That isn’t my experience with Nvidia on Linux. Depending on your distribution it can be annoying to get the proprietary driver installed in the first place, but once it’s running performance is decent and there are no features that are apparently there but are broken like this. This card has been released, GN bought it and tested it. I think it’s pretty disrespectful of the Chinese market and their customers to sell them broken shite and not consider them as real customers as if they were American.
No, it isn’t. Being unable to overclock, Speed Sync not working and the software crashing is not a performance issue. In game performance is fine. These are two completely different properties and it’s very unhelpful to obfuscate the two together.
This is how we end up with shit like “AMD’s driver is shit because they take ages for their Dx11 performance to catch up” “Yeah well Nvidia’s driver is shit because Control Panel looks like a Windows 98 application and GeForce Experience crashes”. They are two completely different arguments.
But that’s a good thing. Once the nvidia driver decides that the proper profile for desktop window manager is power saving and throttles down to 100% GPU usage with nothing open and GUIs are so slow you can see different elements being drawn in between frames, you will be glad the control panel is like a win98 application instead of being a not-webpage
So we’re getting some competition for Radeon Pro and Quadros. This looks interesting indeed. And you can be sure the drivers will be up for the task, because business customers don’t mess with shoddy hardware nor do software companies validate their business products for it.
They start with the lower priced segment. There is no way Intel competing against A6000 and stuff. These are all low-powered cards with 50W and 75W.
I really like the A40 with 50W and single-slot. Probably full encoding package + 4x DP/HDMI. No extra PCIe power. I hope they don’t lock the blower on full-auto for those 50W.
I have not read this post except for your original post, so if this is already been said great. If it hasn’t I’m just going to touch on it
So I’d like to clarify something that Intel has not canceled their arc. It’s that their AIB makers have issued concerns over the quality of the boards and this is going to be pretty normal for somebody getting into this market. It’s very difficult to gain all of that knowledge when it comes to PCB design and interposer design.
I think Intel’s going to continue to focus on their workstation and Enterprise customers and that you should stop the sensationalism about smacking intel over the side, especially when you don’t really understand silicon in the difficulties that come along with it.
So for now if you ask me I think that they should focus on their Enterprise customers and they should focus on getting good at that because that’s where AMD and Nvidia make the most money anyways. Once they figured that out maybe they can take another try at a consumer GPU. I’m not disappointed by this. I partially expected this but it’s not exactly the omg it was canceled. See I told you so sensationalist headline you just made
In a market where there are two competitors with decades of experience compared to you, it’s a pretty hard thing to do and given that raja has come out and said it’s not canceled as well as him exposing the difficulties of having only a small team at Intel working on this, I’m not surprised to see it at all.
I’m still holding up to see whether or not ARC makes a really great Plex server.
If nothing else, all signs point to it being an encode/decode monster. Hopefully some NEEDED competition comes to the VDI space, where there nVidia has been unchallenged for years.
Given that it has all of the iGPU systems and hooks into the rest of the system similarly to an igpu, I’d say it has the capability. Theres a reason they over cooled the A380.
one thought that comes to mind for resizeable BAR being such a big thing would be a texture streaming/caching dma performance hack.
being able to grab a big chunk of system ram probably isn’t gonna be super beneficial for performance if your gpu is running out of ram and you’re essentially just page swapping to system ram.
however, reserving a section of system ram that should theoretically be able to have a dma channel from storage through the chipset could be handy for texture streaming or just caching things that are frequently used (but can be reliably predicted when it will be used, ahead of time) but not enough to justify giving up the higher speed local vram.
im hopeful we can have a player 3 in the gpu space.
kind of curious how linux drivers are gonna shake out, they did wonders with AMD stuff a few years back.
So, interesting news. I picked up one of the Gunnir A380 cards.
My primary workstation is an Epyc 7302P with a 2080TI
My wife’s computer is a Ryzen 3700X which I recently installed the Gunnir card in.
I’m using HandBrake to re-encode some 4K Dolby Vision HDR H265 content to 1080p H264 SDR. With the NVENC encoder I am getting an average framerate of 18 FPS for the source file. CPU utilization is rather low at only 30% and the GPU hovers up and down around 50%.
On my wife’s computer I ran the same file but with the QuickSync encoder. That computer is getting 35 FPS! CPU utilization is about 90% and GPU is about 35%.
Am I crazy?!!? This literally doesn’t make sense.
Okay, so I was crazy. I had HandBrake 1.5.1 on my computer and had the nightly on hers. I’m now encoding at just shy of 60 FPS on my workstation, with much higher CPU utilization. In any case, that 75-watt little card $140 card is about half as fast as a $1000 card, albeit one generation older. That’s a pretty good value! Also kudo’s to whatever change was made in HandBrake to make it a whole heck of alot faster lol
Some additional details. Both of these encode jobs were done with H264 constant quality 17, encoder profile main, and encoder level 4.1. The Nvidia encoder was set to “Slow”, while the Intel one was left at the default “Balanced”
I did several test files, and it seems that the files encoded with the Intel QSV encoder are less than half the size of the ones encoded by the NVENC one. Top half is Intel bottom half is Nvidia
In terms of picture quality, if anything, I noticed fewer compression artifacts from the Intel one. I’ll have to give this a whirl with some royalty free content and share some screens.
