|
|||||||||
|
Thread Tools | Search this Thread |
May 1st, 2005, 05:59 AM | #2791 |
Regular Crew
Join Date: Oct 2002
Location: Netherlands
Posts: 111
|
One thing to consider when choosing a SATA RAID board is whether it does the RAID calculation in hardware or not. This may not be a big deal when doing simple stripe-sets, but when doing RAID 5 with actual redundancy high datarates will actually place a big burden (I've seen 30% CPU load per 100MB/s on Opteron systems) on the CPU.
With Obin's current CPU load issues you wouldn't want to add CPU load for the RAID calculations. Bar3nd |
May 2nd, 2005, 08:10 AM | #2792 | |
Silicon Imaging, Inc.
Join Date: May 2004
Location: Troy, NY USA
Posts: 325
|
Obin,
There is one out there that I know of - I/O Industries. They have a superb frame grabber with an on-board SCSI RAID controller. Unfortunately, they are priced out of your range - $3.5K for the FG, $2.5K for the SW. We were able to record 100MPix/sec, 12 bit on a PCI-32 machine without any problems so for applications where you need the highest rates without any bus headaches, it is ideal. For cost sensitive applications, it doesn't fly. Steve Quote:
__________________
Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com |
|
May 2nd, 2005, 09:27 AM | #2793 |
Regular Crew
Join Date: Mar 2005
Location: Frankfurt / Germany
Posts: 46
|
Hi Steve,
is the I/O controller you mention the one with the 5 times dual 2 GBit FC host ports and CameraLink interface? Its said to write 850 MByte/s. Or is it something different you are talking about? Anyway, is the goal to have Obin's camera on long cable or to have recording "build it". The last one will stay a problem with uncompressed HD for a while :( Axel |
May 2nd, 2005, 12:31 PM | #2794 |
Silicon Imaging, Inc.
Join Date: May 2004
Location: Troy, NY USA
Posts: 325
|
IO Industries - DVR Express:
http://ioindustries.com/express.htm We have bundled this FG with our cameras in the past. Very nice hardware.
__________________
Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com |
May 3rd, 2005, 05:58 AM | #2795 |
Regular Crew
Join Date: Mar 2005
Location: Frankfurt / Germany
Posts: 46
|
Hi Steve,
yep, thats the one I had in mind too. I didn't know its from I/O industries, anyway, it seems to be the fastest direct to disk solution currently available. Do you know if there will be PCIexpress x8 or x16 frame grabbers? The best I found so far do only x4, which is limiting around 500 MByte/s, which is to slow for some projects, which should be enough for this project. PCIexpress has the big advantage that you can use a typical low-cost SATA RAID controller for writing the data to disk, rather than expensive 5 times FC dual 2 GBit (=10 GBit total) bandwidth. The RAIDs controllers & enclosures alone are expensive enough to buy an HDCAM... Axel |
May 3rd, 2005, 06:38 AM | #2796 | |
Silicon Imaging, Inc.
Join Date: May 2004
Location: Troy, NY USA
Posts: 325
|
Axel,
Leutron was talking about PCIe at the Stuttgart Vision show. Matrox is always on the cutting edge of capture speed. Others are planning it. Keep in mind that the camera link buffers are at 85MHz now - that is 85MHz * 2*12 bits per camera link channel. Above that, you run more parallel pathways. Since 500MB/sec is about the full bandwidth of the PCI-64/66 bus, how much more data do you want to move? OK, 3 chip...... Quote:
__________________
Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com |
|
May 3rd, 2005, 07:54 AM | #2797 | |
New Boot
Join Date: Oct 2004
Location: BG
Posts: 19
|
Quote:
so PCIe x 4 should be 2GByte/s ...no? |
|
May 3rd, 2005, 08:49 AM | #2798 |
Regular Crew
Join Date: Mar 2005
Location: Frankfurt / Germany
Posts: 46
|
Hi Marin, Hi Steve,
I am in contact with Leutron and they are not yet ready with the PCIe version (and most important there are no WinXP64 drivers ready yet, which I would need too). I too have in mind that PCIe 4x should be faster than 500 MByte/s, but thats what the Leutron person told me. Anyway, for uncompressed standart HD formats something around 250 MBytes/s for RGB 12 bit data should be enough, what do you think? Viper style... I am more on faster frame speeds for this high bandwidth thingy. Axel |
May 3rd, 2005, 08:53 AM | #2799 |
Regular Crew
Join Date: Mar 2005
Location: Frankfurt / Germany
Posts: 46
|
Hi,
I've just looked at their website (sorry for german here): Mit PicPort-Express-CL ultraschnelle Datenübertragungen von Zeilen- und Matrixkameras mit bis zu 500 MByte/sec. in den Hauptspeicher des PC´s realisieren. Camera Link Framegrabber für den PCI Express x1- und x4-Bus Datenrate x1-Bus: 250 MByte/sec. x4-Bus: 500 Mbyte/sec. So thats 250 MByte/s for x1 and 500 MByte/s for x4 type cameras. I do think that a x16 graphics card can do 4000 MByte/s (bi-directional), though that would mean 1x is 16x/16 = 4000 MByte/s / 16 = 250 MByte/s. In that way the equation makes sense to me, but I may be totally wrong. Axel |
May 3rd, 2005, 09:16 AM | #2800 |
Silicon Imaging, Inc.
Join Date: May 2004
Location: Troy, NY USA
Posts: 325
|
Axel,
I don't know what your level of system and software expertise is but please understand that you are talking about moving a tremendous amount of data. System architecture and RAID design should be very well thought out and benchmarked at these rates. It will be very easy to go through many hardware purchases trying to balance a system design at 500MB/sec. Continuous. Important word to add.
__________________
Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com |
May 3rd, 2005, 09:52 AM | #2801 |
Trustee
Join Date: Jan 2003
Location: Wilmington NC
Posts: 1,414
|
YEs Steve..I am talking with I/O about that board..and the Savant software..looks like we have no luck with our software...I will have a demo of Savant in a few days to see if they can save on our system.if so I could use the SDK and write a GUI around it...argggg I wish this thing would work
|
May 3rd, 2005, 10:27 AM | #2802 | |
Silicon Imaging, Inc.
Join Date: May 2004
Location: Troy, NY USA
Posts: 325
|
A cheaper alternative is that I can get you an OEM version of StreamPix that lets you use their underlying disk code and you can build a GUI around it. Contact me off-line if you want to persue that.
Quote:
__________________
Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com |
|
May 3rd, 2005, 11:51 AM | #2803 |
Regular Crew
Join Date: Mar 2005
Location: Frankfurt / Germany
Posts: 46
|
Hi Steve & Obin,
some points here: a) You are right about these peak limit issues. I guess I assembled enough FC / SAN and HDTV editing systems to know about the difficulities one can run into reaching "expected" speed limits. I've been working on SCSI based systems doing real 800+ MByte/s sustained & random record & playback from disks, given RAID0 conditions, but NTFS! For writing the RAID subsystems caching is the simple (!) key, as long as all other parts comply within the required limits (though stay on average above the required bandwidth). This said, a physical limit of 500 MByte/s means to me to expect something around 350 to 400 MByte/s to be really satisfied. There are technologies that allow "easy" writing of 800/1600 MByte/s, small enough to consider even portable environments. b) When you consider the DVR express for Obin's camera, you are going for the SCSI one, right? Given HD at single chip, SCSI is fine. c) I still wonder if its simpler to reach the goal of writing back 4:4:4 uncompressed with "off the shelf" hardware puzzles compared to developing a "good enough" light compression scheme (I know, that would be a "givin up" compromise somehow). d) Given the thread title of "4:4:4 10bit single CMOS HD project" I still wonder how many pixels you consider to be a 4:4:4 10 Bit HD signal? 4:4:4 would usually means you record 4*13.75 MHz bandwidth per color component (if I remember video electronics well). In total this should be ~165 MPixel @ 10 bit. But then your chip would need true RGB representation in full resolution. Somehow 4:4:4 isn't the best acronym in the context of a single chip camera to me, since a typical single chip Bayer pattern does a "hardware" 4:2:2 (now talking in weighting of pixels). I guess the approach will be to write back the Bayer pattern in realtime and do the processing to RGB afterwards? Excuse me if that has been said somewhere in the thread, I haven't read all 186 pages :( Given a 10 Bit 1920x1080y chip it would be 20736000 bits per frame, ~2.5 MByte/frame Bayer pattern, around 60 MByte/s @ 24 Hz. That should fit in a small housing. Am I wrong in any point of the project? e) Leutron just told me they expect the x1/x4 cards to ship around June. WinXP64 support could be there in time if required (probably not necessary for Obin anyway). Axel (just trying to understand :) ) |
May 4th, 2005, 05:30 PM | #2804 |
Trustee
Join Date: Jan 2003
Location: Wilmington NC
Posts: 1,414
|
I need some help bad.
Our software works. Our CPU and dual hard disk drives are more then fast enough to record and display..my motherboard the DFI 855GME has a problem. we can't record AND display at the same time as things seem to crash inside the chipset...this is also with the Video Savant software, this tells me what we are working aginst is REAL and not a fault of my code writer..I need a new motherboard that has SATAx2 PCI-X x1 and a mobile CPU socket for the INTEL MOBILE cpu does anyone know of such a beast that is NOT running on the INTEL 855 chipset or using the 6300ESB SATA controller?? all Dev stops till we get this fixed ...so please help asap..thanks a bunch gang.. I REALLY don't want this to be "dead in the water" after all the effort I have in it! please post or email direct links if you have any of hardware I can try |
May 4th, 2005, 05:54 PM | #2805 |
Regular Crew
Join Date: Mar 2005
Location: Frankfurt / Germany
Posts: 46
|
Hi Obin,
so you search for a board that has - mobile CPU - PCI-X slot (for the capture card) - Either onboard SATA controllers or place for another controller. Right? I will have a look. If I find something I'll post it here. Btw, is AGB required or would a simple onbaord graphics card OK? Axel |
| ||||||
|
|