Faible performance avec les fichier raw du Sony A6700

,

Je possède plusieurs appareils Sony, dont un A6700 (26Mpx) et un A6600 (24Mpx), mon problème c’est que le traitement des fichiers Raw est considérablement plus long avec les Raw du a6700, un Raw du A6600 se traite en 5 à 7 secondes tandis que ceux du A6700 prennent entre 25 et 30 secondes chacun pour la création d’un fichier DNG! J’ai une carte Nvidia 4060 qui performe normalement très bien. Note: Pour bien sûr des traitements similaires.

Exporting can take more or less time depending on how much detail and/or noise is in an image. To test, I exported with exactly the same settings (no correction) two images from the same camera body, but the files have different size. This is what I got:

The detailed image took 6 seconds, the image with less detail in 2 seconds.
Repeating the export, times were 5 seconds and 1 second respectively.

  • The detailed photo took 3-5 times as long as a less detailed photo.
  • add a few megapixels and some more customising and times will change, they get longer, but the difference can increase or decrease. Applying deepPRIME brought both images to 9 seconds in my little test.

:warning: Depending on what you do and how, results can vary.

Unless you need an DNG file for any reason. I would use a 16 bit TIFF file. After all, a DNG file is only a TIFF file in an Adobe DNG wrapper.

1 Like

This is an oversimplified conclusion of what the DNG specifications say.

Read this and find that

DNG is based on the TIFF/EP standard format, and mandates significant use of metadata. The specification of the file format is open and not subject to any intellectual property restrictions or patents.

“based on the TIFF/EP standard format” does not necessarily mean that a DNG is just a repackaged TIFF. One can build a lot of different things with LEGO :wink:

1 Like

Using RTX 4070/i7-14700KF/PL8.3.1, DNG exports of a6700 are about 10% slower than for a6600, as expected. However, PL uses less CPU/GPU for DNG exports than for JPEG or TIFF exports. It seems that DxO should tune CPU/GPU usage for DNG exports, like it did for JPEG and TIFFs recently. Don’t know if that’s possible, since they probably have to use Adobe DNG SDK, which might make it harder to control computing resource usage.

I did tests with dpreview studio scene raws (https://www.dpreview.com/reviews/image-comparison) taken at ISO 25k, original + 9 VCs, using almost all PL settings. Tested Sony a6600, a6700, Nikon Z6III. The results for Z6III were almost the same as for a6600, so I don’t give them here. Likewise, TIFF 16-bit exports took about the same time as JPEG exports, maybe 5% slower. JPEG exports had quality=90%, full resolution. Average processing time in seconds per photo (read from PL export status window) for various NR settings.

 Camera   Type   XD2s   DP    HQ   NoNR  (PRIME) 
+-------+------+-----+-----+-----+------+-------+
 a6600     DNG   2.7   2.3   3.2   7.2    (7.2)
 a6600    JPEG   2.0   1.0   1.1   0.9    (5.4) 
 a6700     DNG   3.2   2.7   3.6   7.2    (7.0)
 a6700    JPEG   2.5   1.2   1.2   1.1    (4.9)

Several values seem to be off but not more than by 20%.

Approximate CPU/GPU usage in percent read from Task Manager graphs, for various export types and NR settings (e.g. 25/87 means 25% CPU usage and 87% GPU usage on average):

  NR       JPEG     DNG
+-------+-------+-------+
 XD2s     25/87   25/60
 DP       55/40   30/15
 PRIME    75/0    60/0
 HQ       50/35   25/10
 NoNR     40/15   15/5

The “measurement” error in this case can be much greater than for processing times. Note that the proportions may be very different for other CPU/GPU combinations.

@MarioTR I tried to produce some figures for comparison on my Ryzen 5900X with an RTX 3060 (slower than your 4060) so I downloaded an A6600 image (which I cannot easily find!?) and an A6700 Image which was easy to find so I created a Batch of 30 images(1 [M]aster and 29VCs] and started using my standard export to 100% JPG and got this

Followed by a Dump!?

So I think I will return to installing some bits of hardware and leave testing to those better equipped!? I am running other releases of DxPL at the same time so memory might be getting a bit short and @Wlodek has published his results before me anyway !

With the other DxPL Versions shut I reran the test, still exporting to JPG not to DNG, and that worked this time.

and with a “random” A6600 image from the internet I got this

The final timings were

So about 2minutes 27 seconds for JPG and DNG for 30 A6700 images and 1 minute 54seconds for the 30 A6600 images, with the input on SATA (i.e. 1 image with 29VCs) and the output was to the same SATA SSD.

@MarioTR This is way out of line with respect to what you reported??

Perhaps you used DeepPRIME (XD2s) for a6600 exports but used no noise reduction for a6700 exports. Sorry for not making this guess explicit in my post.

I would guess that your CPU is much weaker than i7-14700KF. Also RTX 4060 GPU is about twice slower than 4070. Taking into account also my previous post, it would make sense, imo.