on the approved 'compatible list' for the computer. I'll also look into the possible options for graphics cards. If one wouldn't make a significant difference in shortening the file processing time, it wouldn't be worth the bother.

PSU with 300 Watts is what I had in my Windows XT PC. I think adding a GPU might be asking a bit too much. Replacing a PSU just for this computer is not worth it.
A GPU with 2GB VRAM power consumption is about 30 Watts. It is up to you to decide, but if it took you 8 hours to transcribe a 90 minutes video then this is simply not the way to go.

Well then, it looks like I may just be screwed. I looked over the "Approved" graphics card list for my existing hardware, and the top end of it was something like the Nvidia GTX780 with only 2GB of VRAM. That card is dual slot, whereas my small form factor computers seem to offer just a single slot, and the "Minimum System Power Requirement" is listed as 600 W., whereas the max. power supply for this hardware is 500 W. I want to continue using this particular hardware, for a lot of reasons that I won't go into here. 2GB of VRAM probably wouldn't make the desired difference, anyway. (Highly rated current issue goodies like the ASUS GeForce RTX 3070 Noctua Edition 8GB apparently wants 4 slots and an extra large case ! I'm overlooking the stratospheric price, for the moment.)

The exorbitant amount of processing time that I'm looking at now may just be the price of admission for being able to do this at all. I suppose it's not that terrible a deal if I set one of these jobs to run overnight, while I'm sleeping. Trade-offs . . . .