Wednesday, January 31, 2007

The Secrets of the K10D (Part 3 of 3) – The Brain: The IPU

The IPU (Image Processing Unit) used in the K10D has been a big secret since Pentax announced the K10D in September last year. The photos published officially by Pentax, in the K10D catalogue, Pentax’s official website for digital cameras and Dpreview.com are actually all the same one which shows nothing printed on the chip. Well, here “they” are:-
http://www.dpreview.com/reviews/PentaxK10D/Images/Features/prime.jpg
http://www.digital.pentax.co.jp/en/35mm/k10d/feature.html

As seen in the short descriptions about the PRIME in the 2nd URL above of the Pentax website for digital cameras, Pentax have actually talked nearly nothing about the features and specifications of their new IPU. It only tells that the IPU utilizes DDR2 RAM at a data transfer speed of 800MB/s (but the RAM is actually just the peripheral) and that it is built on a 90nm fabrication process.

While people had been comparing the specifications of the ADC (Analog to Digital Converter) and the IPU used in the K10D and guessed that they are the NuCore products, there was no further information which could actually *confirm* anything until Samsung revealed the secrets, finally, about one month ago, in their camera website about the GX-10 when the photos of the ADC and the IPU were first got exposed:-
http://www.samsungcamera.com/product/pro_view.asp?prol_uid=1923&cat_uid=62

It can be seen in one of the pictures in the above page that two NuCore NDX-2240 chips are being used as the ADC whereas it can be seen in the other picture that the IPU shows a “PENTAX” brand name marked on it with some codes and numbers but no ordinary people know anything about what those represent. Well, there is no “PRIME” printed on the chip anyway, see:-
http://manage.samsungcamera.com/upload/editor/200612/15_1_2006121218134791.gif

Anyway, now, it can be confirmed that the IPU is not probably the NuCore SiP-2290, which is the recommended IPU to pair with the NDX-2240 by NuCore.

So, now it seems the “PRIME” is just a pure marketing name which either refers to the IPU alone or collectively refers to the image processing hardware and software inside the K10D, as a whole. In contrast, Canon’s DIGIC is clearly the IPU chips of their DSLRs and DCs, of which the name “DIGIC” is printed on all of the Canon’s IPU chips, despite that it is also widely used for marketing purpose.

Nevertheless, in this article, I shall still try to give some technical talks about the role of an IPU and why a powerful IPU is crucial for a successful DSLR with good image quality, in a general sense, followed by my views on the issue of "soft jpegs" which was originated from Phil Askey's reviews on the *ist D, DS and K10D.

Okay, let’s first look at the NuCore SiP-2290 technical catalogue (again), for what an IPU is and what it can do:-
http://www.nucoretech.com/nu3/images/80_downloads/pb_sip2290.us.pdf

By looking at the functional block diagram on the page 2, the major essential functions of an IPU shall include: 1. those basic mathematical calculation functions for the mapping between different color models as well as; 2. for basic adjustment of parameters within each of those color models.

For example, for those color models they can be RGB (Red, Green, Blue), HIS (Hue, Intensity, Saturation), YC (Luminance and Chroma(nce), or called the S-Video in the videoland), YCrCb (which is similar to YC but the C is further divided into two “vectors” called Cr and Cb, i.e. the red and blue color (chroma) components; called (one of the) “Component Video” in the videoland) and etc. Actually, all those color models has their own value for use, from for the ease of data processing and manipulation to representation in real-life in practical sense, for the end-users.

Other functions of the NuCore IPU include data compression, SD card interface controller, CCD bad pixel re-mapping function (so that the dead pixels could be ruled out and the end user will never see them, the missed info is to be inserted by interpolated values from the neighbourhood) , analog video signal (format) conversion and so on. It's interesting to note that the NuCore IPU does provide a few useful and popular DC functions like the digital effect filter and the face detection algorithm.

Well, let's focus back on the core function of an IPU, i.e., image processing. In addition to that the IPU supports the conversion between color models, it may or may not support also for the adjustment of parameters’ values within each of the color models, which this task is generally referred as “processing” of the image. If the IPU does not have those image processng functions built-in, higher level software algorithms need to be embedded, in the firmware. To elaborate the “processing” a little bit more, it is just that raw numeric values, for each of the parameters, for every pixel are picked up and then the IPU (or the higher level algorithms) will have all these adjusted, by mathematical formulas and some logics behind, with the aim of achieving certain effects, and it will ultimately “re-insert” all the new values for each of the parameters per pixel into a temporary buffer for further processing or for storage, if it is the final product.

Practically, the high level algorithms are usually defined in the firmware for all those adjustments of the image aspects, like white balance, saturation, contrast and brightness adjustment and so on. The high level algorithms used by Pentax in the K10D are known to be the "SilkyPix", which is originally developed by the Ichikawa Soft Laboratory which marketed their RAW convertors with the name SilkyPix Developer Studio.

It is trivial that it is highly desirable that the IPU should complete all the required calculations as fast as possible. Just like the CPUs and GPUs (Graphics Processing Unit) of the computer industry, different IPUs with different design will perform differently, in terms of the variety of mathematical functions supported, the processing speed and the output quality or even power consumption and so on. But there are always ways to optimize the IPU to deliver a higher performance, e.g., better pipelining / parallel processing, faster clock speed and faster peripherals and buses like faster RAM and RAM bus, etc.

It should be noted that for the same hardware with the same processing power, there is always a conflict between the output image quality and the performance/speed in processing. Just like any computing hardware, if you need to complete the processing job faster, then the calculations are needed to be simplified. As a result, image quality will drop for such a rougher calculation. On the other hand, if there is no compromise made in the calculations and the software algorithms to be implemented in a precise way (well, provided that the algorithms are good and efficient enough in the first place), then the processing time will become longer and this would affect the continuous shooting rate, with this increase in the processing time for each of the frame.

Actually, the amount of image data to be manipulated is quite huge. Just see how long for our PC, of which the CPU is already quite powerful indeed, will need to convert a RAW file. Then you will see actually the on-board conversion and compression must have some compromises. Also, without the hardware IPU for all those, it is nearly impossible to complete the task in “real-time”, as the delay / waiting time will be too long to be acceptable, when the camera is used in the field.

Despite the specs of the Pentax IPU used the K10D are all unknown to we typical Pentax users, e.g., how many *hardware* calculation functions it supports and how fast or slow it processes things, actually the end results in the images produced can tell something. And, it’s actually what it should be counted afterall!

Even though I have examined quite some of those K10D samples over the net for months, I still have tried to be more objective. Well, I started a poll here. The pictures in the shootout were all taken by Phil Askey with the same target, with different current 10M budget DSLRs, namely, Sony A100, Nikon D80 and K10D, which are all using the same Sony APS-C sized CCD sensor. You can first try to play the game and see which one you would prefer and then have a look about the poll results.

The poll is now near the end (which will expire on 2.2.07 shortly) but you are still welcomed to make a vote overthere, as long as you contribute honestly according to the fair rules set.

Indeed, there is only one purpose for such a comparison (or something alike) and the poll results are useful for one thing, mainly: it is to find out how (well) different cameras, which is mostly about the IPUs plus the software conversion/compression algorithms, do the job. I suppose the lens factor is minimal as Phil Askey yet knew to choose the best 50mm standard lens for each of the brands to do the bench tests. So, optically, I won’t expect a huge difference.

Finally, I wish to point out about the (huge) mis-concept and/or simply un-substantiatable argument by certain Pentax fans that “the image quality is not inferior, it just produces softer jpegs which allow the users more room to do further post processing”.

But, IMHO, I’m afraid that this kind of statement is actually totally groundless.

First of all, if post processing is important, why bother to shoot JPEG? Why not shoot RAW in the first place if so? Do note that JPEG format is having only 8-bit per color per pixel which is far less than the typical 12-bit data of RAW. Indeed, the true value of the JPEG format IMHO is just to provide an option for direct out-of-the-camera “instant” product. As the main purpose of the JPEG mode is just for convenience and thus it should be highly preferable that it should look great when it is output.

Secondly, I must point out that “softness” is actually a totally different thing from “blur” or simply “bluriness”. A “soft” image means that it is less sharpened, as this is exactly what people who are defending always arguing about. Blur(iness) means that there is less picture/image information/data contained and that the image is not clearly defined and is somehow in a mess. This is merely a result of inferior and poor processing. The worst thing is that when there is poor processing, digital artifacts will often appear and come out altogether, e.g, obvious zigzag edges in JPEGs.

The truth is when the above happens, sharpening cannot help in any sense. Yes, sharpening can make the picture *looks* sharper, but the lost image data can never be retrieved in any way and an ugly duck can by no means becomes a princess, which can only happen in a fairy tale!

For anyone who is still unconvinced, one can feel free to visit my K100D full review page and download the original “soft” Jpeg image produced by my *ist DS and the corresponding Jpeg image direct out of my K100D and then apply sharpening/USM (or whatever you like including increasing the contrast and etc.) so as to see if you could do the miracle!

Last but not least, I must second the views of Phil Askey here this time as I believe he actually knew very well for how to draw up his conclusions in his reviews, which are widely publicized, and most importantly (to him), how to write his conclusions carefully with his excellent native English. BTW, I have nearly never been able to find anything he said in his review conclusions is obviously wrong. This also explains why he is so authoriative and is widely and highly trusted by the photo gear consumer community.

Nonetheless, I can see that Phil Askey have already been quite restrained in using his wordings and sentences and usually tried to give a leeway for each of his comments which can be considered as “negative”, frankly speaking. Indeed, it is not difficult to see these in his *ist D and K10D and other reviews. Well, by inspecting his “lab” test shots as well as all those real-life samples, I would say that quite some of the (K10D or *ist D) pictures are “blurry” when viewed in 100% or so, but Mr. Askey was still being very kind in just saying “soft” or “lack of edge sharpness”. If I was writing the reviews with Phi's test data, I would have used the word “blur” long time ago to represent things in an even more accurate and exact way. Well, to talk about more positive results, I think Phil is again right for his favourable conclusion on the image quality of K100D, which I have made up a similar conclusion in my K100D full review months before Phil published his report :-) BTW, as Phil Askey mentioned, I too scratch my head hard from time to time why Pentax has ever only done the thing right *once* with the K100D *only*!

So, when there are quite some die-hard blind brand loyalists at the DPR Pentax DSLR forum bashed and bash Phil Askey from time to time for what he writes (which are already in a restrained way), I think those guys should give a big thank to Phil instead, if they could really face the truth by doing more researches and thinking a little bit more objectively.

If unfortunately and yet fortunately you are one of those who are now reading my article to this point, what I can say is only: “Sorry, I just tell my best knowledge and express my truly humble opinions..” If *you*’re still not convinced, please download and inspect thoroughly again for the posted images in the “Compared To” sections of Phil Askey’s review reports. If you still can find something new and different with good evidence, any further discussion in a peaceful way will be much welcomed.

Read also:-
Dpreview puts Low Rating on the Image Quality of K10D

The Secrets of the K10D (Part 2 of 3) – The Bridge: The A to D Convertor

The Secrets of the K10D (Part 1 of 3) – The Heart: The CCD

Related Posts

 
Creative Commons License
RiceHigh's Pentax Blog by RiceHigh is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.