Precision 7530 & Precision 7730 owner's thread

I was referring to HEDT—high-end desktop. Check out the Core i9-9940X, for instance.

Definitely not wrong. The DMI link is fixed, regardless of implementation, and Dell absolutely cannot mess with it.

Let's clear up some misconceptions: the first 16 PCIe lanes are directly from the CPU, to any peripheral. This is typically entirely occupied by one single GPU. If you want SLI, you are limited to 2 × 8 lanes. And this is fixed for all consumer CPUs. HEDT CPUs like the 9940X or whatever are basically highly-binned Xeon parts, with correspondingly more PCIe lanes.

Now, if you look at that graphic I posted, you'll see that the CM246 chipset itself is connected to the CPU by a DMI 3.0 link. Regardless of CM246, Z390, Z370, Z270, or whatever—they're all connected by the DMI link. This link is not really PCIe, but it has the bandwidth of 4 PCIe 3.0 lanes. See link below:

https://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/3

Yes, the chipset exposes 24 lanes, but that's like 24 traffic lanes emerging from a tollbooth, into which go only 4.

As for the implementation on MacBooks, they don't have very powerful GPUs in the first place, so what I think is done is that Apple can afford to allocate only 8 lanes to the GPU (this is typical in thinner notebooks), and use the other 8 to jack up the ThunderBolt controller(s).

At any rate, 4× PCIe 3.0 lanes = 985 MB/s × 4 × 8 = 31.52 Gbps, which is only 9 Gbps short of the 40 achievable in TB3.

What are you doing that needs to saturate that much bandwidth, though? I'm curious.

 

Furthermore, @Aaron44126, I am using driver 419.67, and haven't experienced BSODs—I am using the notebook in discrete GPU mode, exclusively. I haven't experienced any BSODs at all. Perhaps you'd consider getting a board replacement?

Dell's QC on these machines really is terrible.

 

Thing is, it is totally stable on 418.81 (three releases old at this point) and prior, only these 419.xx releases are giving me trouble, so I don't think that it's a hardware issue.
I don't think that I can really complain to Dell unless one of their NVIDIA drivers has the problem...

 

Fair. Have you managed to pinpoint the cause of said BSODs? Maybe some sort of stack trace, or memory dump?

 

I do have the memory dump but I honestly haven't spent much time digging into it at this point. (Too much other stuff to do...)

 

Yes, I have the same issue with my 7730 and TB18DC. It's been that way since new. Even a very light touch on the connector seems to disconnect the dock.

 

Are those nvidia drivers you guys are talking about?...
I can only find up to 411.63 published on the dell website and up to 416.78 on the nvidia website. Where are you getting those more recent drivers from?

 

What are you searching for on the NVIDIA site? Try this form: https://www.nvidia.com/Download/Find.aspx?lang=en-us
If I put in the mobile Quadro P1000 then there are several releases showing up newer than 416.78.

 

Pfff...

If you google for 'nvidia driver', then this is the first link to pop up: https://www.nvidia.com/Download/index.aspx?lang=en-us
Which is apparently not the same driver searching tool... for some reason. Instead, your link leads to some 'advanced driver search' which I can't find an immediate way to get to through the nvidia website. The menu item "drivers -> all drivers" leads to the page I had.

The tool I had only showed one driver as a result at the time. I always chose for 'quadro new feature driver' rather than 'optimal for enterprise'. When I first tried that back in november, this lead to a more recent driver. Now, it's the reverse...

Your 'advanced' driver search tool has an option to show all drivers.

I feel like I've just been fooled by some silly magicians trick; I want to blame nvidia for having a sloppy website, but I also feel like I'm a neanderthal bashing buttons on the keyboard.

Thank you, Aaron, for sharing this life changing link with me.

 

The 'Quadro New Feature' driver has always been a misnomer. The 'ODE' driver (for some reason when I see that acronym I can only think of Ordinary Differential Equations) has generally been the newer one, and I've noticed this well from last year.