Nvidia working with partners to support external graphics solutions

Greg S

Posts: 1,607   +442

The green chip maker is ready to start offering native support for external graphics cards. AMD introduced XConnect to support Radeon series graphics cards over Thunderbolt 3 connections, now it's Nvidia's turn to offer a usable external graphics solution.

At SIGGRAPH 2017 in Los Angeles, California, Nvidia announced that it has been working with partners to develop an external solution for the professional Quadro series of graphics cards in addition to the Titan Xp. External graphics is nothing new, but offering a certified solution is very important for businesses purchasing Quadro series cards. Consumers should be able to look forward to WHQL GeForce drivers with eGPU support.

Nvidia seems to be pushing hard for creative professionals to turn to their hardware regardless of whether they are buying true professional graphics cards or not since a performance patch has been issued for the Titan Xp. Better performance can now be had when working in Adobe Premiere Pro and Autodesk Maya despite the fact that the Titan Xp is not considered a professional grade graphics card.

Gaming is obviously what many people will be interested in when looking at a Titan Xp for use as an eGPU. It is unknown what kind of real world performance can be expected from a Titan Xp connected via Thunderbolt 3 (we found a GTX 1070 external solution apparently maxing out/bottlenecking), but it is definitely far better than anything an Ultrabook or Surface Pro tablet can provide.

Pricing and availability is not yet known, but Nvidia claims that the Titan Xp is ready for use in external solutions already and that the Quadro series cards should be finished within two to three weeks.

Permalink to story.

 
Last edited by a moderator:

Evernessince

Posts: 5,464   +6,145
Now we just need and external CPU.

Likely impossible, if not ill advised. The graphics card can be moved externally because it is not the primary driver of the computer. The CPU on the other hand is and would add additional latency to everything in the best case scenario. Other questions remain about how you would actually initiate the port that the CPU is connected to without a CPU to do so. Kind of hard, yeah.

If you are talking about having an additional CPU as an external card, that has been done in the past and it didn't work out well. Unless you are programming an application to take advantage of it specifically, adding a high latency processor to windows will only serve to screw things up.
 

jobeard

Posts: 14,119   +1,851
Likely impossible, if not ill advised. The graphics card can be moved externally because it is not the primary driver of the computer. The CPU on the other hand is and would add additional latency to everything in the best case scenario. Other questions remain about how you would actually initiate the port that the CPU is connected to without a CPU to do so. Kind of hard, yeah.
Our current multcore and multicpu solutions are in the SMP world of architecture. The "external CPU" is in the MPP world of multiprocessing. It's been around since the late '70s, has a Wiki and is also available from Big Blue IBM, in their cluster design.
 

Evernessince

Posts: 5,464   +6,145
Our current multcore and multicpu solutions are in the SMP world of architecture. The "external CPU" is in the MPP world of multiprocessing. It's been around since the late '70s, has a Wiki and is also available from Big Blue IBM, in their cluster design.

I'm aware of these. Intel also has CPU expansion cards as well. Like I said in my post though, these are intended for a specific use case.
 

jobeard

Posts: 14,119   +1,851
I'm aware of these. Intel also has CPU expansion cards as well. Like I said in my post though, these are intended for a specific use case.
my objection was this errant summarization
Other questions remain about how you would actually initiate the port that the CPU is connected to without a CPU to do so. Kind of hard, yeah.
The solutions are well known
 

jobeard

Posts: 14,119   +1,851
Hmm; IMO, your too focused upon the MS PC. In the MPP architecture, we share NOTHING (including chipset devices or bios); context and data is managed over a highspeed backplane to allow threaded implementations to co-process without resources being access in the other machine(s).

If you add the restriction that you need shared resources in the ext CPU, then you're right back to the SMP system with an expensive co-processor