DIY eGPU experiences

Sure, I'll add in x1E benchmarks if I can get it working this weekend, no problem!

A couple quick feedbacks on the new DIYVidock Setup:

-When running "devcon" on Win 7 64-bit, you have to make sure to use the cmd.exe in C:\Windows\SysWOW64 if you want to use the 64-bit version of devcon. Even on Win 7 64-bit, typing "cmd" from the run/start menu gives the 32-bit version of cmd.exe, and the 64-bit version of devcon won't work.

-My monitor takes a bit to, um, "wake up," on a reboot - would you consider making the timer for menu selection a bit longer in the next version? 5 seconds goes by pretty quick!

Again, great release and thanks much! Really cool to read through and see all that has been accomplished by a couple dedicated individuals.

 

Just edit \config.sys and change the line "menudefault=RUN,5" where the 5 is 5 seconds to whatever delay time you like, eg: 10 for 10 seconds.

Great feedback on the 64-bit version of cmd. Thank you! I didn't even know that was an issue since i run Win7/32. Documentation has been updated.

OK - I've created a package you can use to overwrite the beta grub with the previous grub. See NOTE - BETA grub version here. If this older grub version presents problems still then it's a matter of working through the \core\menu.w7 contents until can get grub4dos to chainload on your system. 'help grub4dos' will bring up the full and lengthy grub4dos documentation to help you. Change 'debug 0' in \core\menu.w7 to 'debug 9' will give more debug messages.

Your system's 9800M GTS occupies 2x256MB (Cxxx.Dxxx windows). Compact can only free up 128MB with 4GB installed using 32-bit compaction. Nvidia allows 64-bit allocation so just need to move the 9800M to 64-bit space to create room for your attached HD5850 (1002 6899) in 32-bit space. Do that, with your HD5850 attached by running one of the following, then can just chainload your os with 'grub4dos w7'.

Code:

compact import \core\devcon.txt force32 1002 6899
compact pciend 310000000 import \core\devcon.txt force32 1002 6899 

Once successful, just add 'makebatch \core\pci.bat' to end of the correct commandline to get a setpci-based output file. So then can just do 'call \core\pci.bat' to setup your PCI configuration without needing to do any further runs of compact with the HD5850 attached.

 

If I understand you correctly, are you saying that my current configuration is set for PCI and not PCI-e? Or are you referring to something different when you say x1E?

I will try to find a pci-e expresscard to test...

 

tried the grub files that created. and it worked perfectly now. I tried running the x2 config and then chainloading it does it successfully. The problem is still in the resources allocation though. I tried both the recommended command you gave but the first one say no resource allocated while the second did say there was 512mb allocation for less than 4 gb but when i chainload it it hangs during windows loading screen. Would it be easier if i installed a 64bit os instead. Also i tried disabling the 9800m gts but it still hangs. Any suggestion as what would the next step would be. Thanks

editing: I also notice that even with 2gb of ram if i shut down my pc and then on it again and boot the card it say i don't have enough resources. Only after i restart after that will it work

 

Hooray, I won the 5870 for $366 including shipping.

 

congratulations. enjoy you improved performance.

 

Does that mean you have a HD5850 setup running at x2 using 2GB only?

Yes. Win7/64 is recommended since we are doing 64-bit NVidia allocation. Win7/32 might work, but I'm not sure if the NVidia driver is PAE compatible (Intel 45000MHD driver is).

So With your 9800M, you can only have a 128MB window after compaction in 32-bit space. So have two options:

1. (32-bit compaction - 'compact pciend 100000000') Try to get the 9800M to use only a 256MB window. For some reason it's using 512mb. Is it a SLI config? If so, can you remove one of the 9800M units to liberate 256MB?

2. (64-bit compaction - 'compact pciend 310000000'). Need to relocate your 9800M to 64-bit. Unfortunately when use compact with this commandline it can move other devices to 64-bit, which may cause your hang. So need to use 'force32 <VID/DID>' to force any of those other devices to 32-bit space.

It would be a matter of trial and error to get a working solution. Ensure you use 'import \core\devcon.txt' for all compact commands as well to ensure freeze-free operation.

 

yes i have tried both a 5770 and a 5850 with x2 on 2gb and they work perfectly except for the for the problem i mention above. It is not sli. its a single card.
just to confirm. The process is i boot the diy vidock boot usb.
i select option 3 which is the dos prompt
Then i type
compact pciend 310000000 import \core\devcon.txt force32 1002 6899
then once it is done i chainload.
I tried the 32bit compaction and i still got the no resource allocated mesage. With the 64 bit one i got the massage "largest empty pci range <4gib @c0000000 size 512.00mb" Then when i chainload as i mention earlier it hangs. Maybe i will try and see with a 64bit os. look llike i have to reformat my pc later to install the os.

 

Means the pci-e configuration space isn't being reset properly. Might need a run of compact to do it for you, even with 2GB installed. ie: 'compact pciend 10000000 import \core\devcon.txt'

The diags you posted were without the HD5850/HD5770 attached. Can you run 'diag' again with either attached and post?

Whoah! Can't wait to get some benchmarks.

Yes.. could try 64-bit allocation, allocating 9800M to 64-bit space and HD5770/HD5850 to 32-bit space with Win7/64. Try each of the following, then use 'grub4dos w7'.

Code:

compact import \core\devcon.txt force32 1002 6899 force32 1002 68b8
compact import \core\devcon.txt force32 1002 6899 force32 1002 68b8 prohibit F0000000 10000000

You want compact to return "1 solution found". 0 solutions means it couldn't allocate all the pci-e devices, with the HD5770/HD5850 missing out.

Otherwise to get 32-bit compaction going would need to shrink the 9800M's 512MB window to 256MB (from C000-DFFF to be just C000-CFFF) to free up a 256MB window to accomodate your HD5770/HD5850. Not sure why the 9800M has been set to use a 512MB window to begin with. Try both compact lines, whichever works. The one with pciend F4000000 does not relocate system devices.

Code:

:: 256MB Window for 9800M GS
setpci -s 0:1.0 24.l=cfffc000
:: perform PCI compaction but don't change 9800M's allocation
compact pciend 100000000 import \core\devcon.txt preserve 10de 0628
compact pciend F4000000 import \core\devcon.txt preserve 10de 0628

* Note: preserve above may still not allocate the HD5770/HD5850 if it still wants a 512MB BAR. Hmm.. trying to figure what the 9800M needs to set it to have only a 256MB BAR defined.

 

I'm currently encountering the same mapping problems with 4GB memory (X200, newest BIOS, Geforce 8800GTS 512MB; everything works fine with only 2GB memory installed), however using the DIYViDock 1.0b USB disk results in a crash (eip=extended instruction pointer):

Code:

'compact' binary crashdumped [...]
General Protection Fault at eip=0000aedf

I'm not using Windows and hence I'm unable to use "devcon" which I assume is the reason "compact" is crashing (no devcon -> no boundaries here). However I'm able to obtain the ACPI memory ranges via "lspci" but don't know how to represent them appropriately to get "compact" working. A "devcon.txt" example would be nice.

More details on Linux soon.

 

hi nando just to inform you that the devcon that you are using for the 64 bit is for Itanium processors. It can't be used with the later processors such as the core 2 duo. The correct version is available here
Sometimes, 64-Bit Is a Pain Visible Invisibles
Still trying to resolve the resource problem. Just installed the 64 bit os. Will try again to see if it works.

edit: here is the diag with the 5770 attached. Still no luck with resource allocation. i would remove the 9800m gts if only it weren't so difficult to remove and reattach

http://www.mediafire.com/?y14co2mzumj

edit 2: Finally managed to remove the 9800mgts and now i don't have the resource allocation problem. So it seem that the culprit was the 9800m gts. Currently i physically removed the card. But cant do this for a permanent solution . Maybe if there is a way to totally disable the card so it will automatically disable at start up so it will only use the external card.

 

Hi Nando
It seems that I have no luck with my Laptop HP Pavilion dv9925nr..I couldn´t relocate the the onboard ATI gfx to 64-bit space with command "compact"

compact pciend 310000000 import \core\devcon.txt force32 1002 9490
grub4dos w7

HP Pavilion dv9925nr.rar

what a shame Nando4!

 

It wasn’t as easy as I thought to get Windows XP (32 bit) up and running, but in the end I succeed. I now have some data from PCI Scope, but no idea what to look for. Maybe you can take a look? The data files are attached.

The system (Dell Studio XPS 1340, MCP79) has two internal graphic cards, the integrated one (9400M G) and the discreet one (9200M GS). They are connected thru a “hybrid SLI” solution. Bus4 (pcie port #4) seems to be PE4H (expresscard) and bus6 (pcie port #3) seems to be the wlan card.

I’m willing to take the chance if there is a little hope that my system can do x2. My biggest concern is the mini PCIe card itself. It looks like the wifi card is a "half length" mini PCIe card and PM3N is a full length card. I see that PM3N has four screw holes, maybe it can be shortened to fit? (see figure below, middle card)

 

See devcon.txt in rar diag.rar file here.

Thank you for pointing out x64 devcon. It will be packaged in next revision. The problem is that the 9800M is occupying SLI-like 512MB of PCI space as set by either the system bios or the NVidia bios. If that can be made to be 256MB (which it should) then problem solved.

Let's see if we can do it. Below we are allowing the HD5770 to be allocated *over* the 9800M's c000-dfff range (HD5770 should be placed at d000) and then quickly set the 9800M to use a 256MB@c000.

Code:

:: perform PCI compaction allowing HD5770 to use 9800M's c000-dfff range
:: HD5770 should be allocated to D0000000
compact pciend 100000000 import \core\devcon.txt ignore 10de 0628
:: set 256MB Window for 9800M at c0000000
setpci -s 0:1.0 24.l=cfffc000

Whoops. I think I see the problem. Remove the following section from your devcon.txt then try running the compact lines suggested earlier. The problem is since you are not using English language, devcon outputs the "PCI BUS" section as "BUS PCI" which confuses compact. compact ignores "PCI BUS" and uses all other items to prohibit. In your case it's using "BUS PCI" (which defines the full 32-bit PCI BUS ranges available) as a prohibit range so there are NO ranges to allocate to.

I made a mistake earlier. You have Nvidia onboard graphics. Try the following two tests then.

Code:

:: 64-bit compaction, forcing HD4670 to 32-bit
compact import \core\devcon.txt force32 1002 9490
grub4dos w7

:: If compact doesn't move 7150M to 64-bit then manually
:: move it from 32-bit D0000000 to 64-bit 300000000
setpci -s 0:12.0 14.l=0c
setpci -s 0:12.0 18.l=3
:: Perform 32-bit compaction. HD4670 should occupy D0000000
compact pciend 100000000 import \core\devcon.txt force32 1002 9490
grub4dos w7

 

@echo off
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
:: Source: template [USB]:\core\startup.tpl
:: Ensure pci.bat is the last line before grub4dos. call=run batch files
::
:: call %DRV%\scripts\igp_disb : disable IGP resources. [only use for XP!]
:: call ramdisk : Copy+run from R:\core. Ensures no freeze due to USB relocation.
:: call iport x2 [1 3 5] -check: set x2 port. Use -check for a true x2 link.
:: call pci : run setpci-based pci.bat to do PCI configuration
:: call grub4dos [w7|mbr|xp]': chainload OS. Runs from ramdisk if available.
::
:: NOTE: instead of 'call pci.bat' can run compact. This requires video card
:: attached. Eg: 'compact pciend 100000000', 'compact force32 1002 68BE'

REM call %DRV%:\scripts\igp_disb
call ramdisk
REM call iport x2 5
call pci
:: 64-bits compaction, forcing HD4670 to 32-bits
compact import \core\devcon.txt force32 1002 9490
grub4dos win7
:end

This is what I typed in Startup.bat, and also i tried with the other test, Am I right? I have doubts because when I load startup.bat in the VIDOCK menu, my laptop freezes in both cases

I was wondering about changing the language to English in my Windows7, could I solve the problem with this solution??¿??

Vidock.rar

 

With the removed "BUS PCI" section compact is actually able to perform PCI allocation. Boot using "DOS based" to do testing at the commandline followed by 'grub4dos w7' if a solution is found. If successful, add your commands you use at commandline to \core\startup.bat to automate the boot. Choose either 'compact' or pci.bat, not both. If using pci.bat you'll need to generate a new pci.bat from your working compact line using extra parameters "makebatch \core\pci.bat". Using pci.bat means you don't have to have your HD4670 attached during the bootup process.

MCP79 x2 capability

MCP79 pci-port 3 link status/control registers are 8c, 90, 92. Can have a play setting data for link width/speed to see if the port responds. Likely there is some master port configuration byte somewhere that sets up the pci-e ports to do x2 (x8,x8,x2,x1,x1 setup) as indicated here. NVidia will not provide the public with a programming guide to explain how to set x2 mode if it happens that the pci-e ports are still set to x8,x8,x1,x1,x1,x1.

PM3N is designed to be snapped in half to make a half-height card, so will physically fit. If you did go ahead and buy the x2 parts (PM3N and another mini HDMI cable) then you'd be hopeful the bios or chipset can autodetect that there is a x2 lane available. The system can

 

the resource allocation is still not working. It seems that even in a 64 bit os the 9800m gts does not want to give up its allocation. It still hangs when ever the resources are removed from the 9800m gts. is there any other device that can have its resources reallocated to the external card other than the 9800m gts. like maybe the resources from the wifi or something.

 

Hi nando4, i got some problems here...

1. I am using win7 64 bits, but i cant generate a devcon.txt from '\devcon\ia64'
I dont know if i can use a devcon.txt generated in '\devcon\i386' or it must be 64bit.

2. In anycase my system language is spanish. So i understand that i need to remove the 'Bus PCI' section from devcon.txt. Or not?

3. You are right, i am using an Nvidia onboard graphics and i have tried these tests:

Code:

:: 64-bit compaction, forcing HD4670 to 32-bit
compact import \core\devcon.txt force32 1002 9490
grub4dos w7

:: If compact doesn't move 7150M to 64-bit then manually
:: move it from 32-bit D0000000 to 64-bit 300000000
setpci -s 0:12.0 14.l=0c
setpci -s 0:12.0 18.l=3
:: Perform 32-bit compaction. HD4670 should occupy D0000000
compact pciend 100000000 import \core\devcon.txt force32 1002 9490
grub4dos w7

The results are....

Test 1: ) The laptop freezes after grub4dosw7

Test 2: ) Windows 7 freezes with Vidock 2 connected.
Without attaching, windows 7 starts but it doesn´t recognize Vidock2

Is Test 2 the right choice? - i guess it is -

Please verify your private messages

 

Hi Nando4

I sent you the results from PCI.Bat

PCI.BAT

 

Did you have the HD4670 attached when you ran compact? Anyway, can try the following to see if your 7150 works successfully in Win7 when using a 64-bit bar. I know 9600GT works, but not sure if the 7150 will.

Code:

setpci -s 0:12.0 14.l=c 18.l=3
grub4dos w7

 

here are PCI.bat and out.txt

OUT.rar

 

Please report with HD4670 attached. Would be a good idea to use add "import \core\devcon.txt" as well. So your commandline would be " compact pciend 310000000 force32 1002 9490 import \core\devcon.txt makebatch \core\pci.bat > out.txt'. x64 compatible devcon posted here.

I don't think it necessary for us to do this troubleshooting in this thread. We have PM communication which we can troubleshoot this further with. Please PM me the out.txt and pci.bat.

 

DIY ViDock Setup v1.0c released. I won't be announcing further releases as a post. Instead I will just add a new entry in the NEWS section on the first page.

 

wow did you guys see the engadet link? They are giving all the credit to HardOCP and that other guy.

PCIe Radeon HD 5850 connected to Lenovo T410s, great gaming ensues -- Engadget

Tip Endgadget and let them know that nando4 and moral hazard and who ever else was involved in the processes gets credit for there work.

 

There is a new rendering software called Octane that uses Nvidia CUDA capable GPUs instead of traditional CPUs.

Here is a link to the soft:

Octane Render - The Interactive Physically Based Render Engine

Your experiences with external cards on laptops caught my eye, and I keep wondering if this kit with the PE4H would do the trick for me as well.

Looks like it should be even easier, as I intend to keep the internal Geforce 8600m GT to drive the laptop display, and use the external gtx 470 only for calculation purpose.


Do you think it would work?

Could any caritative soul with an external nvidia card download the demo and give it a try?

Thanks in advance,

Albert

 

actually if you read my thread over at hardforum, you'll see i directly credit NBR forums and specifically credit nando4 for his awesome work on the booting issues:
So I just hooked up a 2GB 5850 to my laptop... - [H]ard|Forum

still working on full benches for a detailed post here - posted early over at the [H] cause I know they go bonkers for pretty pictures. no idea how it ended up on engadget, however - i had nothing to do with that.

 

Can you do x1E versus x1 benchmarks? I know you have a PM3N so could do the x1E bench using mPCIe port1. You might need the 'iport whitelst save/restore' workaround using DIY ViDock Setup [USB bootdisk].

Intel's x1 1.0 is not delivering 2.5Gbps full duplex - it's underperforming

Your performance benchmarks could then highlight that Intel's pci-e ports are not working to 2.5GBps in each direction as explained in the x1E workaround and here.

Intel could then explain how to get proper x1 1.0 performance (x1E) on any port. The workaround applies only to odd ports so some systems miss out. x1E performance is noticable better, NFS Shift is definitely more playable. Would be moreso with your 2GB card.

 

Has anyone tried this?

Using the ATI card to power the graphics and the built-in Nvidia card to run PhysX would be very nice.

 

I don't think the notebook GPU and external GPU can interact?

 

Im going to buy the PE4H with PM3N adapter but I dont know what Graphics cards are supported. I have a built in NVIDIA GeForce 8600M GT.
is using the minipcie the best choice or should i connect it throught the express card slot? I understand theres a little bit of speed boost but either way I have no clue what cards I can get to work.

 

I just saw the post on engadget.com and the link to this forum. I am using an loving my Lenovo Thinkpad X201 but I hate the built in graphics...

So I just orderd the HWTools PE4H v2.0 PCIe Adapter and EC2C ExpressCard Adapter.

Now I am not sure which desktop graphics adapter I should use...

Does it make sense and does it work if I would go for a ATI HD5970 or NVIDIA GTX 480?

Or should I stick with a normal one ATI HD5870?

Thanks for your help guys

 

X200 has turbo memory slot=port3, expresscard port=4, so could do a x2 link using port3+port4 if you wanted to use it that way. In that case it would be worth considering the faster cards. Might encounter Lenovo whitelisting issues with port3, but that's probably solvable. You'd need another some more parts.. add a PM3N from hwtools and another mini hdmi cable to do x2.

We have no benchmark as to how much pci-e bandwidth the GTX480 uses. A HD5870 is probably more efficient with bandwidth usage based on the GTX260M results we have.

We have a user doing a P8400-2.26 HD5770@x2 and HD5850@x2 implementation, so will hopefully have some comparative benchmarks soon.

 

But the miniPCI slot method for x2 is really untransportable... opening the notebook and so...

With the ExpressCard Solution... I shutdown Win7 or Debian remove the ExpressCard boot the other kernel without the modification and I am good to go, or do I get there something wrong?

So you would recommend me to test it with a HD4970 because of the bandwith? Because the bandwith useage on the GTX-480 is not that good?

Or if I am using X1 taking these hardcore cards spoiles the point?

By the way does it matter if I get Version 1.0 or Version 2.0 of the HWTools PE4H?

 

Great read! Thank you all

I currently have a Lenovo X61 tablet with the following specs:

• Intel Core 2 Duo 1.4 GHz (L7300)
• 2048 MB RAM
• Intel GMA x3100 / 965 (same thing?)
• ExpressCard-slot

And I am about to buy the PE4L v1.4 with EC2C in attempt to use my Sapphire HD4870 1GB on my Lenovo.

Do you guys think it would be possible (considering the guys with the x200 managede to do it)? Or is there anything I will have to reconsider?

I have several PSUes laying around and know the paperclip-trick, so no problem there.

Thanks again for all the cool stuff posted here,
-Stefan

 

At x1 1.0 the pci-e bandwidth will be the bottleneck. The best you can do is get a pci-e efficient card (a HD5xxx series) with as much VRAM to give some bandwidth sparing effects (eg: 2GB HD5850). We don't have benchmarks to show how much increase in performance the 2GB will have. The good bang-per-buck solution would be a HD5770-ATX or HD5750-12V/80W solution.

PE4H 2.0 is more desirable, as has 4x mini HDMI inputs and no reworked fuses.

Your setup is no problems at all. 2GB ram, expresscard. X61t lspci output shows pci-e port1 and port2 enabled by the bios. If can confirm port1+port2 correspond to mPCIe and expresscard slots then could even consider a x2 link ganging the two links together for superior performance. DIY ViDock Setup USB Bootdisk can overcome whitelisting if port1 is your wifi port.

 

Sounds awesome! Thank you for quick response. I think I will stick with the x1 on the PE4L since my cpu (1.4 GHz... ) will probably bottleneck before the limits of the PCIe x1 is reached. Am I right?

EDIT: After a quick search on Google I started to doubt whether or not my tablet does in fact have the ExpressCard-slot. E.g. it says: "1 Type I / II PC Card slot with 34mm ExpressCard via adapter". Is there any way to check it?

 

Is this important for the performance the 4xmini HDMI and the reworked fuses? Because they emailed me that they are currently out of stock with the version 2.0

 

For your purposes, no. PE4H 1.0 will have the right fuses and the 2 mini HDMI connectors if want to do a x2 link. There is like 1% of notebooks that could do a x4 link and likely less folks that would want to have 4 leads coming out of their notebooks. So x4 is more for show than go.

 

nobody wants to help with my renderer thing?

it's a very straightforward test..

 

Thanks a lot for your help.

Other question. I use Terabyte BootIt NG right now as an OS Selector. I have serveral Windows XP / 7 / 2008 Server / Linux / MacOS installations on my X201. How difficult will it be to modify them to use the external graphics card?

 

I will take a look at it tonight. I have an on board ATI Mobility Radeon™ HD 4570.

for testing my 1x link i have the PE4L . due to the location of wireless card and no extra connections.i have my Nvidia 9800gt 1 Gig card in it.
Performance is not the best. 3dmark shows no improvement but RE5 and DMC4 benchmarks show the improvement for the extra horsepower of the add in card.

I will post back tonight with some info.

 

thanks a lot, man..you saved my ***.

all you have to do is download the demo, install, launch the soft and select the external card -your 9800gt card- under:

FILE
--DEVICE MANAGER
----ACTIVE CUDA DEVICES

You should be booting with no special bios or modifs, and your mobility radeon should be driving the system as usual; this way, your radeon card should be listed as "Inactive" in the DEVICE MANAGER.

On the same page you download the demo, there is a demo suite with models. Opening any file with .ocs extension will do.

Once you open a file, you will see a little tag in the screen saying:

"blablabla... (CLICK ME) "


THAT'S IT! If you see an image appear in the screen, that means it works!!

 

It depends on if you X201 requires the DIY ViDock Setup [USB Bootdisk] for any aspect of the configuration of the DIY ViDock (error 12, switching x1E/x2, anti-whitelisting).

If not, then can boot straight in. If it does require it, then the DIY ViDOck Setup can chainload your MBR which will then run Terabyte BootIt NG (TB-NB). You may be able to merge the two so don't need a USB stick by creating a boot floppy image to do your PCI configuration, adding if to TB-NG, or by swapping TB-NG altogether for Linux's grub2 (0.98 or newer) which has a setpci command AND can do memory writes (write_xxxx), so can transcribe the commands the DIY ViDock Setup is doing.

Can you post your details of your P7350 Toshiba A505-S6965 + 9800GT setup so can be added to the first post? We have very little NVidia benchmarks/experiences to draw any performance comparisons from.

 

> sunlooksgreen

ok the good news. It sees the 9800gt Nvidia card and the program works, i selected the card from the list and then selected one of the models from the folder of samples. the chess set was first and then, space ship was select. Now onto How well it works, i know nothing about the program so i have no way to know how long it would take to "render ? " the scenes . it was running about 4fps on the tab and it let them run for about 5 minutes each +/-. i would think that with the 9800gt it might take an hour.

anyway yes it sees the card and the program runs, oh and my CPU load was at 50% while the rendering was running.
without the Nvidia card connected to the add in card the program will not launch. so it looks for the card as soon as it loads.

 

GREAT NEWS! THANKS A LOT!

It was very kind on your part!!!

As for performance, on the examples folder there is a "benchmark scene". If you load that, and change the lighting method from "directlight" to " pathtracing" under "Preview kernel" tab (last icon below Node Inspector), that should give me a good reference point.

The value to look for is Megasamples/sec (next to FPS)

 

HWTools wrote me that my PE4H-EC2C will be shipped tomorrow. I also just orderd a XFX Radeon HD 5970 2.0GB Black Edition.

Can somebody advise me what power supply I need?

Regards,
Sonic

 

I'd advise against a twin-gpu card. It needs a 512MB window (256MB for each gpu) in 32-bit PCI space. Unless your notebook bios has a low TOLUD it probably won't work with a 4GB setup. (might work OK with 2GB).

A 2GB HD5850/5870 is a cheaper option that only requires 256MB window in 32-bit PCI space and will probaby perform better, since the single gpu has access to the full 2GB.

For PSU requirements, each 6-pin PCI-e power connector needs 75W, a 8-pin requires 125W and the slot needs 75W. So just add those up for total max power. Then just ensure the PSU has a floppy molex connector as well as either molex to attach pci-e<->molex adaptors or direct pci-e power connectors, which will be pretty much any ATX PSU.

A HD58xx will still be bandwidth limited in a x1 1.0 setup, so consider if you can do x2 1.0 by combining ports or at least x1E by ensuring your expresscard slot is an odd numbered port with nothing attached on the next sequentially numbered port. See x1E/x2 port details.

 

I just called my supplier and ask if I can change. He told me best single GPU is the GTX480? Should I get this one?

Do you think the NVIDIA GTX 480 is faster than the HD5870?

Regards,
Sonic

 

Can you post your Everest pci-e port screenshot like this and the PCI BUS section in Device Manager like this. What onboard graphics do you have?

We don't know. I know a 2GB HD4870 is slower than a 1GB HD5750 due to the 5750 using less pci-e bandwidth as shown here. On a x16 2.0 link a HD4870 benchmarks at HD5770 levels or slightly faster.

WE don't know how pci-e bandwidth efficient the GTX480 is. If I was to take a guess, I'd say ATI cards are efficient. One advantage of the NVidia desktop cards is their Win7 driver works with 64-bit pci-e PCI resource relocation, so may work when an ATI card wouldn't. Which is why I wanted to see your 32-bit PCI bus memory map OR know if you have 4500MHD or better Intel onboard graphics (which is 64-bit relocatable) to ensure an ATI solution would work with 4GB or more of RAM.

 

In class now, i will post the results tonight. might try to add a screen shot with the info also.