Hacker News Reader

Comment by zanny | Hacker News Reader

Hopefully the kernel driver rewrite pays off, and not just in time saved supporting separate Catalyst and free kernel modules. AMD on Linux is infamously slow, and while their Windows driver situation has never been particularly good, they get crushed by Nvidia.

Thats also bad, because they are also the most freedom hostile major hardware manufacturer on the market. Some other contenders like Broadcom, Intel, and Qualcomm all participate some engineer time somewhere in the open stack, albeit not across all their products, but literally nothing Nvidia makes is open at all, and they even go the extra mile nowadays to use signed drivers to stop Nouveau development.

Anyone who cares about eventually seeing a world where GPU programming is as open as CPU is (was) also cares to see AMD succeed on Linux, being the only major discrete GPU developer out there besides Nvidia, and whose legacy IP has gone into the largest mobile GPUs on Android as well (Adreno). They might have proprietary power management firmware on their cards, but at least they publish the ISA. Can you imagine a CPU without an open instruction set?

View on HN

I'm not sure what applications you're talking about when you say they get crushed by Nvidia.

AMD has faster integer operations so they're much better for hashing operations like bitcoin mining and password cracking.

Nvidia is better at floating point ops which are used in computational biology and neural nets. Also they use CUDA which is the only interface supported by many libs.

AMD has better open source drivers than Nvidia. By far. Their default open source drivers are actually usable on Ubuntu, while with Nvidia you have to install ridiculously unstable Nvidia proprietary drivers to just watch flash videos.

On Windows for games, AMD cards are competitive for almost every given price point. At the top, r9-390x beats GTX 980 by a little.

>AMD has better open source drivers than Nvidia.

Yes, but Nvidia has working proprietary drivers that put anything AMD to shame on Linux. The difference is so big its outright ludicrous, and it gets worse the higher end the card. I have no idea where the "ridiculously unstable" comes from, must be Ubuntu bugs. Or were you talking about AMD's proprietary driver, fglrx? I agree that's garbage.

>At the top

The top is the GTX980 Ti, for which AMD has no answer. Also if you say "competitive at price point" this means totally disregarding power consumption, of course, where again AMD is so far behind its ludicrous.

If one let it count there still is the R9 295X2, which beats the GTX 980 Ti.

Edit: Source http://anandtech.com/bench/product/1437?vs=1496, because this got downvoted (??)

295x2 is a dual GPU card that requires a nuclear reactor to power. It will beat the 980ti in games that will gain 50% when using 2-way crossfire. Considering the usually disadvantages of using crossfire/SLI it's really not worth it.

Yeah I know, the card uses a lot of energy, and the potential crossfire issues are not nice. Still, it is an answer to the 980 Ti (or rather, the 980 Ti was an answer to the 295X2).

980ti wasn't an answer to 295x2, more like the nail in the coffin, it came out couple of weeks before the Fury at a lower price and at about 1000 times more availability and just pulled the rug under AMD's feet.

NVIDIA was smart enough not to get into the fight with dual GPU cards it's a loose loose situation.

> NVIDIA was smart enough not to get into the fight with dual GPU cards it's a loose loose situation

Uhhh... well, look, I don't know much about video cards these days, but that's not quite true: http://www.t3.com/news/nvidia-is-about-to-launch-a-new-monst...

It is quite very true, that dual GPU card if it ever comes out will be a replacement for the TitanZ a niche card not a flag ship. NVIDIA leaned not to repeat that misstake from the 690 which was and be all accounts would be the last time they release a flagship consumer GTX line dual GPU card. When you have a flagship card that can lose in benchmarks to cards 1/4th it's price because SLI/Xfire fails, and when you can't enable all the bells and whistles like Adaptive VSync (GSync/ASync), Temporal AA, DSR, and any other new "tech" that you boast about for 6 months before release because of SLI/Xfire you are going to fail and fail badly. AMD had to release the X2 cards because they could not compete against the GTX780/780ti and the 980 was around the corner, by the time the 295x2 actually was available in the market at any decent amounts the 980 was already out, and while the 295x2 beat it in certain cases it was too little too late, especially considering that the 295x2 couldn't fit in many cases (same issue that the 690 had) needed a 350$ power supply to run which most people don't have (same as the 690), and produced enough hit to start fusion (not that far from the 690). So no NVIDIA isn't fighting in the dual GPU arena anymore, they aren't releasing flagships, and they are smart enough to wait for the very end of their refresh cycle to release a "look how much money I got to spend on the latest Titan" card that will never be produced in any large amounts, will not be needed to beat the competition as far as single card performance goes.

Everywhere I can find is calling the new dual-GPU card Nvidia's "flagship", though. Call it a "niche card" as much as you like, but you're moving the goalposts here.

Their previous Dual GPU card was priced at 3000$ this isn't a consumer card, that's double the launch price of the 295x2 which was priced at 1400$ and received a 40% price cut within 3 months (considering that it was pretty much unavailable on the market until July/Aug where it already has been cut to below Titan Black prices you can say it's actual launch price was ~950$). The normal GTX Titan cards which launch at 1000$ can barely be considered a consumer flagship cards, the Titan X was the "1st" one which was some what financially successful because like with the Titan Black they've dropped the "workstation/professional" features from the Core and released it as a premium Ti placeholder to appease early adopters until the yields are good enough that they do not have to cherry pick cores. As far as GPU market goes there are 2 "flagship" products that the market actually cares about the 350-400$ enthusiast card, and the 600-700$ "flagship", both of them however sell several orders of magnitude less than the entry level 200-250$ gaming cards like the 750/760/850/860 or the 270/370 and 280/380. If you still don't understand the difference then the 295x2 was released and positioned to be a competitive option against 700-900$ cards, just like the old NVIDIA 690 which was a 1000$ Dual GPU card. You need to understand that the 690 was supposed to be the "top" of the line card for Nvidia at the time, and technically in some aspects it's still the most powerful single board card they've released to date (it can even beat the TitanZ in some benchmarks). This was the card that Nvidia released as the definite gaming card, this was the card that Nvidia launched the "Geforce Exprience" on top of it, and this was the card that had more issues on launch since their FX series. The card actually lost in gaming benchmarks to Nvidia's own 680 SLI in pretty much every case, it lost to the 680 which was clocked higher in games that didn't support SLI well, it could not support all of the new Geforce graphical features due to SLI compatibility and worse it lost to the 7970CF in most benchmarks. That was pretty much the point where Nvidia it self said they will focus on single core performance and ensure that the primary numerical GTX series will be the one that will focus on real-world gaming performance, 2 years later the Titan series was released which is their "look at what we can do, now give us 3000$ please". It's not about moving goal posts, its about perception and actual market presence.

You are right, the 980 Ti was probably more of an answer to the expected potential of the Fury, it had a great timing. But lower price? I can get a R9 Fury X for 635€, a R9 Fury for 564€. The GTX 980 Ti starts at 718€.

Maybe the 980 Ti was cheaper in some markets at the very beginning.

The TI was cheaper than the FuryX before the price cut, and the 980 was cheaper than the normal Fury again before the price cut.

On several laptops already, I got a blank screen at boot, after some updates. Looking for fixes usually gave me "purge nvidia-*". The way i see it these days is that you just have to be lucky.

Solution that works for me is to add the ppa from this page:

https://launchpad.net/~graphics-drivers/+archive/ubuntu/ppa

Everything else requires having to turn off lightdm service or having to ssh to the machine to reinstall the drivers when something goes wrong.

NVidia proprietary drivers are slightly less terrible than AMD's but basically unusable on laptops. Desktop is a far smaller niche. There is nobody that is getting these bugs fixed in distros in a timely fashion, because NVidia doesn't care about laptops and distros don't care about proprietary drivers. See eg this 2+ year old unfixed bug in Ubuntu: https://bugs.launchpad.net/xorg-server/+bug/1220426 (hundreds of comments and subscriptions)

AMD open source drivers are on the other hand are pretty stable and work quite well for things like WebGL, light gaming etc.

Good point about laptops, I didn't consider the Optimus mess. That said, when we're talking about gaming cards I'm not sure that desktop is a niche compared to laptops.

I have a laptop with optimus and it works like a charm even in steam using primusrun. For non gaming stuff it is using the intel gpu.

All I had to do was install 2-3 packages in arch linux and that was that. I haven't had any crashes in linux due to nvidia in many years either so I'd say that the proprietary drivers they have are stellar. I'm a big open source advocator, however I will use what works and for me nvidia is light years ahead of amd when it comes to linux, sad as it is.

Funny, Nvidia's binary drivers have given me the only consistently good experience I've had on Linux laptops. AMD's binary drivers have been absolutely atrocious and the open source drivers (until recent years?) didn't have enough features and enough power savings to sway me.

Compared to AMD's drivers, Nvidia's are good.

Compared to Intel's, both are so horrible it doesn't really matter.

Can't say I expected to see Nvidia fanboys on HN but the R9 Fury X exists and beats 980ti at >1080p resolutions in games not gimped by Nvidia Gameworks: http://www.anandtech.com/show/9390/the-amd-radeon-r9-fury-x-... At the same price point using only slightly more power and having nice and quite closed loop water cooling.

> the R9 Fury X exists and beats 980ti at >1080p resolutions in games not gimped by Nvidia Gameworks

I skim-read for comments regarding the R9 Fury X vs. 980 Ti and in particular comments about >1080p and had to go through seven benchmarks before arriving at an admittedly mixed ray of sunshine for AMD. Quotes from the article:

> (Battlefield 4) The R9 Fury X does not take the lead at any resolution, and in fact this is one of the worse games for the card. At 4K AMD trails by 8-10%, and at 1440p that’s 16%

> (Crysis 3) Once again the card trails the GTX 980 Ti, but not by quite as much as we saw in Battlefield 4. In this case the gap is 6-7% at 4K, and 12% at 1440p

> (Shadow of Mordor) These framerates aren’t bad – the R9 Fury X is always over 30fps – but even accounting for the higher variability of minimum framerates, they’re trailing the GTX 980 Ti by 13-15% with Ultra quality settings

> (Civilization: Beyond Earth) ...can’t keep this advantage at 1440p, where they go back to trailing the GTX 980 Ti in minimum framerates by 7%.

> (Dragon Age: Inquisition) The card trails the GTX 980 Ti at all times, by anywhere between 13% and 18%. At this point AMD is straddling the line between the GTX 980 and GTX 980 Ti, and at 1440p they fall closer to the GTX 980.

> (The Talos Principle) ...AMD’s performance once again regresses at 1440p. It’s enough to retain the lead, if just barely, and tie the GTX Titan X

> (Far Cry 4) Overall what we find is that the R9 Fury X has a 9% lead at 4K Ultra, and a 4% lead at 1440p Ultra, making this one of the only games where AMD takes the lead at 1440p. However something interesting happens if we run at 4K with lower quality settings, and that lead evaporates very quickly, shifting to an NVIDIA lead by roughly the same amount.

The article concludes:

> At 2560x1440 the card loses its status as a viable alternative. AMD’s performance deficit is over 10% at this point...

I didn't look up whether those games use Gameworks or not but if six-point-five out of seven games play better on NVIDIA than AMD well ... who is going to look at AMD cards for gaming?

He did not give you a good source for his claims. There are other benchmarks and the driver situation changed after that. http://wccftech.com/amd-r9-fury-x-performance-ahead-nvidia-9... writes about the driver, http://www.computerbase.de/2015-06/amd-radeon-r9-fury-x-test... saw the Fury X in a different game selection a lot closer to the 980 Ti already after the launch. Still, the 980 Ti was ahead.

Regardless, the more interesting card price/performance wise was the Fury without the X (with the current prices that is debatable). To answer your question: AMD is seen as having the potential to beat Nvidia cards later on, with driver improvements and DirectX 12. Also, there is something to be said to favoring the smaller player in the market. Especially now, with the prices being adjusted to the performance difference.

It is not like choosing the slower card will lead to games being unplayable – never on normal resolutions, with reduced settings otherwise.

Can't say I expected to see Nvidia fanboys

There are plenty of benchmarks to support my claim, no need to resort to name calling, just show evidence that you are right. You'll see that the majority of objective evidence disagrees with you.

Who's the fanboy then?

AMD has Nvidia beat at one point: the R390 beats the GTX970, on Windows. Luckily for AMD (and hence for us, yay competition) it's a reasonably popular point.

> Can't say I expected to see Nvidia fanboys on HN

This breaks the HN guideline against calling names in arguments. Please don't do that.

https://news.ycombinator.com/newsguidelines.html

AMD opensource drivers 3D performance is more than bad. Catalyst driver performance is acceptable for me but I can't use them. It shows artifacts in gnome-shell after some time, resume from RAM almost never works etc. It feels like 10 year old linux user experience.

Do you use really the newest driver? Witcher 2 for example worked a lot better with it than with Catalyst when I tried it out, I think that was half a year ago.

My computer is not able to run Witcher 2. It is notebook APU. Some games works on radeon driver but you can't be sure before you try it. Checking minimum spec for game does not help you.

Witcher 2 is difficult with the minimum specs anyway, because the official support does not include AMD at all :/

I tested it in the beginning with a HD 7850, and later/currently with a 7950. Am not surprised that it does not work with a notebook APU, it runs with my cards but not with high-fps. With a slower gpu it is bound to fail.

> At the top, r9-390x beats GTX 980 by a little.

No, but one step down it does. AMDs R9 390 beats the GTX 970 a little, which also has the problem of only having 3.5GB Vram, instead of 4 as initially announced. The R9 390 has 8.

The R9 390X is normally slower than the GTX 980, but also quite a bit cheaper (~100€).

(Source: I have a hardware-recommender at www.pc-kombo.de which aggregates many gaming benchmarks to judge processor and graphic cards performance)

You might be right. I just went to one comparison site:

http://www.gamespot.com/articles/amd-radeon-r9-390x-review/1...

The article catches one core point well: the resolution plays into it. When we go to higher resolutions than 1920x1080, the R9 gets a boost. But it has a very strange way of presenting the data. http://www.computerbase.de/2015-06/amd-radeon-r9-390x-390-38... shows it easier (1080p Performance).

The one metric which nVidia is beating AMD handily in is gaming performance/power consumption.

Open source drivers are meaningless for gaming, the compiler and driver specific optimizations are not open sourced for AMD neither in the old nor the new driver, and do not seem to ever be. While it's true that AMD has "better" open source drivers I would not go as to say by far, because they are usable and that's about it, you can't use them for gaming no more than you can use the OSS Geforce ones, and on the closed source part NVIDIA driver quality trumps AMD, I wish we were back in the old ATI days with my Hercules Radeon 9800 pro but we aren't. AMD at this point constantly lags behind NVIDIA on everything from performance, to feature support, memory utilization and power consumption.

On the GPGPU front NVIDIA is only "better" at FP when you are talking about HP, FullP is just horrid on everything but it's Tesla and 1st Gen Titan cards. And since you don't need to use FP as much because OpenCL and DX-Compute push towards using advanced INT32 operations (mu, mua, sad) it doesn't fair well, NVIDIA dominated when all of the real world applications used FP in their implementations even when you only had to deal with ints (e.g. SHA256) now when you can do INT32 operations it's dragging it's ass behind GCN for most things that people here would care about (Bitcoin, bruteforcing passwords) even if you are buying Tesla's.

> While it's true that AMD has "better" open source drivers I would not go as to say by far, because they are usable and that's about it, you can't use them for gaming no more than you can use the OSS Geforce ones

That is wrong. Since a few versions (since Gallium3D arrived, if I got that right) at least you absolutely can use them to game. I played Witcher 2 on them last week. Also many easier games work fine and stable (Kebal, Mark of the Ninja, Mount & Blade… not hard stuff, but still needing working drivers with 3D performance).

Being capable of running a game and being viable for gaming are 2 different things, i can run Witcher 2 on a MBP13 with Intel Iris Pro on medium settings at an acceptable frame rate. Neither OSS drivers are going to be viable for a gaming machine e.g. SteamOS Box when It's actually intended to be a gaming machine rather than a machine that occasionally can run games older than X amount of years.

You need to check the Phoronix Test Suite numbers, for a number of games on all but the latest AMD hardware the OSS drivers are actually faster than the Catalyst ones. One area where the OSS stack is functionally less than Catalyst is OGL 4.3+. I too use the OSS Radeon drivers and find that they are within a few percentage points of Catalyst numbers and occasionally double digit faster.

OSS drivers for the latest AMD hardware (<1 year old) is hit and miss.

EDIT: clarification

I can check what version I had in SteamOS but 20-30% performance difference in modern games between OSS and closed source AMD drivers as far as 2-3 weeks ago goes on modern AAA titles with Linux support.

I could care less about <1 year old games or synthetic benchmarks tbh.

The versions of the drivers in SteamOS are well behind up to date stable, let alone testing. You can easily pull in one of the many PPAs that track the latest stable and/or testing kernel/MESA/DRM/DRI stacks (such as xorg-edgers, obiaf or others).

If you are not happy with what is provided by stock SteamOS then your issue is with Valve's distribution not the current state of the drivers.

How much of that difference is because of the driver and how much because of a bad port from the game developer?

Remember all the console to PC ports with really bad performance. Now add the lack of a financial incentive for Linux games.

To put this in perspective, there are more Windows XP users than Linux users in Steam stats.

I have a HD 7950 and I can game very well a lot of games.

it's dragging it's ass behind GCN for most things that people here would care about (Bitcoin, bruteforcing passwords) even if you are buying Tesla's.

For Neural Nets NVIDIA is the only game in town.

The non-open source drivers (and CUDA libraries) are a huge deal though - as anyone who has tried to get <insert DeepLearning of your choice here> working.

The (old, "radeon") open source AMD drivers are good enough for a bunch of Steam games, including popular ones like CS:GO or Dota2. Unfortunately they don't scale with higher end cards having the same performance as old ones.

My personal experience with open source drivers versus Catalyst is that Catalyst is generally buggier (Xorg takes 20% memory at times (memory leak)) while the open source drivers work perfectly and have similar FPS (haven't done benchmarks, but it can't be more than 10-20%).

> you can't use them for gaming no more than you can use the OSS Geforce ones

Completely wrong

Open source drivers are critical for the long term maturation of gaming.

Part of the reason that graphics APIs stagnated for a decade was because vendors were using the API in place of actual technical documentation for their hardware.

If we had a vibrant compiler community for GPUs the same way we do for CPUs we would see the development of new and innovative languages to utilize the hardware way better than the C style imperative or C++ style objects we got stuck with in GLSL / HLSL.

APIs like Vulkan are still frontloading the workload of implementing a software stack to compile for specific hardware targets on the hardware vendors. Meanwhile, CPU compilers have been predominantly community projects that target architectures and we have gotten tech like LLVM out of it.

If we ever want to see the ultimate maturation of massively parallel computing devices, they need to be liberated. That doesn't actually mean free software drivers, but it does mean ISA documentation and programming manuals like the computers they are. Nvidia is antithetical to that future with their current extreme hostility to documenting their hardware.

> Open source drivers are meaningless for gaming

Not in principle. See Intel drivers for example. They will not beat a desktop AMD or nVidia card, but I can play Mark of the Ninja and many other games here in my Ubuntu laptop without issues.

I can't play Portal 2 in the laptop with good framerates, but demanding games are not the only games in town.

Therefore, open source drivers matter.

> Their default open source drivers are actually usable on Ubuntu, while with Nvidia you have to install ridiculously unstable Nvidia proprietary drivers to just watch flash videos.

Not just usable, the open source drivers have similar FPS (± 10%) to Catalyst and unlike Catalyst, it doesn't have many bugs and memory leaks.

AMD on Linux is anywhere from 20 - 150% slower than on Windows on either driver depending on workload. Regardless of your title choice, Nvidia will at least have parity with AMD's performance and will usually crush it. It is hard to persuade gamers to adopt a video card with significantly lower performance / $ when they don't care about software ethics.

And its inappropriate to say AMD have "better" open source drivers - Nvidia has nothing to do with Nouveau besides three commits to make their mobile APUs boot right on it so people can immediately install their blob driver, and its a disservice to the community and all the developers that have put huge effort reverse engineering one of the worst technical black boxes in history to give Nvidia any credit for their work. If its any companies driver, its Red Hats, but a super majority of contributions to it are from unpaid or freelance developers.

> Hopefully the kernel driver rewrite pays off

It's not a rewrite, amdgpu is a fork of radeon with support added for newer GPUs plus some cleanups and older code stripped off. Which means if you're submitting fixes, you need to submit it to both drivers (at the moment the two patches you'd submit will usually be identical.)

> AMD on Linux is infamously slow

This has actually changed during the past year, and the open-source drivers are now actually quite good.

I'm coming from the perspective of a 100% radeonsi user. In games where Nvidia gets performance parity between Windows and Linux (mostly TF2) radeonsi will get 50 - 60% of the FPS of Catalyst on Windows on the same card.

Broadcom and Qualcomm? These two are terrible.

Nvidia has actually hired a developer to work on Nouveau: https://www.youtube.com/watch?v=dnG0X0uwLuY

Somewhat related, AMD's slides are claiming over a 100% performance improvement on a wide range of games w/ their latest Crimson drivers (I think these are the latest version of their Catalyst drivers, so not sure where that puts it w/ this OSS release): http://techfrag.com/2015/11/24/radeon-crimson-driver-expecte...

Reply on news.ycombinator.com