Jump to content
214 posts in this topic

Recommended Posts

Hello mercurysquad,

 

I'm glad that i could be of help. Their seem to be more links on the page whic explain the cause of the tsc drift.

 

http://developer.amd.com/documentation/art...1214200692.aspx

 

The article says it only affects the K8 socket of multi core processors (pub date 12/14/2006)

 

I hope they fixed this for their next generation products on AM2

 

I will be definately interested to see how you can apply this patch. I imagine you will need some kind of kernel flag to engage this feature.

 

I woud like to praise you for your work as it will help when apple starts selling AMD machines, amongst other things :)

 

Biggup

 

I'm using an X2 with a K8 board so this certainly sounds like the cause :)

AMD X2 64 5000 + works fine without any flag (i think so...)) but with -force64 flag freezing up while boot and no 64-bit support in cpu_log we can see..) (small question - in theory, or in all - can AMD support 64-bit under MAC OSX?)

In all other ways - all works fine!

Oh..and one, overclocked fsb bus (230mhz) displayed as 200mhz, but i don't see the problem in this..))

 

CPUID_Log.txt

 

MB - Asus M3A (AMD700+SB600)

My MSI-Wind is working perfect. Though it shows up as a Intel Core Solo.

 

1183 using Geekbench up from 1065 Sweeeet

GEEKBENCH is the best CPU bench , i know. But i think that the values are not different between kernels - if they didnt have any problems (spaming lots of errors in the console = takes CPU cyles). A difference less than 5 % maybe happens alyways, even with same kernel running at different times.

I agree with the above ^, but this is like 11% increase - I don't think a benchmark will have an error of ±10% !! While I can't pinpoint the reasons for performance increase on sse3 processors, my guess is that it's because proper cache info is propagated in the kernel for non-apple supported CPUs too, so perhaps the various libs and the benchmark itself is either optimising its test, or re-scaling the scores based on the (correct) cache size info, which was earlier probably hardcoded and/or incorrect in ToH kernels. Actually I expected performance to be just a little slower (<1%) because the timing code now uses the "slow" path for all CPUs. Seems improvement in other areas not only balanced, but actually surpassed it..

I agree with the above ^, but this is like 11% increase - I don't think a benchmark will have an error of ±10% !!

Your´re right - i cant count :) 11 % diff under same condition is more than some 1-5 % of measure fault.

But such benches must be really run under same condition - no background tasks (better boot in safe mode).

The article says it only affects the K8 socket of multi core processors (pub date 12/14/2006)

 

Is that including the AM2 socket? Or only S939? The processor I'm experiencing these issues on is a Athlon 64 X2 5000+ on socket AM2. Can anyone confirm there are no similar "mouse bug/timing issues" with the Phenom series?

Is that including the AM2 socket? Or only S939? The processor I'm experiencing these issues on is a Athlon 64 X2 5000+ on socket AM2. Can anyone confirm there are no similar "mouse bug/timing issues" with the Phenom series?

Athlon 64, Opteron and Semprons are all K8 Processors. The Phenom is K10 iirc, though, I'm not sure if they experience this bug.

Yes, I posted this on the previous page also -- Phenoms are K10, and the documentation specifically says the tsc is invariant.

 

Can a phenom user please confirm that this issue doesn't affect them?

I have tried the notsc flag to no avail.

 

Kernel flag thread: http://forum.insanelymac.com/index.php?showtopic=99891

 

Pease read the comments by mercurysquad in previous pages. He says that many kernel flags like this one are totally kernel dependant. Therfore you cannot expect a vanilla kernel to contain this or other non standard flags.

 

He also goes on to say that the kernel flag thread is full of these types of kernel dependant flags so don't expect them to do anything unless the kernel devleoper has said that these will work or have been implemented in his/her kernel. Hope that makes sense. Please read the orignal comment by mercurysquad as he made this point forward very clearly. :thumbsdown_anim:

 

edit - sorry guys, the commment i was referring to is in this thread. oops -> http://forum.insanelymac.com/index.php?sho...17558&st=60

edit - sorry guys, the commment i was referring to is in this thread. oops -> http://forum.insanelymac.com/index.php?sho...17558&st=60

Heh, I hadn't seen that post before, so thanks for pointing it out :thumbsdown_anim:

 

As for my comment about the notsc flag, I should have stated that I tried that particular flag with every non-vanilla kernel ever used on my system to no avail. At least I now know why :)

I am a phenom 9500 user, and with alpha13 using fsb 200000000 and busratio 11, the system runs smoothly and without errors after hours of work (internet, p2p, usb tdt receiver, and time machine works also. The usb bug with 2.0 devices disappear using the fsb and busratio flags.

system specs:

 

Phenom 9500

gigabyte ga-ma770 ds3 mobo

3 Gb Ram 667 DDR2

xfx geforce 8800 gts 320 Mb (EFI string)

sata hd with sb600 in ahci mode with stock appleahciport

chameleon bootloader

shutdown and restart working

leopard 10.5.5 (ideneb 1.0 updated to 10.5.5 with asu)

Running the Alpha 13 test on an Athlon X2 6000, I get panics when trying to install World of Warcraft, and when running parallels. Installing/running other software (Photoshop, Time Machine, GarageBand, iTunes, etc) is completely stable, including shutdown, reboot, sleep, etc.

 

The only thing that leads me to think it might be partially kernel related are the mDNSPlatformRawTime: last_mach_absolute_time errors that I tend to get during boot up and often right before the system becomes unstable.

 

Other than that, it's working well and I'm looking forward to the beta.

@mercurysquad Guess u were right. Booted with Stock Kernel got scores of 1165-1112. With your kernel I am getting scores of 1184-1154

 

Note the system is freshly booted and with the same exact state while running the tests. I guess AppleDecrypt was responsible for 1065 to 1165

@mercurysquad Guess u were right. Booted with Stock Kernel got scores of 1165-1112. With your kernel I am getting scores of 1184-1154

 

Note the system is freshly booted and with the same exact state while running the tests. I guess AppleDecrypt was responsible for 1065 to 1165

Do you mean the new AppleDecript V12 ? What .kext (dsmos.kext ?) must i remove if i use / install that (faster) decypt ?

Ok for once and for all - AppleDecrypt is not faster than dsmos :-) If anything, it's probably 0.1% slower.

 

Both these decryption kexts simply provide the key and hook back into the kernel's AES decryption routines, so they use the exact same algorithm, with the exact same performance.

 

The difference with AppleDecrypt being that it contains the key in a mangled form, which it has to de-mangle, and this takes a wee bit more time than dsmos which has the key in plaintext :angel:

Thanks for clearing AppleDecript ! So it isnt faster (dev say that!) but mabe little more legal :angel:

 

 

I read now (in google code) that VMWARE 2.0 has problems (one user says crashing start virt machine) with your kernel.

 

I have no problems running newest VMWARE , alpha11 + alpha13 on C2D E4400, GA-EP35 !

I dont know what goes wrong at that user.

REALLY BAD: Updating VMWARE with an supended/paused virtual machine !!!

Mostly a crash on next start of vm !!!

If you update VMWARE always shutdown all virt machines before update !

Bild_36.jpg

Hehe, saying and proving are 2 different things. I say sse3 emulator in voodoo is 3 times faster, but don't believe me. Benchmark it yourself. Link for voodoo sse3 performance analysis. There is also a very experimental patch in our source tree which downgrades the most used sse3 instruction to its sse2 roughly-equivalent opcode. This is about 12% faster than real sse3..

 

Thanks for the heads-up about VMware. Seems like the issue is unrelated to the kernel, but I'm still keeping an eye on it for the beta phase.

 

In case people missed it - the beta is going to be out tomorrow on the Germany Unification Day! Expected 1 week or a few days more of beta testing, and final kernel to be out around 10 Oct.

Thanks for the heads-up about VMware. Seems like the issue is unrelated to the kernel, but I'm still keeping an eye on it for the beta phase.

 

In case people missed it - the beta is going to be out tomorrow on the Germany Unification Day! Expected 1 week or a few days more of beta testing, and final kernel to be out around 10 Oct.

Yeah !

a MUST have : germanuniday_kernel :)

to non europeans: we krauts are one country - before that uni there was a west(cola_ed krauts) and a east(vodka_ed krauts) germany.

Hi,

 

My specs:

Pentium 4 HT 2.8GHz CPU

FSB 200 ratio 14

 

10.5.5 allmost vanilla install, changed only IOATA kext

using mach_alpha13, chameleon EFI

 

I have to set busratio to 28 to get normal spinning wheel behavior.

DVD Playback isn't possible. 100% CPU load.

 

restart and shutdown doesn't work.

 

everything else is working great.

 

DeskSystem Specs:

IBM IntelliStation M Pro 6230

Pentium 4 HT 2,8 and 3.00 GHz

1.5GB DDR 400MHz

nvidia GeForce FX5200 full QE/CI

CPUID_Log.txt

Voodoo 9.5 worked out really gud for me. I always had firefox crashes, probably due to HyperTransport for my AMD (I read this on other forums). Installing this kernel stopped all the crashes, don't know why, sorry, I'm not too techy :blink:. I didn't see much of a performance difference, but that doesn't matter much since its already pretty fast. Will post more results/bugs as I encounter them. ThanX for making this possible.

Got the kernel from here http://groups.google.com/group/xnu-dev/web...-0-kernel?pli=1

 

My specs:

AMD 4200+ dual core

AM2 (940) Socket

2 GB DDR2 667 Mhz

ATI X1650 Pro 512 DDR2

Leo 10.5.2 with EFI

×
×
  • Create New...