New AGPGart

cybland · March 26, 2008

Hi Slice,

I've installed both Tiger and Leopard. With Tiger I've received joblo's AGPGart which seems to be working. However now I have Leopard with CI/QE, without AGPGart, and Tiger without CI/QE but with AGPGart. I've managed to ruin my Tiger install trying to get CI/QE so I guess I'll have to reinstall it tomorrow. In the meantime here are dmesg and ioreg from Leopard and Tiger fresh after the install. Hope that it helps.

cybland_Tiger_and_Leopard.zip

JaS · March 26, 2008

Agpgart (not edited) detect everything it finds

AGP: VRAM=[d0000000, 10000000]AGPINTEL: aperture [d8000000, 08000000] (aperture set to 128mb in bios)

And I see this

AGP: VRAM=[e0000000, 10000000]AGPINTEL: aperture [e8000000, 04000000] (aperture set to 64mb in bios)

AGP: VRAM=[e0000000, 10000000]AGPINTEL: aperture [e8000000, 02000000] (aperture set to 32mb in bios)

AGP: VRAM=[c0000000, 10000000]AGP: WARNING! Please set your aperture to 64 or 128MB in your BIOS for more compatibility AGPINTEL: aperture [c8000000, 10000000] (aperture set to 256mb in bios)

Why is my vram size changing depending on the aperture size? also why does the first value in the aperture size always have an 8 as the second number for vram+aperture no matter what the second value says(true aperture size)?

Setting the values manually in the agpgart plist gives me nothing but a black screen no matter what I try to set it to.

Hope this info helps.

Slice · March 26, 2008

I've installed both Tiger and Leopard. With Tiger I've received joblo's AGPGart which seems to be working. However now I have Leopard with CI/QE, without AGPGart, and Tiger without CI/QE but with AGPGart. I've managed to ruin my Tiger install trying to get CI/QE so I guess I'll have to reinstall it tomorrow. In the meantime here are dmesg and ioreg from Leopard and Tiger fresh after the install. Hope that it helps.

First of all you need to have QE/CI in Tiger without AGPGart and then try to use it for acceleration. But I dunno about Leo. "Common Sense" with AMD64 have no success in Leo while in Tiger he has full acceleration as with joblo's driver as with my.

http://forum.insanelymac.com/index.php?sho...1308&st=254

In your dmesgs I don't see information from AGPGart to understand what is happen.

Agpgart (not edited) detect everything it finds
AGP: VRAM=[d0000000, 10000000]AGPINTEL: aperture [d8000000, 08000000] (aperture set to 128mb in bios)
And I see this

Fantastic images! I never see that. First screen is shifted? May be fboffset value?

It means AGPGart really switched on. Not only bus tuning as AGPlight do. But it works with bad address of framebuffer.

VRAM=[d0000000, 10000000] - plain framebuffer [address, size] works without GA.plugin

aperture [d8000000, 08000000] - AGP framebuffer [address, size] work with GA.plugin

But how other drivers uses those addresses? What is best address for AGP framebuffer?

AGP: VRAM=[e0000000, 10000000]AGPINTEL: aperture [e8000000, 04000000] (aperture set to 64mb in bios)

AGP: VRAM=[e0000000, 10000000]AGPINTEL: aperture [e8000000, 02000000] (aperture set to 32mb in bios)

AGP: VRAM=[c0000000, 10000000]AGP: WARNING! Please set your aperture to 64 or 128MB in your BIOS for more compatibility AGPINTEL: aperture [c8000000, 10000000] (aperture set to 256mb in bios)

Why is my vram size changing depending on the aperture size?

So I see BIOS change VRAM address depending on aperture size!

also why does the first value in the aperture size always have an 8 as the second number for vram+aperture no matter what the second value says(true aperture size)?

First value is address. By my default

AGP_address=VRAM_address+VRAM_size/2

in theory it might be VRAM_ADDR+VRAM_SIZE but usually VRAM_SIZE already twice as real. As I don't sure in the calculation I propose you to change the result manually. "AGP_Base" key.

Setting the values manually in the agpgart plist gives me nothing but a black screen no matter what I try to set it to.

In any case one value must be the same as default named "0". Black screen - bad AGP_Base.

I have no other explanation, except -_- : the framebuffer driver don't work with AGPFramebuffer. I have the result with Callisto which never works with AGP. But you have nVidia like CommonSense.

Hope this info helps.

Very thank you for detailed reports. If I had more such testers my work may finished last year.

Dan Druff · March 27, 2008

Hi

You can look in my sign for my specs. With 10.5.2 (With Graph update) I've tried the method Izmael reported to work with 10.5.1. First, I couldn't find iondrvsupport.kext from official 10.4.7 update, so I guessed that it's a previous version and I took 10.4.5 for my test. Besides that, Leo booted fine and showed AGP in the profiler as you can see by the pics, but boot was longer and feeled more sluggish. I did some OpenMark test and saw that

the method doesn't give the promised results. OpenGL seems to be impaired by this setup, but at least it's working.

My PCI setup is more snappier and get better results on OpenMark. I can't tell what kext I use for this setup cos I was a bit sloppy and kept the working Kext w/o keeping track of the origin, but I installed Vanilla Leo 10.5.0 with PC_EFI v8, update to 10.5.2 and graph update, I just replaced a couple of kext but the graph ones are vanilla except for ATIRadeonX1000 to which I added the 71C2 Dev ID to plist (no other mod). I do not use AGPGart and I also get some tearing from time to time with this setup, but the perf on OpenMark is superior and with near vanilla kext. My other setup is exactly the same but with Izmael's specified kext.

Pics (I tried to take a picture of the average FPS of each setup and not maxed out FPS)

Vanilla kext w AGPGart 2.4.9 tearing exemple

Bottom line, we've got two report with same results, as cyclonefr said Izmael method is flawed and non fully working on i865G chipset but not completly bad.

If I can do some testing I would be happy, but I would need a little coaching first (just a little how-to).

EDIT

My PCI setup works with PC_EFI but without GFX-String, unlike my AGP setup, so only the kext works and there is no turn around there.

JaS · March 27, 2008

With AGPGart 2.6.5.b2 and IONDRVsupport from 10.4.5 I get this

kernel AGP: Coherence support: no 
3/27/08 4:14:28 AM kernel AGP: GART is 32 bit capable 
3/27/08 4:14:28 AM kernel AGP: Found an AGP 3.0 compliant device. 
3/27/08 4:14:28 AM kernel AGP: VRAM=[e0000000, 10000000]AGPINTEL: aperture [e8000000, 04000000] ( 64mb aperture )
3/27/08 4:14:28 AM kernel AGPINTEL trace PCI space 
3/27/08 4:14:28 AM kernel (00)=25708086   (04)=20900106   (08)=06000002   (0c)=00000000	
3/27/08 4:14:28 AM kernel (10)=e8000008   (14)=00000000   (18)=00000000   (1c)=00000000	
3/27/08 4:14:28 AM kernel (20)=00000000   (24)=00000000   (28)=00000000   (2c)=25708086	
3/27/08 4:14:28 AM kernel (30)=00000000   (34)=000000e4   (38)=00000000   (3c)=00000000	
3/27/08 4:14:28 AM kernel (40)=00000000   (44)=00000000   (48)=00000000   (4c)=00000000	
3/27/08 4:14:28 AM kernel (50)=004a0200   (54)=001c8040   (58)=00000000   (5c)=00000000	
3/27/08 4:14:29 AM kernel (60)=00000500   (64)=00000000   (68)=00000000   (6c)=00000000	
3/27/08 4:14:29 AM kernel (70)=00000000   (74)=00000000   (78)=00000000   (7c)=00000000	
3/27/08 4:14:29 AM kernel (80)=00000000   (84)=00000000   (88)=00000000   (8c)=00000000	
3/27/08 4:14:29 AM kernel (90)=00111110   (94)=00333300   (98)=00000000   (9c)=00381a00	
3/27/08 4:14:29 AM kernel (a0)=00300002   (a4)=1f004a1b   (a8)=00000b00   (ac)=00000000	
3/27/08 4:14:29 AM kernel (b0)=00000080   (b4)=00000030   (b8)=03f66000   (bc)=00001010	
3/27/08 4:14:29 AM kernel (c0)=00000000   (c4)=240e4000   (c8)=00000000   (cc)=00000000	
3/27/08 4:14:29 AM kernel (d0)=0e042802   (d4)=00000d0b   (d8)=00000000   (dc)=41400000	
3/27/08 4:14:29 AM kernel (e0)=00000000   (e4)=0106a009   (e8)=00000200   (ec)=00000000	
3/27/08 4:14:29 AM kernel (f0)=00000000   (f4)=00000000   (f8)=00030f68   (fc)=00000000	
3/27/08 4:14:29 AM kernel AGP: Setting 08 data rate 
3/27/08 4:14:29 AM kernel AGP: command written target=00000b12 master=1f00e312 
3/27/08 4:14:29 AM kernel AGP memory 1131c000 length 00001000 offset 00000000 cnt 00000000 
3/27/08 4:14:29 AM kernel AGP memory 1131c000 length 00001000 offset 00001000 cnt 00000001 
3/27/08 4:14:30 AM kernel AGP memory 1131c000 length 00001000 offset 00002000 cnt 00000002 
3/27/08 4:14:30 AM kernel NVDA,Display-A: vram [00000000:10000000] 
3/27/08 4:14:30 AM kernel NVDA,Display-B: vram [00000000:08000000]

The thing to really notice here is I have something new and for the first time look at the NVDA display

3/27/08 4:14:30 AM kernel NVDA,Display-A: vram [00000000:10000000] 
3/27/08 4:14:30 AM kernel NVDA,Display-B: vram [00000000:08000000]

I was able to get to the desktop once with this new combo of agpgart and iondrvsupport kexts. Ci/ QE were still not supported.

Dan Druff · March 27, 2008

Another test on my AGP setup, I used AGPGart 2.6.5.b2 and the result is the same, booted, get to desktop, AGP found but OpenGL impaired and perf below PCI setup with near vanilla kext.

Slice · March 27, 2008

Another test on my AGP setup, I used AGPGart 2.6.5.b2 and the result is the same, booted, get to desktop, AGP found but OpenGL impaired and perf below PCI setup with near vanilla kext.

Show us your dmesg!

2 JaS

<_>

Dan Druff · March 27, 2008

Here are my outputs for my AGP setup:

Leo 10.5.2

AGPGart 2.6.5.b2

IONDRVsupport 10.4.5

atindrv 10.4.5

Graph kext -> Graph update 1.0 with mod to ATIRadeonX1000.kext with 71C2 key added

dmesg_Dandruff_2.6.5.b2.txt

ioreg_Dandruff_2.6.5.b2.txt

heng2006 · March 27, 2008

To Slice, As your requests and I send additional information for helping youThank you very much for your helpsheng2006--------------------------Asrock 4Dual-VSTA InformationGeneralCPU - LGA 775 for Intel® Core™ 2 Quad* / Core™ 2 Extreme* / Core™ 2 Duo / Pentium® XE / Pentium® D / Pentium® 4 / Celeron® D, supporting Quad Core Kentsfield processors- FSB 1066/800/533 MHz- Supports Hyper-Threading Technology- Supports Untied Overclocking Technology- Supports EM64T CPU*When you adopt Quad Core CPU, FSB may be reduced 5%Chipset - Northbridge: VIA® PT880 Ultra- Southbridge: VIA® VT8237AMemory - Dual Channel DDR/DDRII memory technology*- 2 x DDRII DIMM slots- Supports DDRII667/533 non-ECC, un-buffered memory- Max. capacity of system memory: 2GB- 2 x DDR DIMM slots- Supports DDR400/333/266 non-ECC, un-buffered memory- Max. capacity of system memory: 2GB*DDRII and DDR are supported separatelyBIOS - 4Mb AMI BIOS- AMI Legal BIOS- Supports "Plug and Play"- ACPI 1.1 Compliance Wake Up Events- Supports jumperfree- SMBIOS 2.3.1 Support Audio, Video and NetworkingGraphics - n/aAudio - 7.1 CH Windows® Vista™ Premium Level HD Audio (ALC888 Audio Codec)LAN - VIA® PHY VT6103- Speed: 10/100 Ethernet- Supports Wake-On-LAN Expansion / ConnectivitySlots - 4 x PCI slots- 1 x PCI Express Graphics slot (@ x4 mode)- 1 x AGP 8X slotConnector - 2 x Serial ATA 1.5 Gb/s connectors, support RAID (RAID 0, RAID 1, and JBOD) and Hot Plug functions- 2 x ATA133 IDE connectors (support 4 x IDE devices)- 1 x Floppy connector- 1 x IR header- 1 x Game header- 1 x HDMI_SPDIF header- CPU/Chassis FAN connector- 20 pin ATX power connector- 4 pin 12V power connector- CD in header- Front panel audio connector- 2 x USB 2.0 headers (support 4 USB 2.0 ports)Rear Panel I/O HD 8CH I/O- 1 x PS/2 Mouse Port- 1 x PS/2 Keyboard Port- 1 x Serial Port: COM1- 1 x Parallel Port (ECP/EPP Support)- 4 x Ready-to-Use USB 2.0 Ports- 1 x RJ-45 LAN Port- HD Audio Jack: Side Speaker / Rear Speaker / Central / Bass / Line in / Front Speaker / Microphone Other Features / MiscellaneousHybrid Booster - CPU Frequency Stepless Control- ASRock U-COP- Boot Failure Guard (B.F.G.)Support CD - Drivers, Utilities, AntiVirus Software (Trial Version)Accessories - Quick Installation Guide, Support CD, I/O Shield- Floppy/ATA 133 cables- 1 x SATA data cable (optional)- 1 x SATA 1 to 1 power cable (optional)- 1 x HDMI_SPDIF cable (optional)Hardware Monitor - CPU Temperature Sensing- Chassis Temperature Sensing- CPU Fan Tachometer- Chassis Fan Tachometer- CPU Quiet Fan- Voltage Monitoring: +12V, +5V, +3.3V, VcoreForm Factor - ATX Form Factor: 12.0-in x 9.6-in, 30.5 cm x 24.4 cmOS - Microsoft® Windows® 2000 / XP / XP 64-bit / Vista™ / Vista™ 64-bit compliantCertifications - FCC, CE, WHQL------------------------------------PT880 Ultra Chipset informationVIA PT880 Ultra - The Perfect Stepping-Stone Solution for System Integrators and PC Enthusiasts This product is available fully ROHS-compliant.Based on VIA's innovative Flex Express Architecture, the VIA PT880 Ultra chipset provides the ultimate transition solution, offering support for PCI Express and AGP graphics, as well as DDR and DDR2 memory, all on the same motherboard. Compatible with the latest Intel® processors, the VIA PT880 Ultra chipset includes support for 1066MHz FSB and Hyper-Threading Technology.Featuring VIA's revolutionary Universal Graphics Interface (UGI), the VIA PT880 Ultra enables support for the latest generation of PCI Express Graphic cards as well as existing AGP 3.0 graphics cards, providing OEMs, SIs, and end-users much needed flexibility during the transition period. Furthermore, motherboards based on the VIA PT880 Ultra support VIA DualGFX Express, by running both AGP and PCI-Express graphics cards simultaneously, providing the ability to connect to as many as four separate displays.By harnessing the advanced VIA StepUp memory controller technology, motherboards based on the VIA PT880 Ultra can support either DDR or DDR2 memory modules on the same motherboard, providing users the flexibility to use whatever memory standard best suits their requirements.The VIA PT880 Ultra features a high bandwidth Ultra V-Link bus to allow communication between the chipset's North and South bridges at a blazing 1GB/s. When combined with solutions such as the market-leading VIA VT8237R Plus South Bridge and a broad array of VIA companion chips, the VIA PT880 Ultra provides a comprehensive range of storage, multimedia and connectivity features, including native Serial ATA and V-RAID, 6-channel or 8-channel VIA Vinyl Audio, VIA Velocity™ Gigabit Ethernet, USB2.0, and 1394.The VIA PT880 Ultra chipset has received PCI-Express certification, having successfully completed the rigorous testing procedures of the PCI-SIG Compliance Workshop.Features of VIA PT880 UltraFeature VIA PT880 UltraNorth Bridge Processor Support Intel® Pentium® 4, Celeron®, and Core™2 Duo processorsHyper Threading Support YesFront Side Bus 1066/800/533MHzMemory Support VIA StepUp Technology for dual 64-bit channel memory with co-layout support » DDR2 533/400 » DDR 400/33/266Graphics Support VIA Universal Graphics Interface » AGP 3.0 (AGP8X) » PCI ExpressBus Architecture AsynchronousMax. Memory 4GBSouth Bridge VIA VT8237R PlusNorth/South Bridge Link Ultra V-Link (1066MB/sec)Audio VIA Vinyl™ 6-channel Audio (AC'97 integrated)VIA Vinyl™ Gold 8-channel Audio (PCI companion controller)Network VIA Velocity Gigabit Ethernet companion controllerVIA integrated 10/100 Fast EthernetModem MC'97PCI Devices/Slots 6 slotsDual PCI-X bus support through VIA VPX2 I/O expansion bridgeSATA 2 x SATA 150 devicesSATALite™ interface for two additional SATA devices 4 total)V-RAID RAID 0, RAID 1, and RAID 0+1* & JBOD (SATA)PATA ATA133 (up to 4 devices)USB 8 portsI/O Protocols I/O APIC / LPC Super I/0Power Management ACPI/APM/PCI/PMTo Slice, Thank youheng2006

dmesgHeng.txt

Slice · March 27, 2008

Here are my outputs for my AGP setup:

Leo 10.5.2

AGPGart 2.6.5.b2

IONDRVsupport 10.4.5

atindrv 10.4.5

Graph kext -> Graph update 1.0 with mod to ATIRadeonX1000.kext with 71C2 key added

In your dmesg I don't see agp transaction like this

 AGP memory 1131c000 length 00001000 offset 00000000 cnt 00000000 
AGP memory 1131c000 length 00001000 offset 00001000 cnt 00000001 
AGP memory 1131c000 length 00001000 offset 00002000 cnt 00000002

Have you GL.bundle?

What is Provider of your GA.plugin?

			<key>IOProviderClass</key>
		<string>IOFramebuffer</string>

?

Dan Druff · March 27, 2008

In your dmesg I don't see agp transaction like this

 AGP memory 1131c000 length 00001000 offset 00000000 cnt 00000000 
AGP memory 1131c000 length 00001000 offset 00001000 cnt 00000001 
AGP memory 1131c000 length 00001000 offset 00002000 cnt 00000002

Have you GL.bundle?

What is Provider of your GA.plugin?

            <key>IOProviderClass</key>
            <string>IOFramebuffer</string>

?

I don't know what is GL.bundle or GA.plugin. Give me some light, I'll give you some...

Edit

if it's ATIRadeonX1000GA.plugin and ATIRadeonX1000GLDriver.bundle (both v. 1.5.24.9), yes I have them, from Leo Graph Update.

For what I can guess you want to know...

<dict>

<key>IOProviderClass</key>

<string>IONDRVFramebuffer</string>

</dict>

Slice · March 27, 2008

I don't know what is GL.bundle or GA.plugin. Give me some light, I'll give you some...

Edit

if it's ATIRadeonX1000GA.plugin and ATIRadeonX1000GLDriver.bundle (both v. 1.5.24.9), yes I have them, from Leo Graph Update.

For what I can guess you want to know...

IOProviderClass

IONDRVFramebuffer

Yes, they are.

Change IONDRVFramebuffer -> IOFramebuffer

and see result!

Dan Druff · March 27, 2008

I've made the change and see the results:

OpenMark

Edit

I just want to underline that there's no diff in perf from 2.4.9 or 2.6.5.b2 w/o the change to ATIRadeonX1000GA.plugin and from the dmesg I can see that AGP isn't supported.

Then if there's no perf diff and AGP is

Extension "com.apple.driver.AppleACPIPlatform" has immediate dependencies on both com.apple.kernel and com.apple.kpi components; use only one style.
AGP: Found Intel 82865 host to AGP bridge
AGPGart: start
AGP: capptr = a0
AGPgart: BridgeMemoryRange [fe900000, fe9fffff]

or isn't

AGP: not supported 244e
mbinit: done

supported, it may be safe to conclude that AGP isn't used in either setup even if it's present.

dmesg_AGPGart_2.6.5.b2_IOFrambuffer.txt

ioreg_AGPGart_2.6.5.b2_IOFrambuffer.txt

Slice · March 27, 2008

I've made the change and see the results:

Edit

I just want to underline that there's no diff in perf from 2.4.9 or 2.6.5.b2 w/o the change to ATIRadeonX1000GA.plugin and from the dmesg I can see that AGP isn't supported.

Then if there's no perf diff and AGP is

or isn't
AGP: not supported 244e
 mbinit: done
supported, it may be safe to conclude that AGP isn't used in either setup even if it's present.

244e - other device that AGP tried to connect. Not supported - OK!

Real problem later

IOAccelerator driver must makes two calls of AGPGart

createAGPspace(...) - for tuning. The result I see here

(a0)=00300002   (a4)=1f004a1b   (a8)=00000a00   (ac)=00000000   
(b0)=00000080   (b4)=00000020   (b8)=039ce000   (bc)=00001020   
(c0)=00000000   (c4)=040d4000   (c8)=00000000   (cc)=00000000   
(d0)=0e042802   (d4)=00000d0b   (d8)=00000000   (dc)=01300000   
(e0)=00000000   (e4)=0106a009   (e8)=00000200   (ec)=00000000   
(f0)=00000000   (f4)=00000000   (f8)=00030f68   (fc)=00000000   
AGP: Aperture [f0000000:08000000]
AGP: GART is 32 bit only
AGP: Setting 08 data rate

All is fine!

And next call

commitAGPmemory(...) - to perform transactions. It must be like this

 AGP memory 1131c000 length 00001000 offset 00000000 cnt 00000000 
AGP memory 1131c000 length 00001000 offset 00001000 cnt 00000001 
AGP memory 1131c000 length 00001000 offset 00002000 cnt 00000002

But you have no this! And I don't understand why.

May be you try to replace ATIRadeonX1000.kext to Tiger version? It decides on or off commit method...

You have IOAGPFlags=0x10027. May be better to use =1?

And where is

	IOLog("AGP: command written target=%08lx master=%08lx\n", tcom, mcom);

????

Dan Druff · March 27, 2008

I would like to know if using Cinebench (which is UB) could be used to compare the capabilities of my GFX card. I could use it in OSX and compare the results with WinXP, as it is cross platform, I could have a better idea about what to expect from my card.

Also, if someone knows a QUICK cross-platform OpenGL benchmark, I would be pleased to hear about it.

cybland · March 27, 2008

Hi Dan Druff,

You can use Cinebench95 to get an idea. As a matter of fact I done that on my rig and got some interesting results.

@Slice,

I'm still here. Spent last two days testing your AGPGart 264 and IOPCIFamily 244 with Tiger, and I have a lot of data to give you. Let me just organize them first and than I'll post them here. Only one general remark. With this AGPGart my system has some strange problems booting. It goes like:

1. first boot - kernel panic (see log):

Mar 27 18:45:28 localhost kernel[0]: hi mem tramps at 0xffe00000
Mar 27 18:45:30 localhost kernel[0]: PAE enabled
Mar 27 18:45:30 localhost kernel[0]: standard timeslicing quantum is 10000 us
Mar 27 18:45:30 localhost kernel[0]: vm_page_bootstrap: 510047 free pages
Mar 27 18:45:30 localhost kernel[0]: mig_table_max_displ = 71
Mar 27 18:45:30 localhost kernel[0]: Local APIC version not 0x14 as expected
Mar 27 18:45:30 localhost kernel[0]: Enabling XMM register save/restore and SSE/SSE2 opcodes
Mar 27 18:45:30 localhost kernel[0]: ACPI CA 20060421
Mar 27 18:45:30 localhost kernel[0]: AppleACPICPU: ProcessorApicId=0 LocalApicId=0 Enabled
Mar 27 18:45:30 localhost kernel[0]: AppleACPICPU: ProcessorApicId=1 LocalApicId=1 Disabled
Mar 27 18:45:30 localhost kernel[0]: Copyright (c) 1982, 1986, 1989, 1991, 1993
Mar 27 18:45:30 localhost kernel[0]: The Regents of the University of California. All rights reserved.
Mar 27 18:45:30 localhost kernel[0]: using 10485 buffer headers and 4096 cluster IO buffer headers
Mar 27 18:45:30 localhost kernel[0]: IOAPIC: Version 0x11 Vectors 64:87
Mar 27 18:45:30 localhost kernel[0]: ACPI: System State [S0 S3 S4 S5] (S3)
Mar 27 18:45:30 localhost kernel[0]: DEBUG fixing subordinate bus setting on HUB0 device(14, 0) - was 2 now 255
Mar 27 18:45:30 localhost kernel[0]: Security auditing service present
Mar 27 18:45:30 localhost kernel[0]: BSM auditing present
Mar 27 18:45:30 localhost kernel[0]: disabled
Mar 27 18:45:30 localhost kernel[0]: From path: "uuid", 
Mar 27 18:45:30 localhost kernel[0]: Waiting for boot volume with UUID 9F575657-B784-315B-B6C1-C6C91959CFF3
Mar 27 18:45:30 localhost kernel[0]: Waiting on <dict ID="0"><key>IOProviderClass</key><string ID="1">IOResources</string><key>IOResourceMatch</key><string ID="2">boot-uuid-media</string></dict>
Mar 27 18:45:30 localhost kernel[0]: m168efe17
Mar 27 18:45:30 localhost kernel[0]: m168efe17
Mar 27 18:45:30 localhost kernel[0]: m168efe17
Mar 27 18:45:30 localhost kernel[0]: m168efe17
Mar 27 18:45:30 localhost kernel[0]: m168efe17
Mar 27 18:45:30 localhost kernel[0]: AGP: Found NVIDIA nForce3-250 AGP Controller
Mar 27 18:45:30 localhost kernel[0]: AGPBridge buses: pri 0 secBus 1 subBus 1 BridgeDev 0
Mar 27 18:45:30 localhost kernel[0]: AGPgart: BridgeMemoryRange [fa000000, fcffffff]
Mar 27 18:45:30 localhost kernel[0]: AGPgart: BridgePrefetchableMemoryRange [e0000000, efffffff]
Mar 27 18:45:30 localhost kernel[0]: AGPgart: BridgeIORange [0000b000, 0000bfff]
Mar 27 18:45:30 localhost kernel[0]: AGP: saveBridgeState OK
Mar 27 18:45:30 localhost kernel[0]: AGP create nub for 02e110de bus=1 cap=68
Mar 27 18:45:30 localhost kernel[0]: AppleNForceATA: NVIDIA nForce by medevil (CMD 0x9e0, CTR 0xbe0, IRQ 23, BM 0xcc00)
Mar 27 18:45:30 localhost kernel[0]: AppleNForceATA: NVIDIA nForce by medevil (CMD 0x960, CTR 0xb60, IRQ 23, BM 0xcc08)
Mar 27 18:45:30 localhost kernel[0]: USB caused wake event (EHCI)
Mar 27 18:45:30 localhost kernel[0]: USBF:	53.554	AppleUSBOHCI[0x462e800]::start OHCI controller will be unloaded across sleep
Mar 27 18:45:30 localhost kernel[0]: USBF:	53.556	AppleUSBOHCI[0x46b9800]::start OHCI controller will be unloaded across sleep
Mar 27 18:45:30 localhost kernel[0]: m168efe17
Mar 27 18:45:30 localhost kernel[0]: m168efe17
Mar 27 18:45:30 localhost kernel[0]: AGPGart: publishNub VGAG
Mar 27 18:45:30 localhost kernel[0]: AppleNForceATA: NVIDIA nForce by medevil (CMD 0x1f0, CTR 0x3f4, IRQ 14, BM 0xe000)
Mar 27 18:45:30 localhost kernel[0]: AppleNForceATA: NVIDIA nForce by medevil (CMD 0x170, CTR 0x374, IRQ 15, BM 0xe008)
Mar 27 18:45:30 localhost kernel[0]: WARNING: ATA Drive claims FLUSH CACHE EXT feature support but does not claim Extended LBA feature support
Mar 27 18:45:30 localhost kernel[0]: WARNING: ATA Drive claims FLUSH CACHE EXT feature support but does not claim Extended LBA feature support
Mar 27 18:45:30 localhost kernel[0]: m168efe17

Right after that a KP listing.

2. Just hit reset and the second boot goes OK till AGP trace is listed on screen after which the system is booted in the background but the screen is frozen (looks like this):

Mar 27 20:49:05 localhost kernel[0]: NVDA::start(VGAG) <1>
Mar 27 20:49:05 localhost kernel[0]: NVDA::start(VGAG) <1> failed
Mar 27 20:49:05 localhost kernel[0]: AGP: Coherence support: no
Mar 27 20:49:05 localhost kernel[0]: AGP: GART is 32 bit capable
Mar 27 20:49:05 localhost kernel[0]: AGP: Found an AGP 3.0 compliant device.
Mar 27 20:49:05 localhost kernel[0]: AGPAMD: 1 hammers found
Mar 27 20:49:05 localhost kernel[0]: AGPAMD apbase=f4000000
Mar 27 20:49:05 localhost kernel[0]: AGPAMD: aperture [f4000000, 04000000]
Mar 27 20:49:05 localhost kernel[0]: AGPAMD trace hammer PCI space
Mar 27 20:49:05 localhost kernel[0]: (00)=11031022   (04)=00000000   (08)=06000000   (0c)=00800000   
Mar 27 20:49:05 localhost kernel[0]: (10)=00000000   (14)=00000000   (18)=00000000   (1c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (20)=00000000   (24)=00000000   (28)=00000000   (2c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (30)=00000000   (34)=00000000   (38)=00000000   (3c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (40)=00003bff   (44)=00000040   (48)=00000000   (4c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (50)=0180c080   (54)=00000008   (58)=00000000   (5c)=d05b04c0   
Mar 27 20:49:05 localhost kernel[0]: (60)=000000bd   (64)=00000000   (68)=00000000   (6c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (70)=51020111   (74)=50008011   (78)=08003800   (7c)=0000221b   
Mar 27 20:49:05 localhost kernel[0]: (80)=23070000   (84)=21132113   (88)=00000000   (8c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (90)=00000003   (94)=0000007a   (98)=000c5380   (9c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (a0)=00000000   (a4)=00000000   (a8)=00000000   (ac)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (b0)=00000000   (b4)=00000000   (b8)=0000003e   (bc)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (c0)=00000000   (c4)=00000000   (c8)=00000000   (cc)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (d0)=00000000   (d4)=04e20707   (d8)=00000000   (dc)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (e0)=00000000   (e4)=07530320   (e8)=00000519   (ec)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (f0)=00000000   (f4)=00000000   (f8)=00000000   (fc)=00000000   
Mar 27 20:49:05 localhost kernel[0]: AGPAMD trace via PCI space
Mar 27 20:49:05 localhost kernel[0]: (00)=00e110de   (04)=00b00006   (08)=060000a1   (0c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (10)=f4000008   (14)=00000000   (18)=00000000   (1c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (20)=00000000   (24)=00000000   (28)=00000000   (2c)=02501462   
Mar 27 20:49:05 localhost kernel[0]: (30)=00000000   (34)=00000044   (38)=00000000   (3c)=000000ff   
Mar 27 20:49:05 localhost kernel[0]: (40)=02501462   (44)=01c0c008   (48)=11110022   (4c)=000000d0   
Mar 27 20:49:05 localhost kernel[0]: (50)=003f0523   (54)=00000003   (58)=00002000   (5c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (60)=00034732   (64)=00044566   (68)=00000666   (6c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (70)=00000000   (74)=00000000   (78)=00003030   (7c)=00881111   
Mar 27 20:49:05 localhost kernel[0]: (80)=00888813   (84)=000000c8   (88)=00000070   (8c)=0000003f   
Mar 27 20:49:05 localhost kernel[0]: (90)=00000000   (94)=00000000   (98)=00000000   (9c)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (a0)=00000000   (a4)=00000000   (a8)=00000000   (ac)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (b0)=00000007   (b4)=00660066   (b8)=00000000   (bc)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (c0)=00300002   (c4)=1f00421b   (c8)=00000001   (cc)=ffffffff   
Mar 27 20:49:05 localhost kernel[0]: (d0)=00000000   (d4)=00000000   (d8)=00000000   (dc)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (e0)=00000000   (e4)=00000000   (e8)=00000000   (ec)=00000000   
Mar 27 20:49:05 localhost kernel[0]: (f0)=00000000   (f4)=00000000   (f8)=00000000   (fc)=00000000   
Mar 27 20:49:05 localhost kernel[0]: AGP: Setting 08 data rate

3. Just hit reset button and the third boot is successful (booted to GUI).

All this is with 64Mb Aperture Bios setting. With 128Mb and 256Mb setting it takes more resets to boot to GUI. Also with 64Mb and 128Mb setting I have CI/QE but with 256Mb setting no CI/QE.

Anyhow, with AGPGart there was no improvement in OpenMark and OpenGL Extensions Viewer, but there was significant improvement in Xbench and Cinebench95. If you are interested I can attach the results of the benchmarks.

Dan Druff · March 27, 2008

May be you try to replace ATIRadeonX1000.kext to Tiger version? It decides on or off commit method...
You have IOAGPFlags=0x10027. May be better to use =1?

And where is
    IOLog("AGP: command written target=%08lx master=%08lx\n", tcom, mcom);
????

Here are my results with:

AGPGart -> 2.6.5.b2

ATIRadeonX1000.kext -> 10.4.5

IONDRVFramebuffer -> IOFrambuffer in ATIRadeonX1000GA.plugin

I had difficulties to get to the GUI (two reboot), apple logo not showing in taskbar (shown in image), menu is translucent (more than usual) and I got (once in each case) a reboot signal (the gray film on the screen with the infamous message) while trying 1. to get to system profiler 2. when starting GIOFX. Also, GIOFX doesn't start, well the program start but there's no image and the GFX card don't get hotter (no fan startup as usual).

The dmesg seems empty...

here's the desktop image

Edit

I would like to know about a good free image hosting site, mine is crappy.

Re-Edit

Problem solved for hosting

dmesg.txt

ioreg.txt

Dan Druff · March 28, 2008

I've made some other OpenGL test with my PCI setup with near vanilla 10.5.2 / Graph update kext and my AGP setup (described in the post #417). Results from GioFX is the same, but as you can see, Cinebench describe the

opposite pattern of data, higher in AGP and lower in PCI. Look at the pictures...

PCI CB = 1051 AGP CB = 1142

In each case, the difference is only by a small amount, but in reverse pattern, and these scores are reliable and repeatable. It should be noted that the movie created by Cinebench is less fluid with the AGP than the PCI despite the higher score.

riws · March 28, 2008

Dan Druff, why no OpenMark (GioFX) Benchmark runs?

Slice · March 28, 2008

Hi Dan Druff,

You can use Cinebench95 to get an idea. As a matter of fact I done that on my rig and got some interesting results.

@Slice,

I'm still here. Spent last two days testing your AGPGart 264 and IOPCIFamily 244 with Tiger, and I have a lot of data to give you. Let me just organize them first and than I'll post them here. Only one general remark. With this AGPGart my system has some strange problems booting. It goes like:

1. first boot - kernel panic (see log):

All this is with 64Mb Aperture Bios setting. With 128Mb and 256Mb setting it takes more resets to boot to GUI. Also with 64Mb and 128Mb setting I have CI/QE but with 256Mb setting no CI/QE.

Anyhow, with AGPGart there was no improvement in OpenMark and OpenGL Extensions Viewer, but there was significant improvement in Xbench and Cinebench95. If you are interested I can attach the results of the benchmarks.

Yes, we interesting. What is mean "significant"?

Slice · March 28, 2008

It is good idea to compare benchmark by Cinebench95 in Windows and MacOSX.

With my Radeon9000IGP I have no yet HardwareOpenGL, QE/CI. Test with softwareOpenGL.

Dan Druff · March 28, 2008

Dan Druff, why no OpenMark (GioFX) Benchmark runs?

If I knew, I would solve the problem... :wallbash:

cybland · March 28, 2008

Hi all,

Here are test results. Everything is done in Tubgirl's Tiger (10.4.10) with AGPGart 264 and IOPCIFamily 244. Remark: Yesterday my Software Update decided to update my Tiger to 10.4.11 so after some workarounds I was able to boot. Therefore don't be surprised that some of tests state that my system is 10.4.11, now it is, but there are no differences in results. After full two days of testing I can say that AGPGart is working with AMD64 on nForce3 if you set 64Mb or 128Mb aperture. With aperture set to 256Mb there is no CI/QE as I've already reported. Haven't tested the 32Mb aperture setting. Also I've seen no difference in setting AGP_Mode to 0x0fffffef or to 0xffffffff (results marked with Fastwrites or FW). Also included are my Cinebench 9.5 results from WinXP with 128Mb and 256Mb aperture. Definitely there is still some space for improvement but I can't honestly say if the problem is with AGPGart or graphic kexts in OSx86. On a subject of quirky boot I've noted that this pattern (1. KP, 2. Screen freeze, 3. Normal boot as described in my previous post) happens only after I've changed aperture setting in BIOS. I've been able to overcome this quirks by always booting with -f switch (added to com.apple.Boot.plist) which have lessened the KP occurrence and by booting WITHOUT -v switch which lessened the occurrence of Screen freeze (maybe a framebuffer issue?).

@Slice: Should I do the same tests in Leopard or you want me to test something in Tiger (maybe AGPGart 265b)?

EDIT: I'll go and test Tiger by playing some NFS Carbon while I wait for your answer.

cybland_AGPGart_Test.zip

BikerS · March 28, 2008

Hi guys!

How i run my ATI Radeon x1250 mobility? I probably AGPGart version 2.4.9, but sustem not loaded this kext and resolution 1024x768!

Please help me!

Thanks advance!

Dan Druff · March 29, 2008

Hi guys!
How i run my ATI Radeon x1250 mobility? I probably AGPGart version 2.4.9, but sustem not loaded this kext and resolution 1024x768!

Please help me!

Thanks advance!

First, there's no magical solution and you may have to test a lot. Second, as your card seems recent, it may be PCIe and we could not provide you any means to use it properly, you should then go elsewhere in the forums to find your answer. If your card is AGP, then we could use a little more background (MB chipset, GFX card spec...) if you want us to help you.

@slice

You may put a little how-to in your first post for the new testers. With a how test, what test, how to give results, which benchmark (with a direct download for GioFX as it is no more available)... It should be good for everyone and avoid you to repeat your instructions to any newcommer.

@everyone

Why do you use Cinebench 9.5 as the 10 is out. Is there a reason?

New AGPGart

Recommended Posts

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites

Link to comment

Share on other sites