Windows XP Ram Limit - Page 4 - Overclock.net - An Overclocking Community

Forum Jump: 

Windows XP Ram Limit

Reply
 
Thread Tools
post #31 of 101 (permalink) Old 03-05-2014, 08:53 AM
New to Overclock.net
 
achm3t's Avatar
 
Join Date: Feb 2014
Posts: 10
Rep: 0
Finished testing on other hardware with 8GB of RAM:
German XP Pro SP3
+English ntkrpamp.exe SP3 5.1.2600.5512
+German halmacpi.dll SP1 5.1.2600.1106
works like a charm (I can see and use the full 8GB of memory)!

using German halmacpi SP3 5.1.2600.5512 ->crash during boot process (as we expected to happen)

So
conclusion 1: I can verify that kondras patch is actually working! (which is really really really great!!!!)
conclusion 2:You can put an english patched SP3 Kernel into a localized XP and the system is still working fine!!! (tested at least with German localization)


So next thing for me is to try to upgrade halmacpi to a level that does not suffer from USB problems any longer (either by intermediate SP1<->SP2 versions) or by using kondras proposal SP3halmacpi+rxvstor.inf!

After that I will try to patch newer ntkrpamp - not decided whether I will use english version (for generality) or german...
As far as I know, there shouldn't be any Kernel display-outputs to the GUI, so maybe patching the localized kernel is a waste of time....
achm3t is offline  
Sponsored Links
Advertisement
 
post #32 of 101 (permalink) Old 03-06-2014, 01:51 AM
New to Overclock.net
 
friendship7's Avatar
 
Join Date: Mar 2014
Posts: 9
Rep: 5 (Unique: 2)
Found a proper solution!
Using the HAL dll from SP1 caused BSOD in my case, and it sounded like a bad idea to begin with.

The problem with XP SP3 HAL is that when you have physical addresses > 4GB, the DMA transfer data may be copied to the wrong memory address. You can easily apply a quick and dirty fix to XP SP3 HAL DLL as described here:

http://vm1.duckdns.org/Public/Windows_XP_SP3_Remove_PAE_Limit/Windows_XP_Remove_PAE_Limit.htm
friendship7 is offline  
post #33 of 101 (permalink) Old 03-06-2014, 09:25 AM
New to Overclock.net
 
achm3t's Avatar
 
Join Date: Feb 2014
Posts: 10
Rep: 0
Awesome folks within this forum!

friendship7, I am not sure whether I understand things right:

Does this mean, using the patched hal will result (A) in a slight performance impact only when using this on a machine with <4GB (since double buffering will be used unnecessarily) or (B) will give a slight performance impact for addresses <4GB compared to kondras solution with SP3halmacpi+rxvstor.inf?

In other words: Is this check performed per memory pointer or once during boot?

Because I think this solution seems cleaner to me than using an additional driver within your setup that fixes the SP3halmacpi issues "by accident"...

Well as clean as a hacked kernel can be *lol*

Did somebody process some benchmarks, e.g: Sisoft Sandra Memory or Winrar (which is also very sensitive on RAM latency and bandwith) comparing
32bit XP SP3 out of the box
vs
kondras patch + SP3halmacpi + rvxstor.inf
vs
kernel + patched SP3halmacpi from friendship7
?

Sounds very interesting to me.


However I think the intermediate SP1<->SP2 hal's are out of the race now, the two known alternatives are much more appealing - both! USB issues -> solved

What do you think? This boot.ini will not force you to workaround SFC(WFP) and risk file replacement due to Windows Updates.
It may also work without fixing the PE header (not verified)

[boot loader]
timeout=5
default=multi(0)disk(0)rdisk(0)partition(1)\WINDOWS
[operating systems]
multi(0)disk(0)rdisk(0)partition(1)\WINDOWS="Windows XP OutOfBox" /noexecute=optin /fastdetect
multi(0)disk(0)rdisk(0)partition(1)\WINDOWS="kondra" /fastdetect /PAE /kernel=kkondra.exe /HAL=hkondra.dll
multi(0)disk(0)rdisk(0)partition(1)\WINDOWS="friendship7" /fastdetect /PAE /kernel=kfriend.exe /HAL=hfriend.dll
achm3t is offline  
Sponsored Links
Advertisement
 
post #34 of 101 (permalink) Old 03-06-2014, 10:17 AM
New to Overclock.net
 
friendship7's Avatar
 
Join Date: Mar 2014
Posts: 9
Rep: 5 (Unique: 2)
achm3t,
for systems with more than 4GB RAM, this patched XP SP3 HAL means that each time a driver that supports 64-bit addresses will perform a DMA transfer, double-buffering will be used unnecessarily.
Note that >4GB systems always require double-buffering for drivers that do not support 64-bit addresses.

for systems with 4GB or less, this patched HAL will also use double buffering for 32-bit addresses, where it's also unnecessary.

keep in mind that memory-to-memory copy is extremely fast, and the performance impact of double-buffering is very small.
friendship7 is offline  
post #35 of 101 (permalink) Old 03-06-2014, 10:27 AM
New to Overclock.net
 
achm3t's Avatar
 
Join Date: Feb 2014
Posts: 10
Rep: 0
Thank you for your quick and clarifying reply.
So the performance penalty will appear for systems having less than 4GB (which is ok, we do not need the patched hal for them - they won't even crash with modified kernel)
and for all systems: system drivers that are capable of handling 64bit addresses (which is a pity)

I understand that memcpy is really fast - but would still be interested whether this can be measured by any synthetical or real-world benchmark (I am just curious).
This is interesting especially compared to the unpatched SP3halmacpi + rxvstor.inf workaround (kondras actual solution)...

Still got one question:
When I do binary comparison of the patched SP3 kernels, I can see that (let PE checksum aside) frienship7 patches 2 locations, but kondra patches only a single location - why this difference?
achm3t is offline  
post #36 of 101 (permalink) Old 03-06-2014, 11:52 AM
New to Overclock.net
 
friendship7's Avatar
 
Join Date: Mar 2014
Posts: 9
Rep: 5 (Unique: 2)
achm3t,
The XP SP3 kernel has two limits: physical address limit and RAM limit, that's why two locations are patched.

Regarding measurements: you can simply benchmarking an SSD read/write speed, but I suspect the cost of the additional memory copy operation would be almost negligible.
Writing to the GPU memory would be more interesting, because it's much faster than SSD and will put more strain on DMA transfers, so any additional cost will have more impact,
there are some programs that can benchmark writing to the GPU memory using Open-CL / CUDA, I'd be interested in the results.
friendship7 is offline  
post #37 of 101 (permalink) Old 03-06-2014, 12:04 PM
New to Overclock.net
 
achm3t's Avatar
 
Join Date: Feb 2014
Posts: 10
Rep: 0
Quote:
Physical address limit and RAM limit:
What's the difference between them?
kondras patch does not touch location 0x15df20 in ntkrpamp, but is working too...
Is physical address limit the per-user-process-limit of 2GB?
Quote:
benchmark writing to the GPU memory using Open-CL / CUDA
Does anybody know a specific benchmark which is targeted to GPU DMA?
achm3t is offline  
post #38 of 101 (permalink) Old 03-06-2014, 12:14 PM
New to Overclock.net
 
friendship7's Avatar
 
Join Date: Mar 2014
Posts: 9
Rep: 5 (Unique: 2)
achm3t,
Physical address limit = A user with a dedicated GPU with 1GB of RAM, could only utilize up to 3GB of system RAM (physical address limit of 4GB)

GPU memory write benchmarks:
AIDA64
fillratetest
friendship7 is offline  
post #39 of 101 (permalink) Old 03-06-2014, 02:06 PM
New to Overclock.net
 
friendship7's Avatar
 
Join Date: Mar 2014
Posts: 9
Rep: 5 (Unique: 2)
achm3t,
My mistake, the additional location relates to the construction of the PFN database and nonpaged pool,
it is constructed differently when support for 64GB is enabled,
it's hard to tell what the effect will be if only the first location is patched.
friendship7 is offline  
post #40 of 101 (permalink) Old 03-07-2014, 04:36 AM
New to Overclock.net
 
friendship7's Avatar
 
Join Date: Mar 2014
Posts: 9
Rep: 5 (Unique: 2)
Ran both fillratetest and AIDA64's OpenCL Memory Read/Write/Copy, could not find any consistent decrease in performance between the original XP SP3 HAL and the patched one.
After a review, I came to the conclusion that the patched HAL will simply allocate map register buffers when it's not necessary (and only during initialization).
double-buffering will only be required when writing to physical addresses > 4GB.
friendship7 is offline  
Reply

Quick Reply
Message:
Options

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in



Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off