Topic-Free Mega Thread - v 1.11.2020

General
2803

A combination of several problems and shortages that are slowly being remedied.
They hired a number of new employees for both the GPU and CPU side but also lost a lot of former talent from the GCN and earlier days far as I’ve heard.

There’s a lot of open issues and unlisted problems as well and regressions on earlier GPU models and several issues took far too long to be acknowledged even with Vanguard and some only got discovered after this new telemetry inclusion and crash reporting utility.
(Curious though as the months before that had AMD’s Reddit and their forums with plenty of issues mentioned that they claimed they weren’t aware about.)

Linux driver situation has improved and the Windows drivers are bettering but the code stack and stuff is from what I can tell quite messy and the 5000 series only now has the hardware to leverage some of the features that NVIDIA could do if I understood that correctly.

NVIDIA is also very non-standard using their own tech and methods so with OpenGL for example it works for them but not the others and it’s fast but very non-compliant with standards whereas AMD’s will work but as a result it’s also slower.

Working with Microsoft and Sony for yet another console generation and further improving RDNA plus the new employees and current staff and engineering talent working out the hardware problems of RDNA1 and resolving the software bugs should at least (Theoretically.) make for a much smoother launch now for RDNA2 and the 6000 series.
(There is a ton of early driver changes and workarounds so AMD was quite aware about some of the issues they claimed to not know about from workarounds in memory controller issues to timing tweaks and other stuff glanced from the Linux code submissions done.)

EDIT: As a average user I only have some insights and what I read and can find and much of it’s probably only internal from AMD themselves so the full extent of problems and issues between the Radeon Technology Group and AMD and the problems between the GCN and shift to RDNA and much more we’ll probably never know the full extent of.

2804

About as expected for the 6000 series GPU’s but possibly somewhat of a confirmation that the situation will be a fairly limited initial supply.

https://www.reddit.com/r/Amd/comments/jp773u/asus_rep_indicates_rx_6000_launch_will_be_just_as/

1 Like
2805

Yep… Once again, until Radeon gets its shit together, I am going to stick with GeForce.

Fun fact, I never had as many issues with NVidia drivers in Linux as I did with Radeon. This casts a very strong doubt at Linus Torvalds fiery middle finger at NVidia for me.

2806

That’s more about NVIDIA not having a open source driver on the same level as AMD I think although AMD has some closed core components too I believe and NVIDIA has opened up more compared to earlier far as I know.

Quality wise it seems they’ve had a number of issues but also worked quickly on resolving them either through beta drivers, hotfixes or branches like the Vulkan one for priority VLK issues and resolving several of these ahead of it being merged into the main driver code for Windows and Linux every few months. :slight_smile:

Another issue being the number of different distributions and kernels and updates for these and support for anything on a older core it’s tough even for the larger corporations to support this effectively from my understanding.

EDIT: As it stands though until a few months ago I had a tough time recommending AMD as a GPU choice because the software situation and outstanding unresolved issues it’s mostly the later Summer and Autumn now almost Winter months that have seen a much better turnaround.

Some important fixes from 20.2.x and 20.4.x but also a bit back and forth on fixes, open issues and long term problems potentially from unverified and hard to confirm problems on a hardware level for the Navi10 GPU’s which AMD has attempted to work around, resolve in later models or just bypassed or hopped over same as Vega and Fury.

EDIT: NGG stack on Vega seems very problematic (Mostly broken on a hardware level.) and was touted as a big feature among other promises though Navi10 / RDNA1 has since resolved many of those issues among other key improvements.

If RDNA2 and the upcoming RDNA3 improvements with the Navi20 and Navi30 cards can do what Terascale 1 → 2 and GCN 1.0/Gen1 1.1->Gen2 did that’d be really impressive and quite a comeback after some of these issues.

But that still leaves the software side and there’s still problems here too.
Getting better though but it’s taken a lot of time so hopefully this situation won’t be a recurring issue.

EDIT: Polaris made several key improvements over Fury and then Vega over Polaris and then Navi and RDNA1 just had a lot of features and revamped functions and key improvements.

Bodes well but if it’s going to be similar to before then the update as with TeraScale or GCN iterations are going to be a lot better.

5000 series of GPU’s already push well above their price class and hardware would suggest though they can’t match NVIDIA’s earlier product launch and high-end lineup and then the 2000 series after that although catching up to within 5 - 10% of the 2070S is a strong effort.
(Even winning against it in some cases.)

Consumer wise AMD and NVIDIA going at it more evenly and without one having a major lead in performance and market availability could also be really beneficial though I have doubts as to the prices coming down though somewhat would be nice and normalizing on a slightly lower level. :slight_smile:

2807

The history there has been too lengthy. AMD needs to solidly prove to me that they are actually learning from their GPU cycles and not hiding behind “features” or “we do this other thing better”. In fact, if it wasn’t for Ryzen, I would say the same for the CPU line.

Good grief was the Bulldozer line a complete joke!

1 Like
2808

Yeah they’re looking at future possibilities and scaling and the market situation but Bulldozer had a lot of problems and as seen from the GPU lineup AMD’s doing a similar thing pushing towards D3D12 or Vulkan even kicking it up on PC with Mantle to get things rolling but there’s been a bit of a lull for a few years of D3D11 still dominating and backwards compatibility and code practices against Windows 7 holding back some of this and then their hardware isn’t as effective anymore.

CPU wise Zen+ and Zen2 really made a difference with Zen3 further improving and pushing against Intel but it’s not going to be the same against NVIDIA.

Back then I think AMD had several miscalculations and bad deals GPU wise too the early move away from TeraScale to GCN hindered them a bit and GCN for gaming never really became as efficient as AMD would have hoped even as they tried to rectify several of the architectural problems and GPU hardware limitations.
(Best case the final resulting GCN GPU the Radeon VII achieved a 67% effectiveness in Wolfenstein New Colossus which yeah that’s not very great and overall results in general were worse.)

EDIT: Zen would have been in problems too if Intel hadn’t made several bad moves in a row as it is they have some time until 2022 almost before Intel is going to be showing what they’ve been doing on the desktop market.

Still big in the server segment and elsewhere plus their actual revenue and funding level compared to AMD, saying Intel is being defeated is a bit premature.

From NVIDIA the upcoming GPU is I think called Hopper and that could be something against AMD’s RDNA momentum too.

Well Ampere on TSMC 7nm+ with the 3080Ti could also be a surprise if that’s what’s coming up next against the 6800XT or 6900XT.
(Likely followed by additional models if that works out well for NVIDIA.)

2809

Software and application wise, server and workstation segment and such NVIDIA also has CUDA.
That’s big and very established while AMD had OpenCL support but little else.
(Giving NVIDIA years to developer and build up support here which paid off really well for them.)

NVIDIA isn’t in much of a threat here from AMD yet even if RDNA2 and newer compete a bit on the desktop and gaming segment, CDNA for the compute cards don’t have much info at all about them yet either.

2810

What even uses OpenCL in semi-professional cases?

2811

Nvidia’s problem with Linux isn’t so much that the driver isn’t open source (that is a problem for some) or even it’s quality/stability (although they tend to lag behind kernels so you have to delay updating your kernel if you are on say a rolling release), but more Nvidia shear stubbornness in not working with the community. Wayland support has been a recent major sticking point. Some Wayland implementations have managed to work with the proprietary Nvidia driver, but they had to go out of their way to do that because Nvidia refuses to implement the APIs that every other driver manufacturer did, because they “know better” or whatever their excuse is. So now there is a whole branch of code that is only for Nvidia, which is hell for maintaining compatibility. Some smaller projects have already downright said they won’t support that, because they just don’t have the manpower to maintain a separate implementation just to bend to Nvidia’s whims. The Linux driver has been fine otherwise, has been for years. Heck Nvidia was supporting Linux way before AMD or Intel decided they could be bothered (at least as it came to GPUs).

1 Like
2812

When was this? I was using ATi hardware on Linux in 2002.

2814

Back then AMD’s Official Driver was a buggy mess, and you were lucky if it would work, and even luckier if you could do anything more than desktop usage (not that there was much “gaming” on Linux back then, at least as you think of it these days). I think there were community written drivers, and I believe they were more stable, but they weren’t supported by AMD directly. These days there are two drivers the proprietary one and the open source community written drivers. Even AMD says to use the Open Source drivers over their proprietary ones for basically any non-workstation card. AMD directly contributes to that driver quite heavily and provides the necessary Binary BLOBs for getting full use out of modern hardware.

Nvidia does contribute to Nouveau drivers some, but not much, and they also refuse to provide the Binary BLOBs, so you are locked out of 3D clocks on anything newer than I think the 700 series (which is when that requirement was introduced to the firmware). That basically cripples the open source driver. No one in their right mind is gonna use that on a decent Video card, and at the low end where such things don’t really matter, most people will just get an AMD or stick to integrated graphics, which both Intel and AMD have viable Open Source drivers.

2815

@Kaldaien I realized that the download link on the top haven’t been updated with v0.11.1 – and there’s no 32-bit build with the gamma fix applied either, I believe.

Can you do a recompile of everything and release it in in a separate thread? It’s confusing for everyone trying to track down the latest versions of each DLL file :sweat_smile:

2 Likes
2816

Plan was to release 0.11.1 this weekend anyway. I’ll sleep on it and probably release something tomorrow.

I need to simplify the HDR visualization menu, I went a bit crazy with stuff the end-user doesn’t need because it interested me :slight_smile:

2817

What’s the easiest way to explain raytracing to someone who doesn’t even understand how PS5’s SSD can be faster than the Series X, if it’s smaller in capacity?

I’m not kidding, i have these sort of friends :man_shrugging:t2:

2818

1000×1000 132 KB

EDIT: Probably not though.

EDIT: Although not entirely incorrect though ray gets shot out you get response that gets used it’s just really, really simplified.

2819

Bounce, bounce, pass.

1 Like
2820

:laughing: yeah what makes it weird is explaining what the current approach is (rasterization) and how it’s different. I can say raytracing is basically true(-er?) to life, but then i’d have to explain how we’re not already there.

Worth noting, this same person asked me to find him a surround sound setup, then asked me why there’s so many speakers…

2821

I’m using this image from the link and hoping (why am i hoping) he understands it from there
image

2822

I don’t even understand that and I’ve written Ray Tracing software :stuck_out_tongue: Why are those rays terminating in mid-air?

The idea of Ray Tracing is that surface properties come along for the ride with rays of light as they bounce from one surface to another. I don’t know what that image is, but it’s showing something different … ray marching, maybe?

2823

I don’t know what it is either, i just see bounce lighting, lol. And i know my friend well enough to know a diagram won’t work on him.