DIY eGPU experiences

[PanzerHauptmann]

The core i7s may be an improvement over c2d, but you're giving them way too much credit for better fps and such. A 2.8ghz C2D for example is better than a 2.4ghz (2.9 turbo) i5 at straight up FP calculations, but a 2.5ghz (3ghz turbo) edges out the 2.8ghz C2D.

Clock for clock, yes the i5/i7 will beat a C2D, but if we're talking about severely bottlenecked GPUs stuck in an expresscard slot, major differences in fps are likely not as much due to the CPU as they are to the efficiency of the expresscard slot.

Yes I realize what you are saying in regards to the expresscard bottleneck, but regardless of this, if you have a CPU that processes data faster & more efficient this will result in obviously better benchmark results, as the data being taken in thru the expresscard port is processed faster. DIY vidock experience as a whole will improve if every aspect of the PC is better, whether it be RAM, CPU clock, etc.

Anyways, if that's not the explaination, then what is? (for the impressive Agentyuri's results?)

[Eggs Scrambled]

Yes I realize what you are saying in regards to the expresscard bottleneck, but regardless of this, if you have a CPU that processes data faster & more efficient this will result in obviously better benchmark results, as the data being taken in thru the expresscard port is processed faster. DIY vidock experience as a whole will improve if every aspect of the PC is better, whether it be RAM, CPU clock, etc.

Anyways, if that's not the explaination, then what is? (for the impressive Agentyuri's results?)

You're misunderstanding what I'm saying. If you look at nando's chart, it looks as if just using a newer CPU is what improves the score so drastically. The real improvement is just using a better CPU, regardless of generation. The other cpu that hits 10k under optimus is like 2.2ghz and Agent Yura's cpu is the tippity toppity top of the line latest i7 dual core clocked at 2.66ghz and turbo goes as high as 3.33ghz O.o . Any upgrade that drastic will result in a huge performance differential.

My point was if someone gets in here and uses a p8700 (2.8ghz) overclocked to like 3-3.2ghz (relatively easy assuming unlocked PLL or such), the difference wont be as drastic. So it's not about an architecture change, it's instead about a raw power change. I just don't want people coming in here, thinking they want a laptop that supports a vidock, and thinking they have to get a latest generation notebook just for the i5/i7 architecture as opposed to something last-gen but with a good C2D inside.

[ithildin]

Yes I realize what you are saying in regards to the expresscard bottleneck, but regardless of this, if you have a CPU that processes data faster & more efficient this will result in obviously better benchmark results, as the data being taken in thru the expresscard port is processed faster. DIY vidock experience as a whole will improve if every aspect of the PC is better, whether it be RAM, CPU clock, etc.

That's why we favour benchmarks focusing on GPU rather than general system performance. If you look again at the Nando's performance data, his "creaky old" Core 2 Duo T6600@2.2GHz is keeping up with AgentYura's i7 in both 3dmark Vantage GPU marks and DMC4 scene4 fps. My "ancient" P9300@2.5GHz can keep up with JohnnyEM's Asus M60J with i7-720QM (a quad core!).

Of course a machine equipped with a i7 will feel snappier and generally perform a lot better but in the context of a DIY Vidock the CPU is not the major performance determinant - expresscard/mPCIe bandwidth, GPU architecture and driver optimization seem to be.

Anyways, if that's not the explaination, then what is? (for the impressive Agentyuri's results?)

Optimus.

[pterodactilo]

Hello, I want to use vidock with a Precision M4500 notebook. Is it possible to display the signal on the internal LCD ?

[jamesbond007]

@pterodactilo It is possible but with frame rate caps. There is a full explanation in the first page of the post under mini faq question number 3.

[antichamp]

Hi.
I've been trying to read through all of this, but there's just hundreds of pages through these threads and I was wondering if someone would help me.

I know from opening up my laptop (acer 5920g) that there is an unused pcie, and it also has an expresscard slot.
When I check it only shows 1 unused pcie:






Does anyone know how to figure out which port is which?

[ithildin]

It's hard to tell how the PCIe lanes are wired for a particular machine just by using software. The easiest way of finding out is by moving mPCIe cards around and connecting a device to the expresscard slot. Everest should be able to tell you in which port the device is now connected and you can then figure out how they're all connected.

For instance, in your case try moving your Intel Wifi card to the unused mPCIe slot, boot up Windows and check which port it shows up in Everest. As for the expresscard, can you get your hands on any expresscard peripheral, like a wifi adapter or USB/firewire card? If you plug any of those to the expresscard slot, Everest should be able to recognize the device and tell you which port that slot is wired to. Have a got at it and let us know what you find.

[PanzerHauptmann]

Is there any way (with adqueate knowledge of soldering and tech knowledge of course) would it be able to detach laptop onboard GPU chip and re-solder some sort of makeshift device that would route data from that old GPU "port" or socket and send it to the PE4H w/ external GFX card attached?

Or, perhaps in other words, what is the absolute best method of achieve highest bandwidth xfer rates with DIY vidock currently?
Assuming x2 w/ optimus drivers has been proven best interface as of yet, right?

[ithildin]

Is there any way (with adqueate knowledge of soldering and tech knowledge of course) would it be able to detach laptop onboard GPU chip and re-solder some sort of makeshift device that would route data from that old GPU "port" or socket and send it to the PE4H w/ external GFX card attached?

A few people here in NBR thought of hooking up a desktop GPU to the laptop GPU's MXM slot. The GPU would have to be supported by a more complex device than the PE4H but then it could make use of full x16 2.0 bandwidth and even use the laptop's LVDS link to display the output on the LCD panel. Unfortunately, this idea hit a roadblock - the detailed MXM specs required to design such a device are not open to the general public.

[PanzerHauptmann]

A few people here in NBR thought of hooking up a desktop GPU to the laptop GPU's MXM slot. The GPU would have to be supported by a more complex device than the PE4H but then it could make use of full x16 2.0 bandwidth and even use the laptop's LVDS link to display the output on the LCD panel. Unfortunately, this idea hit a roadblock - the detailed MXM specs required to design such a device are not open to the general public.

I see, thanks for that info....

On another note, I've been doing some stress level testing with FurMark using my PE4H and GTX470, due to some "blackouts" during gameplay. I'm using stock clock speeds so it couldn't be due to OCing, which leads me to think that it may have something to do with heat levels.

After several stability tests with FurMark, with settings at 1920x1200 Fullscreen, 0xAA, each test got to about the 2:00 mark and "blackscreened" at 64C each time. I had fanspeed at 100% during the duration of the test and no background programs running. (Except firewall). So it seems as thou, for whatever reason, once 64C is hit it automatically shuts down. Im using Corsair VX450 as PSU so I doubt it's a power supply issue. In addition, what facts further support this theory is that of the other several tests that I ran with card overclocked (715/1720) performed without error, and temps never passed 61C during each test.

Any ideas what can be done to workaround this issue? Perhaps it's the PE4H shutting down due to overloading? I just don't know, this card should be able to reach well into the 80C+ zones and perform fine.

Doing a benchmark test, the GTX performed quite well, considering other results with comparable cards scored the same or lower with much more powerful rigs, overall. The results of a bench run at 1920x1200 fullscreen can be seen here

EDIT: Just performed another quick benchmark test with FurMark with settings at 1920 x 1200, with 8xAA. 30000ms test (pretty quick), avgeraged about 50 FPS. Temps never exceeded 60C. Then I did another test, changing AA from 8 to 32xAA. Temps never exceeded 60C, and half way thru the test it "blackscreened" again, freezing up like it does when temps pass 64C. So now Im just stumped...

Only thing I can think of would be that I have my PSU plugged into a surge protector with about 6 other power-hungry devices plugged into it. Does this have a wattage reduction effect, due to an overload??? I have no idea, maybe that's a stupid theory but to me it makes sense, in a way. I will try plugging Corsair PSU directly into it's exclusive wall 110V outlet and replicate test to see if results are the same.