Matrox Parhelia 256 PCI-X Performance

[feipoa]

I was able to run 3DMark2001SE on a Matrox Parhelia 256 PCI-X card in, both, a conventional 33 MHz - 32-bit PCI slot and a PCI-X, 64-bit - 66 MHz slot on the same computer.

The test computer was a Dell PowerEdge 750 with Pentium 4 HT 2.8 GHz and 4 GB of dual channel DDR400 memory. The chipset on this motherboard is an Intel 875P/E7210.

The settings for 3DMark2001SE were:
1280x1024x32
Triple buffer
z-buffer 24-bit
D3D Pure Hardware T&L

Windows 2000 sp4 was installed with DirectX 9.0c. No sound card. ATA100 HDD.

With the Parhelia in the PCI-X slot, the score was 3854. With the Parhelia in the conventional PCI slot, the score was 4580. Any ideas why the score would be higher with the card in the conventional PCI slot?

[Unknown_K]

What speed slot do you have, they can be 66mhz, 100mhz, or 133mhz (and possible setup in the BIOS).

What is the max speed of the card (does it use faster PCI-X bus speeds)?

[feipoa]

From my internet research, the Dell PowerEdge 750 contains a 66 MHz, 64-bit PCI-X slot. The Matrox Parhelia 256 PCI is a 66 MHz, 64-bit PCI-X card.

[Logistics]

I would look into which version PCI the Parhelia requires, and which version the motherboard supports. (2.1, 2.2, 2.3, etc.)

[feipoa]

Are there different PCI-X specifications for 64-bit/66-MHz slots? I thought the different specification came with differing speeds.

I also ran the Matrox Parhelia PCI-X in two other systems as follows.

IBM eServer 306m with a 3.0 GHz Pentium D. The result was a score of 603.

IBM eServer 336 with dual 3.0 GHz Pentium Xeons. The result was a score of 5170.

Unfortunately, both the IBM eServers didn't have a PCI-only slot to compare with the Dell PowerEdge. Perhaps I can put a piece of tape over the contacts to get it working in 32-bit only (EDIT: hardware malfunction error after taping).

The eServer 336 has a 133 MHz PCI-X slot, but it slows down to 66 MHz (according to my oscilliscope) when using the Parhelia card. At least it doesn't slow down to 33 MHz. I am still puzzled by the slower results using the PCI-X slot compared to the conventional PCI slot.

By way of comparison, my Parhelia 128 AGP card scores 4115 in a dual PIII-Tualatin 1.5 GHz in XP Pro.

[feipoa]

I have now confirmed that the Matrox Parhelia PCI-X card works on a dual PIII Tualatin board in a PCI-X slot. The IBM eServer x330 contains a PCI-X slot and is based on the ServerWorks ServerSet III LE chipset. The drivers installed without any problem in W2K SP4. Quake III ran fine, but is markidly slower than the Parhelia 128 AGP on my dual PIII Tualatin MSI board. 51 fps vs. 109 fps.

[PhilsComputerLab]

Very nice!

Question: Does this card have any form of AA in the driver?

[dr.zeissler]

I have an PCI-X Version to sell. If someone is interested.

[obobskivich]

feipoa - thank you for testing this out. I've always wondered if the PCI-X Parhelia was "special" for PCI-X systems, but based on this it would seem that conventionally "good" PCI cards (e.g. FX 5200, Radeon 9250, etc) would be comparable, or better, and likely much cheaper.

I'm also curious about AA and other features vs the AGP card too, if you know/get a minute to check.

[feipoa]

What exactly do you want me to test in regards to AA?

I am planning on testing some other PCI graphics cards in that particular dual Tualatin IBM rack mount server. This would make for a fairer comparison. Were there 33 MHz PCI-X slots (64-bit of course)? If so, this system might have one. I should check for this as well. In any case, having a PCI sound card on the bus would bring it down to 33 MHz. Does having that sound card on the bus also reduce the bus width to 32 bit instead of 64-bit?

EDIT: According to my scope, this PCI-X slot runs at 33 MHz, with or without the PCI sound card. At 64-bit / 33 MHz, it is the slowest PCI-X slot made. Better than 32-bit / 33 MHz, I suppose.

[Logistics]

Do all the boards you tested the PCI-X Parheliainn make use of dual-channel memory architecture, like the Serverworks chipset?

[feipoa]

If I recall, the Dell PowerEdge 750 with P4 HT 2.8 GHz makes use of dual channel DDR400. I beleive the IBM eServer 336 with dual 3.0 GHz Xeons is dual channel. The PIII Tualatin with ServerSet III LE is single channel. Ideally, I'd like to get a PIII Tualatin with dual channel and a PCI-X slot which works with the Parhelia PCI-X board. Those are based on the ServerSet III HE-SL chipset. But that chipset also supports AGP, though not all boards implemented AGP.

These single height PIII Tualatin servers are rather quiet, especially compared to the other ones I tested. The loudest item in there is the worn out U160 SCSI drive. I have 7 units that I am going to rotate through to find the lest noisy. I'll probably keep the best 2 and scrap the other cases. These would make for some nice garage or basement servers if you can adapt the onboard U160 SCSI to SATA. Some photos.

[feipoa]

At 1280x720, the dual PIII 1.4 GHz Tualatin IBM eServer x330 with Matrox Parhelia 256 PCI-X scores 3115 in 3DMark2001 SE (W2K)

By way of comparison, at 1280x1024, my MSI dual PIII 1.5 GHz Tualatin with Matrox Parhelia 128 AGP scores 4137 in 3DMark2001 SE (XP Pro).

The reason for the different screen resolutions is due to the monitor. I turned my old Samsung LCD TV into my garage test monitor.

[feipoa]

There seems to be quite a deviation in benchmark results with different Matrox Parhelia driver versions. I will need to hunt down the best performing drivers. The 3DMark value reported earlier, of 3115, has, for example, increased to 4952 when using the driver version that came on the Matrox CD, which was supplied with the card in the box. Unfortunately, there are dozens of Matrox driver versions for the Parhelia.

I will also compare the Quake III and 3DMark2001 results with other PCI cards of the time, such as the GeForce 6200, FX600 (another rare card), FX5500, ATI Radeon 9250, Radeon 9000, GF2MX400, and ATI FirePro 2260.

[feipoa]

Enclosed are the results for the Matrox Parhelia 256 PCI-X card in comparison with other PCI graphics cards I have in possession. All cards were run on the same system, the IBM eServer x330, which is a PCI-X-only system, containing 2 expansion slots.

Cards tested
Matrox G450
Matrox Parhelia 256
NVIDIA GeForce 2 MX400
NVIDIA Quadro FX600
NVIDIA GeForce FX5500
NVIDIA GeForce 6200 (eVGA)
NVIDIA GeForce 6200 (PNY)
ATI Radeon 9000
ATI Radeon 9250

Test System
Windows 2000 Pro SP4
IBM eServer xSeries 330
Broadcom ServerWorks ServerSet III LE chipset
Dual Intel 1.4 GHz Tualatin CPUs
1 GB registered ECC memory
Creative ES1373 sound card

Notes
3DMark2001 SE results for Matrox G450 in software mode
3DMark2001 SE requires DirectX 8.1
PCI-X slot running at 33 MHz, with and without the sound card
Graphic results the same, with and without the sound card installed

Quake III
1024x768x32
Default OpenGL
Highest quality settings
Bilinear
Sound enabled
Command used to start the benchmark: "timedemo 1", <carriage return>, followed by "demo four"

3DMark2001 SE
1280x720x32
Triple buffer
z-buffer 24-bit
D3D Pure Hardware T&L

Using driver versions 2.0.x had a detrimental effect for the Matrox Parhelia PCI card. Using driver version 1.00.x, in particular 1.09, yielded a 60% increase in performance for 3DMark2001SE and a 40% increase in performance for Quake III. A similar situation often occurs with NVIDIA graphic driver versions.

The Parhelia PCI-X results were a little disapointing. Perhaps the 33 MHz speed of the PCI-X slot on the test computer were holding it back some because the results for the Parhelia 128 AGP card on a different, but similar, Tualatin system were quite a bit higher. In both benchmarks, the Parhelia PCI card fell behind a Radeon 9250, GeForce 6200, and Quadro FX600. More tests will need to be done to dermine if the PCI-X slot on this particular IBM eServer, or on all boards based on the the ServerSet III LE chipset, are holding the card back. Also, tests with a more powerful CPU would be of interest. I do have a motherboard which contains a PCI-X slot and an Operton CPU, however I do not have high hopes for much improvement. Unless you are interested in tri-monitor applications/gaming, I see little value in a Matrox Parhelia PCI-X card. In addition, others have reported buggy game play, at least with the AGP version, particularly for earlier card revisions. I'm not sure how this information affects the PCI-X version of the card.

My main attraction to the Matrox Parhelia PCI-X card was the novelty in running a PCI-X graphics card on non-AGP and non-PCIe motherboards. Perhaps there are some CAD applications that run fantastic with the CAD-optimised drivers.

[udam_u]

I performed following Parhelia 256 PCI, Parhelia 128AGP and Quadro FX4000 tests in 3DM 2001SE v330 benchmark (default settings 1024x768x32 Pure Hardware T&L, Antyaliasing NO).

Test platform
Motherboard: Asus A7M 266-D
CPU's: 2x Athlon MP 2800+ (2133MHz)
RAM: 2GB ECC Registered 266MHz CL2
HDD: 2x Seagate 15k 300GB raid 0
System: Windows XP SP3

driver used for Parhelia cards: 1.13.0.158
BIOS installed in Parhelia 256PCI: 5.1-93
BIOS installed in Parhelia 128AGP: 2.2-47

driver used for Quadro: 307.45

The results are as follow:


As you can see results from Parhelia 128 AGP and 256PCI working in PCI 64bit 66MHz slot are almost identical. Parhelia 256 PCI installed in PCI 32bit port suffer from bandwith limitation. @feipoa It seems that there is some problem with motherboards that you used in tests. Did you update Parhelia Bios to the latest version?

BTW how to upload image to Vogons?

[feipoa]

My PIII-S 1400 with only a 33 MHz, 64-bit PCI-X slot scored 4983 at 1280x1024 in 3DMark2001 SE. Your significantly faster Athlon MP 2800, with faster PCI-X slot, and at a lower resolution scored only 6159. Could you run 3DMark2001 SE at 1280x1024x32 with triple buffer for comparison? Perhaps your result is the GPU limit for this card.

I'm not sure what was going on with that Dell PowerEdge 750 P4 server board concerning the better results with the card in a conventional PCI slot.

I have not check the Parhelia's BIOS date recently, but when I did some initial tests 4 years ago, I'm sure I'd have checked the BIOS revision. I will check again to be sure. However, I think this graphics card is just very particular about which motherboards it will function on. I also doubt that many motherboard manufacturers considered support with a PCI-X graphics card.

I was not aware of any upload restrictions for files on Vogons. Perhaps there is a quota based on post quantity now?

[udam_u]

My PIII-S 1400 with only a 33 MHz, 64-bit PCI-X slot scored 4983 at 1280x1024 in 3DMark2001 SE. Your significantly faster Athlon MP 2800, with faster PCI-X slot, and at a lower resolution scored only 6159. Could you run 3DMark2001 SE at 1280x1024x32 with triple buffer for comparison? Perhaps your result is the GPU limit for this card.

My "new" Parhelia 256PCI has probably damaged some ram chips because I am not able to perform tests at 1280x1024x32 with triple buffer enabled (serious graphics glitches and very low performance). I tried older drivers and replacing pink glue under radiator with Arctic MX-4 thermal compound but without positive results. I was able to perform mentioned tests on Parhelia 128AGP and the result is 5106 3D Marks so scores around 5000 points are probably Parhelia limit.

I'm not sure what was going on with that Dell PowerEdge 750 P4 server board concerning the better results with the card in a conventional PCI slot.

Maybe block diagram of your motherboard could clarify situation. I remember that AMD 760MP which was first dual processor chipset for Athlon CPU has 64bit PCI but working only at 33MHz. Its successor 760MPX supports 66MHz 64 bit PCI bus (chipset installed on my motherboard).
We have to remember that in vintage motherboards PCI bandwidth was used not only for communication with PCI cards but also in most cases with South Bridge. Furthermore I believe that in some motherboards utilizing chipset not designed strictly for server purposes PCI-X may be implemented only to meet electrical requirements of PCI-X interface and physically be connected to 32 bit PCI via some bridge.

I have not check the Parhelia's BIOS date recently, but when I did some initial tests 4 years ago, I'm sure I'd have checked the BIOS revision. I will check again to be sure. However, I think this graphics card is just very particular about which motherboards it will function on. I also doubt that many motherboard manufacturers considered support with a PCI-X graphics card.

The latest bios is dated Dec. 19, 2012 (for Parhelia DL256PCI ).

I was not aware of any upload restrictions for files on Vogons. Perhaps there is a quota based on post quantity now?

Do you have special button in "POST A REPLY" menu to upload image to Vogons? I am afraid that after some time images uploaded on external servers will disappear like image I used to have in my signature. (;

[feipoa]

Click "quick reply" then click "full editor".
I checked and my Parhelia PCI already has the latest BIOS version.
I tried the card in my Opteron motherboard in a standard PCI slot, but it seems the Vista x64 drivers do not work with Windows 7 x64.
Too bad about your Parhelia not working as 1280+

[PCBONEZ]

With the Parhelia in the PCI-X slot, the score was 3854. With the Parhelia in the conventional PCI slot, the score was 4580. Any ideas why the score would be higher with the card in the conventional PCI slot?

.
PCI-X like conventional PCI is a shared bus.
If the specific PCI-X slot you used happens to be shared by something else on the motherboard (drive controllers and LAN are common) then the full bandwidth is not available to the video card.

It is not uncommon for there to be two or even three independent PCI-X buses on a motherboard.
If you want to use PCI-X video it would be a good idea to find out which PCI-X bus has the least on it.
Manuals for boards with PCI-X usually show what is on what bus in the manual.
.
On more recent boards with PCI-X the PCI-X slot may actually be on a bridge chip that is already on PCIe .
.