Recommeded memory speed for supercompact?

[GT]

My P&S camera when saving RAW + Full size JPG comes to 1.1 frames per
second,
NON-stop until the card is full. If I had access to faster cards (I'm
already
using the fastest available) that would greatly speed up non-stop
continuous
shooting. I can verify this by running an internal script (in the camera)
to
benchmark test any card in it.The bottleneck IS the cards in the newer and
faster P&S cameras.

You are basing your DSLR speed on the in-camera buffer until it has to
pause to
dump it to the card (the pause reducing the overall speed to your slowest
norm
when taken into account for total shooting session). After your in-camera
buffer
is full then your DSLR becomes much slower than my P&S.

You haven't provided enough information to say that the DLSR or the P&S is
faster.
That would depend on how many MPixels the P&S is saving at 1.1 frames per
second - you didn't say! As I said, the 350d (once the buffer is full) drops
to around 1 to 2 x 8MPixel frames pers second with my Kingston Elite Pro
card. I can't be bothered to count the time accurately as it really doesn't
matter!

This equates to something between 8-16MB/s write.

You need to do 1.1 * number of MPixels to see how fast your camera is
writing to the card. Then conpare the figure with 8 to 16 MB/s.

A faster card will not help your DSLR, as others have verified. In a DSLR
the
camera is the weakest link for write-to-card speeds, using a faster card
doesn't
change their burst shooting rate. In my P&S its burst-rate performance
increases
equally to the speed of the card.

No its not the 'camera' its the component that transfers memory from the
buffer to the card. The 'camera' is the whole device, which containes the
card reader (and therefore card).

 

[Richard Polhill]

You haven't provided enough information to say that the DLSR or the P&S is
faster.
That would depend on how many MPixels the P&S is saving at 1.1 frames per
second - you didn't say! As I said, the 350d (once the buffer is full) drops
to around 1 to 2 x 8MPixel frames pers second with my Kingston Elite Pro
card. I can't be bothered to count the time accurately as it really doesn't
matter!

This equates to something between 8-16MB/s write.

I doubt it. What is the size of each image (in MB)?

You need to do 1.1 * number of MPixels to see how fast your camera is
writing to the card. Then conpare the figure with 8 to 16 MB/s.

No that gives the number of megapixels per second. A truly arbitrary measurement.

MPx size <> file size.

The test is actually 1.1 * number of MB per file.

No its not the 'camera' its the component that transfers memory from the
buffer to the card. The 'camera' is the whole device, which containes the
card reader (and therefore card).

Where do you think the relevant buffer resides? Between sensor and processor
or processor and storage?

 

[GT]

I doubt it. What is the size of each image (in MB)?

No that gives the number of megapixels per second. A truly arbitrary
measurement.

MPx size <> file size.

The test is actually 1.1 * number of MB per file.

A fair point. Smallest filesize is 6.63MB, largest is 7.26MB. Lets take
6.9MB as average, so actualy throughput to card is something between 6.9MB/s
and 13.8MB/s.

Where do you think the relevant buffer resides? Between sensor and
processor or processor and storage?

Hadn't really considered this - does it matter if the slow component is the
device that performs the transfer between buffer and flash storage?

 

[Richard Polhill]

A fair point. Smallest filesize is 6.63MB, largest is 7.26MB. Lets take
6.9MB as average, so actualy throughput to card is something between 6.9MB/s
and 13.8MB/s.


Hadn't really considered this - does it matter if the slow component is the
device that performs the transfer between buffer and flash storage?

It does if the "device" is the image processor.

Basically, if the queue is INTO the processor and the files are being written
to card as fast as the processor can generate them then the bottleneck is the
processor.

If the queue is AFTER the processor because the card cannot be written as fast
as the processor generates files than the bottleneck is in the storage.

 

[GT]

It does if the "device" is the image processor.

Basically, if the queue is INTO the processor and the files are being
written to card as fast as the processor can generate them then the
bottleneck is the processor.

If the queue is AFTER the processor because the card cannot be written as
fast as the processor generates files than the bottleneck is in the
storage.

Well the image processor in the 350d is Canon's DIGIC II. I don't know how
the internals are configured - are you saying the heart of the camera - the
digital processor (DIGIC II) is the bottleneck?

 

[Richard Polhill]

It does if the "device" is the image processor.

Basically, if the queue is INTO the processor and the files are being
written to card as fast as the processor can generate them then the
bottleneck is the processor.

If the queue is AFTER the processor because the card cannot be written
as fast as the processor generates files than the bottleneck is in the
storage.

As an aside, I just thought I'd look at what a desktop PC can do.

I created a 3200x2133 pixel (6.8MPx) image containing just noise and saved it
as a 19.5MB bitmap (no compression) in about 6s.

I then saved a 771KB JPEG image in about 4.5s.

You'll note that the 25% saving in time is nothing like the 96% reduction in
file size.

The test would be better carried out converting RAW to JPEG but I have neither
RAW image nor converter to try.

Now there are those that believe that the dedicated image processing engines
in the cameras are much faster at processing images. This is utter tosh.
The primary aim of an in-camera processor is to convert RAW images within
tight power and heat efficiency requirements, not being the most performant
device.

The task profile of JPEG conversion is already well suited to a general
purpose processor with floating point processor such as the P4 chipset in my
desktop PC.

The processing requirements seem to be commonly underestimated. The delay is
all about getting the data into an image file. Writing it to the card happens
afterwards so the time is additive, it's just that the differences are
negligible compared to the overall process.

In a hypothetical system, we will assume that 15% of the time is taken getting
the RAW image from the sensor to the processor, 80% is taken converting and
compressing it, and then the final 5% is taken writing it to the card.

Yes you will see slightly greater throughput with a 20% faster card because
the final 5% of the overall time, the write part, is shortened by 20%. The
overall effect would only be a 1% improvement in the total process time.

 

[Richard Polhill]

Well the image processor in the 350d is Canon's DIGIC II. I don't know how
the internals are configured - are you saying the heart of the camera - the
digital processor (DIGIC II) is the bottleneck?

Yeah well that's where I'd expect it to be.

 

[GT]

[snip]

As an aside, I just thought I'd look at what a desktop PC can do.

I created a 3200x2133 pixel (6.8MPx) image containing just noise and saved
it as a 19.5MB bitmap (no compression) in about 6s.

I then saved a 771KB JPEG image in about 4.5s.

You'll note that the 25% saving in time is nothing like the 96% reduction
in file size.

The test would be better carried out converting RAW to JPEG but I have
neither RAW image nor converter to try.

Now there are those that believe that the dedicated image processing
engines in the cameras are much faster at processing images. This is utter
tosh.
The primary aim of an in-camera processor is to convert RAW images within
tight power and heat efficiency requirements, not being the most
performant device.

No, the dedicated image processing engines in cameras ARE much faster at
processing images than general purpose CPUs, because that is the specific
task that they are designed to do. They are also designed to do this using
as little power as possible! A 'PC' CPU is an all purpose Jack of all
trades, master of none. It is large in comparisson, draws significant power
and puts out lots of heat.

This is supported by the figures that you didn't have access to when you
made your sweeping statement:

8MPixel RAW image processed in camera at 3 frames per second into JPEG.
Somewhere along the line there is a bottleneck which slows this down to 1-2
frames per second, while taking more shots at the same time!. I doubt if the
bottleneck is the DIGIC II image processor as the image processor's job is
to process images, not store and retrieve chunks of member between buffer
and flash storage. This same image processor is found in the EOS20d, capable
(I think) of 8 frames per second!

On a Core 2 Duo, same image conversion takes over 4 seconds!

 

[GT]

Yeah well that's where I'd expect it to be.

Well guess again. The EOS 1d Mark 2 uses the DIGIC II and can process 8.5
frames per second at 8.m MPixels! So it is not the bottleneck!!

 

[GT]

If I get a modern super-compact digital camera such as a mid-range Canon
IXUS (US: Canon Powershot) for general use then ...

... what minimum speed of SD memory should I get to avoid delays?

I see "x133" mentioned a lot on memory adverts. Is x133 fast, slow,
just right for the sort of camera I am thing of?

I would probably get 1GB or 2 GB from this store:
http://www.jacobsdigital.co.uk/index.php?thequery=sd+memory

Look at 7dayshop.co.uk too - anything under £18 is VAT free!

 

[Richard Polhill]

Well guess again. The EOS 1d Mark 2 uses the DIGIC II and can process 8.5
frames per second at 8.m MPixels! So it is not the bottleneck!!

Ah so the EOS 1d Mk II uses faster cards then?

 

[GT]

Ah so the EOS 1d Mk II uses faster cards then?

If you say so - I don't have one so can't comment.

 

[Richard Polhill]

If you say so - I don't have one so can't comment.

The 1D Mk. II uses the same DIGIC II processor as the 350D, yet is capable of
almost 8x the speed of the 350D, so you contend that it is not the processor
that is the bottleneck, but the card.

What makes you think that the card is the bottleneck, when the same cards are
capable of an eightfold improvement in speed in a different camera?

Or are you saying that the cards used in the 1D are always 8x faster?

 

[GT]

Well the image processor in the 350d is Canon's DIGIC II. I don't

The 1D Mk. II uses the same DIGIC II processor as the 350D, yet is capable
of almost 8x the speed of the 350D, so you contend that it is not the
processor that is the bottleneck, but the card.

What makes you think that the card is the bottleneck, when the same cards
are capable of an eightfold improvement in speed in a different camera?

Or are you saying that the cards used in the 1D are always 8x faster?

I never said the card is the bottleneck, I said the component that transfers
chunks of memory (frames) from the buffer to the card appears to be the
bottleneck. I was simply surprised that a significantly cheaper compact
camera would be faster than an SLR from the same company, costing (with
lens) about 2-4 times the price!

 

[Richard Polhill]

I never said the card is the bottleneck, I said the component that transfers
chunks of memory (frames) from the buffer to the card appears to be the
bottleneck. I was simply surprised that a significantly cheaper compact
camera would be faster than an SLR from the same company, costing (with
lens) about 2-4 times the price!

Ah right. I misunderstood, sorry.

 

[FrankLM]

I doubt it. What is the size of each image (in MB)?


No that gives the number of megapixels per second. A truly arbitrary measurement.

MPx size <> file size.

The test is actually 1.1 * number of MB per file.

RAW + corresponding Super-Fine JPG file averages about 12.5 megs every 1.1
seconds on this P&S camera, with no slow-down for the total capacity of a memory
card (up to 8 gigs).

If I had a faster card the rate would increase (as verified by in-camera
benchmark tests on various cards, the faster the card, the faster the camera can
write to it, RAW data-saving being the swiftest so far, coming in at 10.1 megs
per second on my fastest card, the JPG processing/saving slows down the camera).
I'm using the fastest cards available to be able to keep up with the camera. As
I said, and others have stated, dSLRs are not designed for storage speed. That's
why they give them huge buffers for burst modes so they can s-l-o-w-l-y dump
them to the card when done shooting. They don't have to be fast when they don't
have things like 60 fps video modes with 44.1khz stereo sound recording built
into them.

The OP's question was whether or not a fast storage card will make a difference
in today's more advanced P&S cameras.

That's why I replied.

Storage card speed DOES matter. In some modes with slower cards the cards can't
keep up. If you want top performance out of today's P&S cameras invest in the
fastest cards you can find. They will still work with the slower cards but some
things like hi-speed burst modes or saving larger RAW files may get bogged down
to half-speed once the in-camera buffer is maxxed-out.

I wasn't replying to help the rest of you try to justify why you wasted your
money on a slower and less-features dSLR.