Gigapixel question

[PeterAit]

I am writing Topaz about this but thought someone here might know something. When I first started using the program, Task Manager would show my 8 logical cores being utilized, but nowhere near 100%, and my graphics processor going along at 100%. With the latest program update, all 8 cores are at 100% but the graphics unit at only 20%. And my subjective impression is that processing is slower. What gives?

[Bart_van_der_Wolf]

I am writing Topaz about this but thought someone here might know something. When I first started using the program, Task Manager would show my 8 logical cores being utilized, but nowhere near 100%, and my graphics processor going along at 100%. With the latest program update, all 8 cores are at 100% but the graphics unit at only 20%. And my subjective impression is that processing is slower. What gives?

Hi Peter,

Check the Advanced Settings section in the File>Preferences. On the first run, Gigapixel tries to determine which of the memory settings is the best. The user has the option to change those CPU/GPU settings. A few minor version upgrades ago, there has been an addition to the CPU mode, which specifically is aimed at Intel CPUs, called "Intel OpenVINO". That option uses highly optimized code libraries which run several times faster than before. How much faster depends on one's specific configuration. Rher are some users who report faster processing with this CPU mode, others still get faster results with GPU mode.

Cheers,
Bart

[julianv]

Other than doing benchmarks, with various combinations of preference settings, is there a way to determine which of the options are supported by my computer?  This is a Retina 5K iMac "Late 2014", with 4 GHz Intel Core i7.  I checked the processor spec search page at Intel and found no clues.

For Mac users, I'm guessing that the best choice of options might also be dependent on the OS version. I'm still running 10.12.6.

[Alan Goldhammer]

I have a simplistic question.  Now that Canon has updated the firmware for my Pro-1000 I can print panos.  If I want to maintain the best quality of 600ppi which is the upper setting for the printer, I have to up rez images.  In the past I simply let LR do this for me which is convenient as I don't have to convert the RAW file.  If I use Gigapixel AI I guess I have to export a TIFF file out of LR and then use Gigapixel AI to get to the desired PPI for my printer.  I assume that what I would export would be the final image that has been sharpened do the up resolution and then import back into LR for printing.  Is this the correct approach?  I watched the video on the Gigapixel site and they didn't cover this part of it.

[Bart_van_der_Wolf]

I have a simplistic question.  Now that Canon has updated the firmware for my Pro-1000 I can print panos.  If I want to maintain the best quality of 600ppi which is the upper setting for the printer, I have to up rez images.  In the past I simply let LR do this for me which is convenient as I don't have to convert the RAW file.  If I use Gigapixel AI I guess I have to export a TIFF file out of LR and then use Gigapixel AI to get to the desired PPI for my printer.  I assume that what I would export would be the final image that has been sharpened do the up resolution and then import back into LR for printing.  Is this the correct approach?  I watched the video on the Gigapixel site and they didn't cover this part of it.

Hi Alan,

Yes, while not as convenient as a direct print from LR:
1. TIFF output (from LR or similar)
2. Gigapixel AI rescaling to output dimensions in pixels for 600 PPI
3. Optionally, add after rescale sharpening
4. Import in LR for printing

Gigapixel AI can already modify the sharpening, but it might be marginally improved by using a dedicated sharpening run in step 3.

Note, when outputting from LR or a similar photo editor, be very careful with adding sharpening (especially with too large a radius), because Gigapixel may pick up very slight sharpening halos and turn them into sharper visible artifacts. Gigapixel doesn't need overly sharp input, it will create higher resolution upscaled results anyway. Also, using very low settings for noise suppression in Gigapixel will retain more subtle surface detail.

I've also noticed better image quality with a CPU enabled preset, compared to GPU enabled. It might be worthwhile to test that for your specific hardware.

Cheers,
Bart

[Bart_van_der_Wolf]

Other than doing benchmarks, with various combinations of preference settings, is there a way to determine which of the options are supported by my computer?  This is a Retina 5K iMac "Late 2014", with 4 GHz Intel Core i7.  I checked the processor spec search page at Intel and found no clues.

For Mac users, I'm guessing that the best choice of options might also be dependent on the OS version. I'm still running 10.12.6.

Hi Julian,

If your Processor is of a recent Intel CPU generation series, you could very well benefit from the OpenVINO settings. I'm using an older generation Intel CPU, but I still benefit from OpenVINO. Or if your Graphics card offers a recent GPU, that may be faster, but do check for image quality differences.

I'm afraid that only benchmarks on your specific hardware can tell the difference, although Gigapixel does try to find the best settings when rescaling the first image after installation.

Cheers,
Bart

[Alan Goldhammer]

Hi Alan,

Yes, while not as convenient as a direct print from LR:
1. TIFF output (from LR or similar)
2. Gigapixel AI rescaling to output dimensions in pixels for 600 PPI
3. Optionally, add after rescale sharpening
4. Import in LR for printing

Gigapixel AI can already modify the sharpening, but it might be marginally improved by using a dedicated sharpening run in step 3.

Note, when outputting from LR or a similar photo editor, be very careful with adding sharpening (especially with too large a radius), because Gigapixel may pick up very slight sharpening halos and turn them into sharper visible artifacts. Gigapixel doesn't need overly sharp input, it will create higher resolution upscaled results anyway. Also, using very low settings for noise suppression in Gigapixel will retain more subtle surface detail.

I've also noticed better image quality with a CPU enabled preset, compared to GPU enabled. It might be worthwhile to test that for your specific hardware.

Cheers,
Bart

Thanks.  I'll give this a try.  I will need to set up CPU enabled as I still have a rather lightweight GPU with only 2GB RAM (I've been meaning to upgrade this and maybe this will be the impetus).

[PeterAit]

Thanks.  I'll give this a try.  I will need to set up CPU enabled as I still have a rather lightweight GPU with only 2GB RAM (I've been meaning to upgrade this and maybe this will be the impetus).

FWIW, you can get great graphics cards on ebay, cheap. The gaming nebbishes always want the latest and greatest, so their 1-generation-old cards flood the market. I got a 20GB NVidea card that was recommended by Topaz for ~$100.

[Garnick]

FWIW, you can get great graphics cards on ebay, cheap. The gaming nebbishes always want the latest and greatest, so their 1-generation-old cards flood the market. I got a 20GB NVidea card that was recommended by Topaz for ~$100.

Hi Peter,

That sounds interesting, but please keep us in touch concerning the GPU performance and life cycle.  Just wondering if the intense gaming tends to perhaps have any negative effects on the GPU.  Of course for $100 it really isn't much of a gamble.  Also, which platform, Windows or Mac and which OS if you wouldn't mind sharing that information.

Gary 

[MichaelEzra]

I got a 20GB NVidia card that was recommended by Topaz for ~$100.

I am curious, which card is that?

[Alan Goldhammer]

FWIW, you can get great graphics cards on ebay, cheap. The gaming nebbishes always want the latest and greatest, so their 1-generation-old cards flood the market. I got a 20GB NVidea card that was recommended by Topaz for ~$100.

Which card is this?  I just looked on Newegg and looking at either workstation or gaming cards by NVidia I only see 16 & 24GB cards.  Given that these cost around $2K new, $100 seems like a rather strange price.  New NVidia GTX 1660 with 6GB RAM run about $250 these days which is almost 1/2 the entry level price that they were released at.  That's the one I'm looking at.

[PeterAit]

It's the NVidea GeForce GTX 960.

[MichaelEzra]

It's the NVidea GeForce GTX 960.

That would be 2GB, not 20GB

[Alan Goldhammer]

It's the NVidea GeForce GTX 960.

That's the card I have in my current PC.  As noted it's 2GB

[PeterAit]

Perhaps I am reading this wrong, but dxdiag seems to be telling me 20GB. Or am I supposed to look at the display memory figure, which in any case says 4GB not 2.

[Alan Goldhammer]

Perhaps I am reading this wrong, but dxdiag seems to be telling me 20GB. Or am I supposed to look at the display memory figure, which in any case says 4GB not 2.

Maybe there was a manufacturer who increased the RAM.  The NVidia specifications state 2GB which is what shows up when I run specs on my PC.

[faberryman]

If you Google the video card, it takes you to the NVidia site, which list the memory specs at 4GB.

[Alan Goldhammer]

If you Google the video card, it takes you to the NVidia site, which list the memory specs at 4GB.

The 980 is 4GB, the 960 is 2GB according to the website.

[Terry_Kennedy]

Perhaps I am reading this wrong, but dxdiag seems to be telling me 20GB. Or am I supposed to look at the display memory figure, which in any case says 4GB not 2.

"Display memory" is on the graphics card. Shared memory is host memory used for shuffling things back and forth between the host system and the graphics card. Most modern graphics cards can only compute with their on-board display memory. The shared memory is used for overlays, textures, and prepping data to be copied to the video card's video memory.

I'd recommend TechPowerUp GPU-Z as it will tell you exactly what your card has and is capable of. Here's mine (a not particularly high-end card for a new system I'm building):



If GPU-Z says your card has 4GB, it has 4GB. It may be a variant sold by one of the companies that create a number of variant models out of a single reference design.

[Timur_Born]

Gigapixel makes less use of both system and GPU memory than you might think. Here is memory usage of an image being upsampled to 25200 px via GPU, with allowed GPU memory consumption set to "High".

Out of my 8 gb GPU RAM, only 55% are used (about 4.5 gb). Out of my 16 gb system RAM only 2.5 gb are used during processing. Only at the end of processing (100%) does system RAM usage shoot up to about 12 gb for maybe 2 seconds and then drops down again while the destination file is compressed (if compression like TIF ZIP or PNG is used).