Let's figure out how to make a DIY eGPU (previously DIY ViDock)

Quite impressed, but I'd like to know what settings he used. The settings for the vidock test are
1280x960
AA - off
motion blur - off
shadow - med
texture - low
overall - med

 

Thanks.

And +rep for you as well, for all your work in this thread.

Lol, a genius? Thanks!

Yeah. Exactly.

----------------------------------------
UPDATE: I have PM'ed someone on the forum with the ViDock, asking him to test this theory. I will post back the result as soon as I know myself.
.
UPDATE 2: He has responded and is going to do the tests as soon as he can.
.

 

Guys... it doesn't really make sense for the ExpressCard slot and the miniPCIe slot share bandwidth... I don't think that's true.

And btw, since I see that you guys are discussing opening up the laptop to hi-jack some extra PCI-e lanes.. wouldn't it just be easier to hi-jack the docking station port? I guess that would only work for laptops with docking capability though.

But I would suggest not to get that ambitious too quickly. It's better to get a baseline scenario working, and then move on from there. I think we should focus more on getting the 1 card connected with the 1x PCI-e from the expresscard slot.

 

I think they've already succeeded with that http://forum.notebookreview.com/showthread.php?p=5122549

 

Does the docking station port have PCI pins?

 

But connecting it with X2 would require it being plugged into Expresscard and Mini-PCIe at the same time right?

If no one else does, when the PE4 is available and i build it, i will test weather they share the bandwith.
I will be building it when the PE4 is available. And will run a ton of tests and all that.

The hardest part seems to be finding something to put it all in... that will hold the PSU and GPU.

Also, do you think getting the 100cm cable vs. the 30mm. cable with the PE4 will hurt the bandwith/performance, because of the longer cable?

 

Ok Ill buy this tomorrow since the PE4 is the same thing... I cant wait to have it and test it!!!: .

Well for the power adapter I plan to use something like this:
http://shop.ebay.com/i.html?_nkw=12...trksid=m270&_odkw=12v+5amps+adapter&_osacat=0

Later Ill buy the PE4...

 

Edit - Summarizing about 6 of my posts here:

-Converting ExpressCard into PCI-E is much like converting SATA to e-SATA. There is no signal processing which needs to be done, we're just changing the shape of the connector and re-arranging the wires. In other words, the cost should be very minimal and we should make sure that we don't get over-charged

-PCI-e 1X has 2.5Gbit of bandwidth, which translates to 300MB/sec. However there is control signal overhead which drops the effective bandwidth down to 250MB/sec. Here is a comparison with old-school AGP:
AGP 1x: 133MB/sec
AGP 2x: 533MB/sec
AGP 4x: 1066MB/sec

-A tip on using our limited 1x bus more efficiently: using a card with more onboard memory will reduce the number of times which the card will access the PCI-e bus to load more data/textures into memory. Therefore, for example, getting a card with 1GB of memory could be better than 512MB. The gain here should be more than what it would have been with a 16x bus.

-I pointed out that this will likely be much cheaper than those over-priced USB DVI adapters, so this isn't necessarily for gamers only! And needless to say, this will perform much better..

-This will NOT work in Windows Vista.. even vidock2 does not work in Vista because it does not allow having 2 video drivers installed simultaneously. This project will only work with Windows 7!

- The Expresscard slot has both a 1x PCI-e lane, and a USB 2.0 connection. So we are adding a USB port + hub without sacrificing any bandwidth from the video card!

- Docking station ports likely use PCI-e, so for those of us who have business-line notebooks from Dell/Lenovo/HP/etc. *might* be able to find a way to hijack the docking station port, if we managed to figure out the pin-out.. Considering that these docking stations provide dual video outputs, eSATA, gigabit ethernet and a bunch of USB ports, it seem like the only way this can be done is with PCI-e.

Wow nice, that's a huge chart! Very useful, thanks

I think we really need a page to store all this information in, cause this thread is just so large it's getting out of hand lol

 

Faruk, if u said that Vista does not allow having 2 video drivers installed simultaneously, then can I just delete the notebook graphic driver and install the video card driver that I bought?

 

Unfortunately no, because you're still going to have some sort of "default" driver installed for your laptop display to work.

The only way you can *probably* get around this is if you have an ATi card in your laptop, and then you get an ATi external video card. So then you just need to have ATi Catalyst installed.

Or similarly, if you have nVidia in your laptop, and you get an external nVidia video card, you would just need to have the nVidia Forceware driver installed.

If you have Intel integrated graphics, then you and I are in the same boat - you gotta have Windows 7

According to the vidock page, Windows XP is supported, but only Windows 7 supports the hot-pluggable feature. (i.e. plugging it in and out while the laptop is on)

 

But, depending on what card you got (4850 right?), 12v/5A is only 60 Watts. Now, after you add the pci-e power connector, that adds 75 watts. But in the end, thats only 135. Which isnt enough to power a 4850, and simmilar cards. That can take up to 200 Watts+... :/

 

Correct, but also Ill power the card with a PSU, Ill use the adapter to power up the PCI E lane and the PSU to power up the 6 pin adapter of the gfx card.

These cards use less than 150W, also a "real" PCI E lane adds only 60w like this adapter, the card will take the needed watts from the PSU by using the 6pin adapter.

BtW Ill use this connector that feeds 150W to the card.

http://cgi.ebay.com/Molex-4pin-to-8...L?hash=item5d257d8add&_trksid=p3286.m63.l1177

 

I got 2 good news and 2 bad news .

My HD4850 is here!!!!!! (this thing is OMG HUUUUUGE).
The bad thing is that it uses a 8pin connector , Ill have to buy one...

Also I just bought the PE2L using a friend´s paypal account
The bad thing is that his shipping address was wrong but we emailed HWTOOLS telling them about the issue and giving them the real shipping address, hopefully they will answer soon.

 

so is there any difference in performance in the pe2l vs the pe4?

 

In theory no, there wouldnt be any difference since both work at X1.

 

No.. the PE4 will just have a USB port and a molex connector that's all. Performance will be the same.

And seriously, we either need to clean this thread up, or we need to make a new thread to just contain the facts/summaries, so that we can move on our discussion in here. The 25 pages will be pretty discouraging for newcomers, and people often end up jumping in and ask questions that were answered earlier on in the thread. Or even if you do go through the whole thread, it's pretty easy to miss some things.

 

Hmm. But if you look HERE, it shows the 4850 using up to 236 watts on load...

And AFAIK, cant the PCIe power connector only provide 75watts?
This makes no sense...

Molex connector seems pretty important to me...

With the PE2L, you need to also buy a DC Adapter. And with a power supply, that starts getting to be alot of stuff altogether.
With the PE4, you'll be able to only use a PSU to power everything. no need for another adapter. I find this a big plus, no?

 

Reading that i would say thats the total system load, not just the GFX card. I believe a 4850 pulls 110 watts max, though the amps and a stable rail are far more important. If your planning on building around a 4850 would you simply not be better off with a 4770? its 40nm creates less heat, uses less power and matchs the 4850 for performance.

Btw came across this while searching for something else, another product said to do the same thing as Vidock http://www.magma.com/products/pciexpress/expressbox1/index.html

 

yeah I'v seen that, costs $730 lol.

 

Hi
I was boring and I did some test with HD3850 256mb and HD4870 1gb comparing bandwitch on X16 2.0 - X1 2.0 and X1 1.0 on this gpu's.
Test's were done on desktop but to make it more similiar to laptop's spec I downclocket it. Processor E8400@ 2.4GHZ FSB =266MHz RAM 4Gb 800MHz. First Resident Evil 5 benchmark:
It's done in two resolution 1.1280X800 and 2.1680X1050 anti-aliasing Off, all details on High, DX9 ( windows Xp SP3).

HD4870
X16 2.0, X1 2.0, X1 1.0

 

Hd3850
X16 2.0, X1 2.0, X1 1.0

 

I assume that frames per second on X16 2.0 = 100% and count loose of performance on other bandwitch.
Compare HD4870

HD3850


And I did Crysiss benchmark too. All details on high in res 1280x960 and 1680X1050. I leave in results only most important : res, Average FPS, min and max FPS and cut rest.

HD4870 x16 2.0

DX9 1280x960, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 53.11s, Average FPS: 37.66
Min FPS: 16.68 at frame 151, Max FPS: 60.90 at frame 876
DX9 1680x1050, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 61.03s, Average FPS: 32.77
Min FPS: 22.97 at frame 138, Max FPS: 47.50 at frame 976

HD4870 X1 2.0

DX9 1280x960, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 61.19s, Average FPS: 32.68
Min FPS: 20.52 at frame 139, Max FPS: 47.84 at frame 869
DX9 1680x1050, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 73.40s, Average FPS: 27.25
Min FPS: 14.91 at frame 145, Max FPS: 37.94 at frame 993

HD4870 X1 1.0

DX9 1280x960, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 99.47s, Average FPS: 20.11
Min FPS: 4.80 at frame 137, Max FPS: 28.71 at frame 1006
DX9 1680x1050, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 115.62s, Average FPS: 17.30
Min FPS: 11.82 at frame 146, Max FPS: 25.62 at frame 993

HD3850 X16 2.0

DX9 1280x960, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 74.52s, Average FPS: 26.84
Min FPS: 13.66 at frame 279, Max FPS: 36.56 at frame 987
Average Tri/Sec: 27370072, Tri/Frame: 1019774
DX9 1680x1050, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 99.56s, Average FPS: 20.09
Min FPS: 9.97 at frame 94, Max FPS: 24.82 at frame 1003

HD3850 X1 2.0

DX9 1280x960, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 106.29s, Average FPS: 18.82
Min FPS: 7.82 at frame 1954, Max FPS: 27.08 at frame 984
DX9 1680x1050, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 223.51s, Average FPS: 8.95
Min FPS: 3.68 at frame 1941, Max FPS: 18.84 at frame 94

HD3850 X1 1.0

DX9 1280x960, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 181.61s, Average FPS: 11.01
Min FPS: 0.95 at frame 1981, Max FPS: 18.88 at frame 132
DX9 1680x1050, AA=No AA, Vsync=Disabled, 32 bit test, FullScreen
Global Game Quality: High
Play Time: 414.20s, Average FPS: 4.83
Min FPS: 0.00 at frame 1981, Max FPS: 14.65 at frame 91

And offcourse I compare this all in the similiar tables like before
HD4870 1GB and HD3850 256MB

 

Yes pcie X1 is very limited but going to X2 make thing much much better and it's really worth to play. Besides sorry for maby little to long and to much writing but I think that it's really usefull compare those two cards; one with only 256MB ram- we can call it middle class card and hi end 4870 with 1GB.

 

darnok44 - VERY useful information summarized below from your screenshots. Many thank yous for this!

pci-e scaling analysis

HD3850-256MB and HD4870-1GB compared at x16 2.0, x1 2.0, x1 1.0

Game	Resolution	Video card	x16 2.0	x1 2.0 (x2 1.0)*	x1 1.0
Resident Evil 5	1280x800	HD3850-256MB HD4870-1GB	71.6 84.6	57.0(-20.4%) 72.7(-14.1%)	34.0(-52.5%) 33.8(-60.0%)
	1680x1050	HD3850-256MB HD4870-1GB	53.0 78.5	43.1(-18.7%) 61.3(-21.8%)	28.1(-47.0%) 30.9(-60.6%)
Crysis	1280x960	HD3850-256MB HD4870-1GB	26.84 37.66	18.82(-26.1%) 32.68(-13.2%)	11.01(-59.0%) 20.11(-46.6%)
	1680x1050	HD3850-256MB HD4870-1GB	20.09 32.77	8.95(-55.5%) 27.25(-16.8%)	4.83(-76.0%) 17.30(-47.2%)

Source: darnok44's results here
* = equivalent to expressport 2.0 performance
Much more RE5 benchmark results here.

We see why there is little REAL ViDock2 performance data out there. x1 1.0 mode has 46-75% performance loss compared to x16 2.0 mode. We can generalise by saying x1 1.0 will give 50% of 16x 2.0 performance levels.

We see that in Crysis a HD4870-1GB at x1 1.0 performs better than a HD3850-256MB at x1 2.0. I would assume this is due to it's higher amount of vram requiring less pci-e bandwidth to move 3D textures to/from the card once loaded. I believe repeating that benchmark with a HD3850-1GB (more vram) in x2 mode would see it outperforming HD4870-1GB in x1 mode.

In Resident Evil 5, there is little incentive to go a HD4870 over a HD3850 at x1 1.0 speed, but there certainly is at x1 2.0 speeds. A HD4870 in x1 2.0 mode (or x2 1.0 mode) has only 13-22% performance loss, we could average it as 16.5% .

If I was to guestimate a card choice based on result, for x1 mode I'd suggest more RAM being most important with higher performance cards giving diminishing returns. We'll see if anything more than a HD4670-1GB is a waste of money as we get more practical examples of performance.

Running a x2 link changes that altogether so I would invest more in getting a x2 mode link working than a faster card. A HD4670-1GB in x2 mode would outperform a HD4870-1GB in x1 mode. A x2 link speed opens the door to faster card giving faster performance as there is much less of a pci-e link bottleneck than at x1 1.0. A x2 link worthy of further investigation as shown in Wiring up x2 pci-e link mode.

HD4350 performance

ATI HD4350-512MB	8x 2.0 (16x 1.0)	1x 2.0 (2x 1.0)	1x 1.0
Crysis 1680x1050 "low" settings 3dmark06 3dmarkvantage	16.56 2747 649	15.01 (-9.34%) 2568 (-6.52%) 631 (-2.77%)	13.63 (-17.69%) 1911 (-30.43%) 540 (-16.80%)

Source: HD4350 compared: 8x 2.0, 1x 2.0, 1x 1.0.

Further reading

- 8600GT 16x 1.0, 1x 1.0 (inc Phys-X with 9600GT)
- 7300GT 16x 1.0, 1x 1.0
- Tom's Hardware PCI Express Scaling Analysis (2007).

 

My card is a PALIT 512mb GDDR3 SONIC edition, its factory OCed (20%) it can match in performance a HD4870 thats why I chosed this card, I also thought about getting the HD4770 but its hard to get in my country, also I didnt see the card at ebay.

Ill try to summarize my posts too to make it easier for people, also Ill take a look to mod the PE2L to make it work at X2.
When I saw many results comparing X2 and X1 I was SHOCKED it seems that this is the most important bridge in performance...

 

Wow, that performance data kinda hits you like a bucket of ice water to your face...the performance drop is insane going from x2 to x1

I wonder how the later, more powerful models of the Vidock will keep performance loss to their "6 percent" margin.

 

Urmmm I got a GOOD idea, why opening the laptop like this?:
http://hwtools.net/Adapter/PM2C.html

When you could have it closed?
Just mod and solder the pcie pins to an "unused" modem port (if your laptop has one) mod a telephone connector and there you go!, just solder the modded connector to the DIY vidock!...
Hehehehe this remembers me those days modding my EEEpc netbook for fun ...

 

Oh, yeah, ok. I forgot, you can just solder your own connector on.
And one thing i need to verify... The PE4... How many molex connectors will it have to power the whole thing to max power?

lolol, wow. I feel pretty dumb.


And one more thing I must be driving you people crazy.
Do you guys think i would see a difference between a 1GB 650MHz 4850 (GDDR3), and a 1GB 750MHz 4870 (GDDR5)? It is a big speed difference. But do you think the difference would be negligible at x1? The cost difference is only $17... but i dont want to waste money on something i wont notice...

 

Well in my humble opinion, even if you dont notice the improvement, the HD4870 is a better card and it will be easier to sell your card later. To be honest $17 isnt that much but its your money after all .

 

Hmm, so, i guess you dont think there'd be an improvement then?

Yeah, i know. But its just adding to the overall price, that I gotta convince my uncle to buy. $133 does sound cheaper than $150. Im probbly not going to sell the card later for like 3-4 years. So, ill think it over and see what he thinks as well.

EDIT: Wow, I never expected that much of a performance drop between x1 / x2 / x16 !
This sucks. So, does this mean, that say, a 4850 loosing 50-60%, could we buy a card thats literally Half as good as the 4850, and achieve the same results then, or would that also be cut in half and be even less?

 

Oh I see urmmm then you could try the HD4850 the price seems less deadly .
Well if you chose a less powerful card the X1 will also cut the performance (almost in half), but you could also try the 4670, wich is not that bad and you could nicely game with it.

 

Ok, i get it now. I think...

So, now that im thinking, the difference between a 4850 and 70 could be higher. If each is cut in half, it could matter more, right?

lets see, if you look HERE, say, look on the top chart for 1920x1200 on high settings/AA.
With a 4850, it is 47 FPS
A 4870 gets....... 56 FPS
Which isnt a noticable difference here....

Now, just estimating half performance loss...
The 4850 gets 23 FPS
A 4870 gets 28 FPS
Now, thats a little more noticeable.

Just saying... Trying to help out.
To me it sound like on paper 40-70% loss sounds like alot.
But if you get a pretty good card, like the 4850/4870 or higher, the noticable difference is lower. Because past 35-40 FPS, there isnt much difference you can see... This also will vastly differ with what resolution you plan to run at.

Personally, the TV's res i plan to use this for is only 1280x720. So, I think i would see almost no difference between the 4850/70.

So, I think if your running games at a higher res than 1680x1050, it will matter more. If your running at that or lower, it wont really matter too much.
As you can see in the Third graph down HERE. (Cutting them in half), the difference between the two are 27 and 33 FPS. Which is much less of a difference. And this would be less and less as the resolution gets lower. And higher and higher as the reolution gets higher from 1680x1050.

At least this is the way ive figured it out. If im wrong, please tell me. But i think this is something we should put on the front page as a good breakdown of things.

 

I like the initiative and modders may consider this way. I was trying to think of a discrete and effective way of transmitting the x2 signals without having to disassemble the notebook.

The pci-e scaling analysis shows
- Compared to it's x16 2.0 performance, a HD4870 at x2 1.0 loses on average 16.5%. At x1 1.0 it loses over 50%!!
- a HD3850 at x2 1.0 speed outperforms a HD4870 at x1 1.0 speed. Only exception is with Crysis which I believe is due to the HD3850 only having 256MB of vram putting more load on the pci-e bus for data transfer to/from vram.

Chipset support and port allocation for x2 pci-e link speed confirms a ICH8M/ICH9M chipset allows for a x2 pci-link to be created by electrically combining the notebook's pci express port1 and port2. The rest of this post proposes a couple of ways of connecting those x2 pci-e lines.

Wiring up a x2 pci-e link mode

DIY x2 mode by connecting mini pci-e to pci-e adapter's x2 lines with flexible extension lead

mPCIe pin / Video card pin 21 (GND) - A23 (GND) 23 (PERn0) - A22 (PERn1) 25 (PERp0) - A21 (PERp1) 27 (GND) - A20 (GND) 29 (GND) - B21 (GND) 31 (PETn0) - B20 (PETn1) 33 (PETp0) - B19 (PETp1) 35 (GND) - B18 (GND)

Pins for x2 mode from the PE4H and PM3N schematics

Cheapest practical way of trying x2 mode is buy the US$55 PE4L-EC2C + ebay $5 Mini PCI-e PCI Express to SATA SSD adapter + $1 IDE cable. I'd remove BOTH the USB and the SATA header and solder IDE cable as shown by the MAGENTA /GREEN bars. Connect those 8 wires from the IDE cable to the mPCIe blank and the video card's x2 pins as shown above. Only issue is the middle GREEN GND, as there should be two wires to act as a shield to either side. 1 may be OK, but I'd try to tie the wire on the end and squeeze two in that space. It really depends on the pitch of the IDE cable versus that of the SATA header. These wires must run parallel to each other to ensure zero parasitic inductance. The interconnect being:

If it works can the consider buying the PE4H then for a proper cabled setup. I guess one advantage of the PE4H is you can swap the pci-e x1 and pci-e x2 ports around to be either expresscard or mPCIe whereas above hack can only use pci-e x2 port as mPCIe.

hwtools request to integrate a x2 port in PE4H/PM3N

hwtools could make adding x2 link speed easier by providing an open ended x2 pci-e slot with either (i) solderable pads for the additional pci-e lines for x2 mode or (ii) a USB connector already wired to the x2 pci-e slot pins. My analysis below suggesting the USB port has 4 lines , plus a GND. Meaning each end could have the 4 pci-e lines required, plus the 4 GND lines trunked and attached to the USB port (4 GND lines all trunked to 1 GND line on the ends).

I've already asked Gerry from hwtools to consider some convenient way of wiring in the additional signals for x2 mode as shown in point 11 of Specification/Requests for PM3/PE4 product. Response so far suggesting he has a preference to leave existing PE2L/PE4/PM3 products as x1 and await expressport 2.0 to double bandwidth. I will ask again with benchmarks showing a good reason to do so but I can see that a ready made solution would be fiddly so we may be on our own wiring it up. An integrated product may be something like:

USB Port Pinout

1 Red 5 Volt 2 White Data - 3 Green Data + 4 Black Ground

The above being a suggestion, but least optimized implementation using another PM3 product and cable as input into the x2 pci-e slot adapter to give 2x bandwidth. A optimized cabling solution might look to link port1 and port2 pins on the system side and just have a single cable running to the x2 pci-e board.

How to wire up a x4 port? [hardcore modders only - need special equip to do this] You'd be looking at the southbridge chip (front and back)? It would have Intel 82801HUB or 82801xxx printed on it, like shown shown from a Macbook Air.. with the test points around it documented here. Right: pci-e pins of interest marked in color. From ICH8M datasheet

16 pci-e pins that could be tapped off to make a x4 port. How to do that? It would mean *lifting* the BGA southbridge chip, attaching tiny leads to the pins, and then run them in the DIY fashion out to a x4 pcie slot hosting your pci card. A bit of effort, but hey - this is a DIY ViDock thread, so someone might wanna have a go at it.

While having the BGA chip off, there are 3(ICH8M) or 4(ICH9M) SATA ports in there. Just needs 4 wires per port to access them. Can then add netbook style mini pci-e SSD in there, running 4 wires plus 3.3V and GND to it and putting it somewhere in your chassis.

 

Well nando this is my original idea, basically and at early alpha state, I cant see why this couldnt work...
You will not lose any bandwidth if you use low loss cables, in most cases the data loss is due the system design of the connector not the connector itself so a regular low los cable could drive a 2.5gbps signal without issues just make sure it is the right size to allow the amps flow.

To avoid soldering issues we would need a custom mini PCIE card with the correct test points in order to drive them to the custo adapter...
Nando we should make more research to allow X2 mode cuz this seems interesting .

 

Guys, I've read through the majority of the thread here. I am very fascinated with this project. One thing that I wanted to suggest was that in doing all these notebook projects, the closest desktop to a notebook is a 1U rack mount server. So whenever we're looking for small compact parts, 1U servers are an excellent place look for parts.

Consider a 1U power supply.

There is a way that you can modify a power supply to power without a motherboard.

This PE4 sounds exciting if it has a USB port. We can leave our USB hub plugged in with our desktop keyboard + mouse plugged into the hubs. So when we are at home wanting to dock, we just plug in power, network, audio, and video/USB. So technically one would be plugging in one less plug. Which isn't a big deal to me, but it is for naysayers


I really really want to do this to my old sony vaio. Only because I spent hours wishing that it had something better than GMA950

Also, what if we removed our WLAN mini-pcie and also tied into our expresscard interface. Would this ever allow us to enable SLI mode?

 

^^^
Good Idea. But wont all PSU's give power to plugs if you just flip the on switch?
I think...At least I sure hope so.


And another Q: Even if a PSU is old, and it is 200W, regardless of Rails and all that, it still has 200W, right?
Because i have an old 200W PSU i plan to use. But, i want to make sure its going to work. I know that 12V Rails and stuff like that is important, and im sure it doesnt have many . But 200W is 200W, right?

 

the exresscard and miniPCIe might share a 1x port. This needs to be tested.

 

ok, that's good to know

I feel like looking for a datasheet for my toshiba "express port replicator" (aka docking port) to see if it has it's own PCI-e port.
That would be the only way for me to get sli

 

Unfortunately not possible using the "express port replicator" shown here. The ports on it are: Rear: DC In, USB 2.0 (4), Serial, Parallel, VGA2, DVI-D2, LAN (10/100/1000Mbps3), Audio In and Audio Out. No expressports/mini pci-e ports on it.. So the naming may be a bit misleading. I believe they mean it's a quick (express) way of getting to your ports. So you could still get SLI, but if you have only 2 ports then it would be x1 1.0 SLI, which is a waste of time. x2 1.0 HD48xx would outperform it anyday of the week so better to figure a way of getting a x2 port from 2 mini pci-e and/or expressports or a combo of both.

 

no, I meant using the actual port on the bottom of my laptop that you connect the port replicator to.

It must have access to a PCI bus. how else could you connect so many devices through it?

Edit: I found a mantenance manual for a toshiba tecra a7, I have a tecra a9 so this is very simmilar. It has a lot of info and I don't have time to read through it all now. I did find an interesting diagram though. Will post it here tommorow when I have time.

 

nando, I downloaded baredit but I'm not sure what exactly to do with it - could you give some step-by-step instructions on what i should do to check what you requested?

sorry for the little disappearance btw, i've been getting a bit busy lately!

 

So... To really see if the ViDock's claim of only 6% performance loss is true. SInce it uses X1, just like our will, how could it be so much different, with a lower card?

Well, no, it looks like thats BS. At least to me. Look at this...

Look here, at "regular" HD 4670. Playing Crysis, it gets 27.95 FPS Taken from HERE



------------------------------------------------

Right from the NBR ViDock Review HERE. On the NP5797, which probably has the most similar specs to the desktop tested above. Gets 14.57 FPS. On the same settings, and basically the same resolution.




So, thats about 13 FPS Difference, which is 47.87% Loss. Just about like on ours. At X1. The reason ours is higher is because we compared higher cards. So, the difference will be larger than on a slower card, Always.

I figured this could help us out. VillageTronic is full of crap. Where they got 6% from, i have no idea.

...I think im going to go post this on the thread. lol

 

Youre Welcome.

And yeah, i think youre probably right there.
Well, before i read about the DIY, i was considering the viDock. because it is cheaper than a new notebook.
I just couldnt figure how it got 6%, when we figured 40-60%. So, i figured id look into a little more.

So, it definetly looks like we remade the ViDock. Even if we cant make it better, we make it for under Half Price.
...I just wonder how they essentially got the PE4, before we have.


Also, how many Molex connectors is the PE4 going to use. i have a feeling one isnt going to be enough to power ~65W for the card, and i need to know if I need to buy a Molex Splitter to have enough plugs...

 

Architectural Limitations affecting performance for both DIY ViDock and the ViDock

Expressport/mini pci-e port are architectural limited to x1 1.0 speed. Thats going to be your performance wall. ViDock and PE2L/PM3/PE4 are using the same expressport pci-e pins so we will get exactly the same performance using either solution. There is nothing that Villagetronic can do to enhance this performance except:

• petition ATI/Nvidia for reduced pci-e bandwidth. ATI have already done a bit of work there.
• tweak the drivers . Realistically only giving < 10% better performance if can be tweaked.
• wait for expressport 2.0 which will provide a x1 2.0 link by doubling the transmission clock rate

Why the DIY ViDock's flexibility makes it is an overall better solution over ViDock

• external 12V PSU: can add one to match your video card
• unrestricted video card sizes. Not limited by chassis dimensions.
• mini-pci-e AND expressport systems are supported. ViDock only catering to expressport community.
• a NBR/global creation. It came from a global community to serve a global community

With it's BIG benefits being:

• x2 pci-e link capable to give a massive performance injection, doubled in bandwith again with expressport 2.0.
• costs a lot less. There's also enough info here so you could make the wiring interconnect for < $20US if you have more time than money.

I acknowledge that ViDock pioneered and popularized the idea, sorting out some compatibility issues along the way and am thankful for their efforts. NBR DIY now removing the inconveniences from their solution and working towards a whole new performance level with a x2 pci-e link mode.

 

^^^
yep.
Those are mainly the reasons im going to build one.

And, any info on the Molex Connectors Question?

 

This is a long shot, but is it possible to change the two bits from 00 to 01 or 11?
I mean if I change 00 to 01, will port 1 become x2?
If I change 00 to 11 will port 1 become x4?

 

The expressport/mini pci-e port has one set of PERn0, PERp0, PETn0, PETp0, which we can say is a x1 port as shown here. To get a x2 link we need to combine two such ports (full set of signals from one and only the 4 lines from the other). The chipset will automatically negotiate the a x2 link if all connected correctly. So x2 is possible if port 1 and port 2 are a physical expressport/mini pci-e port as described here. Your Toshiba Tecra A9 unfortunately only has a single x1 mini pci-e port so you're limited to x1 1.0 speed.

 

ok, I understand. But what would happen if I did change the bits?
I assume nothing?

Edit:
One more question:

from here:
http://en.wikipedia.org/wiki/PCI_Express

Does that mean that soon there will be another PCIe lane in miniPCIe ports or is the extra lane already there?

 

Interesting, there are 6 reserved lines on the mini pci-e slot as shown here which could be the Future extension for another PCIe lane. Now the vendor could have wired another X1 port to those pins, which could be combined to be a x2 port. Do you have any more technical documentation from Toshiba? Either that or guestimate a port connections. Would be interesting if indeed there was 2 x1 ports on your single mini pci-e connector.

 

Is there any way I could check with a DMM?
Maybe HWtools could check?

I will look for more technical docs.

My first guess would be:
pin45 = PCI Express RX -
pin47 = PCI Express RX +
pin49 = PCI Express TX -
pin51 = PCI Express TX +

I don't think we need the other two reserved pins, do we?