merci glacias

feipoa wrote:

Is there a bridged AGP card from NVIDIA which has Flash acceleration that you would recommend?

The last AGP card with an nVidia chip was the 7950GT AGP. Not sure if it has Flash acceleration or not. It would be a generation or two behind the last Radeon AGP cards, so probably not.

Since i've been working in different linux distros trying to find one I like to make as a router, i dawned on me this morning I could just use linux to read the spd info and not have to go through the trouble of installing XP and wasting all that time.

long story short, they are programmed as follows:
pc100 2-2-2
pc133 3-3-3

honestly I'm not surprised, these dimms were sets designed for HP/compaq servers. So anything beyond set standards is not wanted.

Sometimes the chips are PC133 CL2, but the on-module buffers may only be PC133 CL3. I've run into that before. I've also run into the case where the memory chips are PC133 CL2, the buffers are PC133 CL2, and yet the SPD is programmed for CL3. I've read online that some people try to reprogram the SPD. The first account I read about was unsuccessful, so I did not bother to find more.

it would be very easy (relatively) if you had the proper tools. just remove the spd chip, reprogram, and go nuts.

As I prep. my Supermicro to test out Flash GPU acceleration, I did some initial memory tests, the results of which were a little surprising.

1) 3x 1GB Elpida shielded registered/buffered ECC memory, SPD programmed as CL3, I manually configured BIOS to CL3 and ECC off.
36 hrs memtest OK at CL3. CL2 setting fails.
Cachechk read: 694 MB/s
Cachechk write: 395 MB/s

2) 4x 512MB Kingston registered/buffered ECC memory, SPD programmed as CL3, however the onboard registers and memory chips are 143 MHz. I manually configured BIOS to CL2 and ECC off.
23 hrs memtest OK at CL2.
Cachechk read: 726 MB/s
Cachechk write: 395 MB/s

3) 4x 512MB Micron unbuffered, non-ECC memory, SPD programmed as CL3.
28 hrs memtest OK at CL3.
Cachechk read: 726 MB/s
Cachechk write: 395 MB/s

4) 4x 512MB Micron unbuffered, non-ECC memory, SPD programmed as CL2.
46 hrs memtest OK at CL3. CL2 fails.
Cachechk read: 726 MB/s - this is the same read speed as CL3 memory (makes sense)
Cachechk write: 472 MB/s - this is the same write speed as CL2 memory (doesn't make sense?)

Observations:

1) registered/buffered memory at CL2 with ECC off demonstrates the same throughput as non-registered, non-ECC memory at CL3. So if you aren't going to use the ECC feature of the registered/buffered DIMMs, is there any real usefullness to them, aside from the fact that some motherboards/chipsets require buffered memory above a certain threshold?

2) Why when CL2 memory is set to CL3 in the BIOS does the write speed stay at CL2 speeds (refer to bolded entries above)? So the CL2/3 setting in the BIOS is only for the read speed of memory modules? The BIOS takes the write speed from the SPD without the option to manually set it? Seems odd. Both cachechk and Memtest indicated that the CL2 memory set as CL3 was faster than the CL3 memory set as CL3. For CL3 memory at CL3, Memtest recorded the combined speed as 386 MB/s. For CL2 memory at CL3, Memtest recorded the combined speed as 440 MB/s.

I am concerned about item 2) because if these BGA memory modules I ordered are programmed as CL3, then setting them to CL2 still won't allow for increased write speeds. If I am going to run Windows 7, is it better to run 2 GB of memory at 726/472 MB/s (read/write) or 3 GB of memory at 726/395 MB/s?

feipoa wrote:

2) Why when CL2 memory is set to CL3 in the BIOS does the write speed stay at CL2 speeds (refer to bolded entries above)? So the CL2/3 setting in the BIOS is only for the read speed of memory modules? The BIOS takes the write speed from the SPD without the option to manually set it? Seems odd.

IIRC, I had a similar experience while manually changing from CL3 to CL2 in my bios. The write speeds don't vary because they are mostly dependent on other memory timing parameters, tRP & tRCD. So when you change CAS latency from 2T to 3T in BIOS, the values originally set by SPD earlier(tRP=tRCD=2T) remains as it is like you said and doesn't reduce the write speeds. I just rechecked it now with my BIOS. As the option is not there to vary the tRP & tRCD timings in my BIOS (most likely in your case too), I used to manually set the relevant northbridge chipset register to 'tight' timings using the dos PCI config tool(really love it) and then boot the OS. You can also try the System Information Viewer tool (or other such tools) to verify the current memory timings for your system.

Do you have the name and web link for the PCI config tool you used?

I tried to install WinXP with these fast-writing RAM sticks, but it hung during hardware detection. I rebooted, it hung again during hardware detection, though it got a bit further. I rebooted, it hung again after hardware detection, while trying to save the configuration.

My first thoughts are that the memory write speeds are too fast and are causing stability problems, though MemTest did not detect any, or the ATI HD 4350 is causing problems.

BTW, it is the same tool that I mentioned when the issue related to your Matrox Parhelia with Intel Server Set system was being discussed and hence it is not a specific tool for tweaking memory timings. You have to know the relevant northbridge register set and its various possible values to use the tool effectively for modifying the memory timings. I saw the SuperMicro P3TDDE manual mentioning the Northbridge as VIA VT8633/8653, but I couldn't get detailed datasheets for them. However if you want, I have some possible indirect options.

feipoa wrote:

My first thoughts are that the memory write speeds are too fast and are causing stability problems, though MemTest did not detect any, or the ATI HD 4350 is causing problems.

If the PC133 memory is rated for CL2(2-2-2-5) then there should not be any problems and normally should hang well before hardware detection, though I might be wrong. I have even booted into Windows and ran Quake 3 with timings set to 3-2-2-5 for PC133 CL3 memory rated as 3-3-3-6 by SPD.

Probably the hardware detection error is due to your HD 4350 in which case running a Ubuntu/Lubuntu live CD with 'nomodeset' enabled under F6 may throw some light on it.

I was thinking that the write speed may be too fast for the VIA chipset, not necessarily for the memory. I doubt Supermicro and VIA did much testing with unbuffered, non-ECC CL2 memory with a combined size of 2 GB. I do recall a single stick of 512 MB CL2 memory working even when set to CL2, so I suspect the chipset cannot handle the memory size. Not throwing a MemTest error is odd though.

When I have a moment, I will A) setup XP with a different graphics card, and pending failure, B) change the memory to CL3.

As suspected, it was the combination of memory speed and quantity being too much for the VIA chipset. I reduced the memory size from 4x512MB to 1x512MB and XP installed flawlessly.

I first tried a Matrox G200A AGP card using 4x512MB, and install froze.

Confirmed, when using CL2 memory and setting CL3 in BIOS, I get these results,

CL2 memory set to CL3
CL = 3
tRCD = 2
tRD = 2
tCAS = 6

CL2 memory set to CL2
CL = 2
tRCD = 2
tRD = 2
tCAS = 6

CL3 memory set to CL2
CL = 2
tRCD = 3
tRD = 3
tCAS = 6

Back to the original topic. I tested the ATI HD 4350 in Windows XP SP2 using Google Chrome. I have ATI Catalyst version 10.7 installed.

Is Youtube fully onboard with HTML5 now? None of the videos I played used Flash. Anyway, in fullscreen and non-fullscreen modes, GPU-Z doesn't indicate that the GPU is being used at all. My CPU is fully utilised and when in fullscreen mode, there are obvious skipped frames.

When I play videos using a Flash-only player (not using Youtube), I still do not get any GPU usage and my CPU is utilised 100%.

Do I need to use a Catalyst version newer than 10.7 for acceleration to work? I choose 10.7 because I read online somewhere that someone had to use 10.7 for the HD 4350 to work on their older system.

luckybob wrote:

look for BGA sdram. It is on a newer process and overclocks well. I bought three of these: http://www.ebay.com/itm/121288981097 (6gb) for my dual tualatin. It ran 24 hours of memtest and 24 hours prime at 2-2-2 I don't think a better deal exists for name-brand 1gbb ecc/reg ram

How much RAM does Windows XP report with that much RAM? I thought 3GB or 3.2GB was the limit?
How much RAM does your BIOS report? I cannot seem to get the BIOS on any dual Tualatin board to report more than about 3.6GB

Tests using the AMD HD4350 in, both, Windows XP Pro and Windows 7 Pro indicate no change in processor utilisation when playing Youtube videos (HTML5 and Flash). Fullscreen and non-fullscreen modes were tested in, both, Firefox 19 and Chrome 34. Some videos had "HD" on them, while others did not. I tried Catalyst versions 9.12 and 10.7. I can only hope I am doing something wrong and the answer is trivial. It seems gandhig had more luck with his GT520 with respect to on-board Flash acceleration.

I never used XP, but in win 2k it reported the full 6gb. I had to manually enable PAE, but it all was there.

I kinda wish I had the disposable income to play with a few of these pci-x to pci-e adapters: http://www.startech.com/Cards-Adapters/Slot-E … -Card~PCIX1PEX4

2 of those, my dual tualatin board, and CROSSFIRE ATI CARDS. ( or sli, whatever)

feipoa wrote:

Tests using the AMD HD4350 in, both, Windows XP Pro and Windows 7 Pro indicate no change in processor utilisation when playing Youtube videos (HTML5 and Flash). Fullscreen and non-fullscreen modes were tested in, both, Firefox 19 and Chrome 34. Some videos had "HD" on them, while others did not. I tried Catalyst versions 9.12 and 10.7. I can only hope I am doing something wrong and the answer is trivial. It seems gandhig had more luck with his GT520 with respect to on-board Flash acceleration.

Hmm. A while back it was thought that GPU decoding of H.264 required a CPU with SSE2. I wonder if AMD cards do actually require SSE2 in order to do hardware decode? Obviously, NVIDIA doesn't, since gandhig was able to get hardware accelerated 1080p playback in DirectShow players and in Flash.

I'm curious though--have you tried playing 720p or 1080p MKV files in MPC-HC or WMP11 with CoreAVC as the decoder? With an older version of CoreAVC, my single Tualatin at 1.59GHz has just enough CPU horsepower to play 5Mb/s 720p MKV files without any hardware acceleration. CoreAVC is multi-threaded, so your dual PIII just might be able to to handle lower-bitrate 1080p files in a DirectShow player without help from the video card.

feipoa wrote:

Tests using the AMD HD4350 in, both, Windows XP Pro and Windows 7 Pro indicate no change in processor utilisation when playing Youtube videos (HTML5 and Flash). Fullscreen and non-fullscreen modes were tested in, both, Firefox 19 and Chrome 34. Some videos had "HD" on them, while others did not. I tried Catalyst versions 9.12 and 10.7. I can only hope I am doing something wrong and the answer is trivial. It seems gandhig had more luck with his GT520 with respect to on-board Flash acceleration.

As per my understanding, if your GPU is capable of delivering hardware acceleration for HD Videos playback(offline), then it should theoretically aid online flash playback, atleast in full screen(only youtube HD videos, didn't have any success with other flash video websites or html5 playback).

When I bought the card in 2012, I tried a HD movie trailer(.mov) on MPC-HC with DXVA and it played flawlessly. Then I tried youtube flash video, it was a complete failure even at low resolution levels. Two years passed and then lately I tried Firefox's recent version(29 or 30), flash HD video acceleration worked under some odd conditions.

If I were you, I would try offline HD video playback first. Did you try the latest version of Firefox?

Does anyone know where I can download an offline HD video file?

luckybob wrote:

I kinda wish I had the disposable income to play with a few of these pci-x to pci-e adapters: http://www.startech.com/Cards-Adapters/Slot-E … -Card~PCIX1PEX4

2 of those, my dual tualatin board, and CROSSFIRE ATI CARDS. ( or sli, whatever)

Those certainly look tempting, but not at that price. There is probably a high risk of it not working properly on a PIII.

luckybob wrote:

I kinda wish I had the disposable income to play with a few of these pci-x to pci-e adapters: http://www.startech.com/Cards-Adapters/Slot-E … -Card~PCIX1PEX4

2 of those, my dual tualatin board, and CROSSFIRE ATI CARDS. ( or sli, whatever)

Damn. I could use one of those with my dual 604 board if they weren't so expensive.