The AMD Radeon RX 580 and RX 570 Review | Hacker News Reader

The AMD Radeon RX 580 and RX 570 Review

anandtech.com

55 pointsnedsma31 comments

They effectively KabyLake'd it. They released a slightly better (5% quoted from Anandtech article) for the sake of a seasonal/yearly release cycle. Vega is supposed to be the big fish.

The only upside from this from my point of view (a recent RX 470 owner on Arch) is that Linux drivers are likely very close to "just working". Currently somethings have just made their way in to the kernel (Audio bitstreaming, 4.11 I think, and I think I saw it first hand I was running the 4.11 release candidate a few days ago) and soon FreeSync (FWIW) and HDCP are coming soon.

Upside: No proprietary kernel blobs. I applaud AMD on this front and it completely motivated my purchasing decision of my RX 470 (vs nVidia 1060).

> They effectively KabyLake'd it. They released a slightly better (5% quoted from Anandtech article) for the sake of a seasonal/yearly release cycle.

No, it's the same chip, just different clocks. This is nothing new or out of the ordinary.

https://news.ycombinator.com/item?id=14140412

The new chips are on a 14nm process. They might have a similar architecture but there's room for higher clock speeds according to AMD. Ars Technica said they expect partner cards to have higher performance than the reference numbers AMD gave out.

The 480 was 14nm as well. It's literally the exact same thing, same process. New default clock/voltage, new model number. The process did improve over time, sure, but a late 480 is pretty much the same as a 580.

Oh, I misread. They said it's an improved process, but I guess it's at the same node as the last one.

A process can be drastically better while being on the same node - e.g adding FinFet or other technologies. Meanwhile everyone uses different measures of node size such that they're not comparable across fabs and are kind of marketing numbers these days.

what about power consumption and noise?

Well, for me it looks good: market price is $20 less than RX480 [1] (or 10% lower) for 7% more performance. Near silent operation and 7% more power draw at full load (less power draw at idle and video playback).

It is as if Kabylake was cheaper than Skylake.

[1] https://www.newegg.com/Product/Product.aspx?Item=N82E1681412...

They have been doing for ever. Remember the R9 280 and the 280X? What's the difference between the 280X and the 7970?

I have a gtx 1070 and a gtx 1060 system and things "just work". On linux I have bluray playback, various older games on steam, random webGL games or demos, various desktop effects (standard unity).

Do things like that work with the AMD? Do the AMD drivers still have issues with playback? Or leaving yourself logged into for a week with 100 windows that triggers a mem leak so that all new windows are a black rectangle? Pixel corruption? Performance worse than a 3 generation old nvidia?

I have a drawing error with tooltips occasionally, and 4k video could play more smoothly, but apart from that, AMDGPU is working great.

I have a 1070 and Chrome crashes all the time watching streams/youtube, got flickering on windows when running a 144hz monitor

Related: is there an AMD card which can smoothly run modern Linux desktops (with all of the effects) using a built-in kernel OSS driver? Preferably fanless. I want to try Ryzen but don't want to buy a loud power hungry monster and deal with proprietary nVidia drivers.

Thanks!

If all you need is desktop + browser then likely any AMD card 7XXX series and newer. Keep in mind that some of features for AMDGPU such as Freesync or HDMI audio are depend on out-of-tree kernel module which is open source, but add important not yet merged code.

Or for myself, any way of getting 2560x1440 running with an amd card. The hd 6450 specifically but I would by a new card for full support with a Linux desktop. I've tried a few distros, xrandr configs, drivers and the amd catalyst.

Anything after the 7000 series should work just fine OOB.

I have a Gigabyte RX 470 on Arch Linux. The mainline kernel `amdgpu` driver works fine with only minor issues on Xorg. My current setup is a RX 470 with two LG27UD58s via DisplayPort and 4k@60Hz works fine. I do mostly desktop productivity stuff, but have fired up games briefly(Dota2, Cities: Skylines). There are RX 460 variants that are fanless and can be had for a good price with the RX 500 series release.

Known issues:

* DisplayPort + HDMI audio bitstreaming is in the 4.11 kernel, I believe I saw the ALSA audio devices on a 4.11 release candidate I ran briefly for another issue.

* HDCP isn't implemented yet (can't confirm though)

* FreeSync isn't implemented yet (can't confirm either, not sure of the utility other then gaming and maybe video vsync improvements)

Follow the news @ http://www.phoronix.com/scan.php?page=search&q=AMDGPU The "AMDGPU" driver is the OSS and the AMDGPU-PRO is the proprietary driver.

Side note, I moved from the Radeon 6870 I bought years ago (or Bitcoin mining, I have two blocks in the chain to that card's credit... a different time). I dropped it not for performance but for deprecated driver support because of the the `radeon` driver. Review this table[1] and try to pick a card that falls under the `amdgpu` driver to avoid support issues in years to come.

[1] https://www.x.org/wiki/RadeonFeature/#index5h2

[2] https://www.x.org/wiki/RadeonFeature/#index1h2

Hehe, small world. I think I saw your review of it on NewEgg earlier, thanks! :)

Fanless is hard -- there's an XFX fanless RX 460 but there's no need for it really. Most cards have a "zero fan" concept so unless you are driving the card real hard it simply won't spin the fans. I am using a single slot RX 460 and it practically never runs the fan, you probably will use a dual slot one with beefier heatsink -- even better.

Just got an R3 250 (fanless) with a Ryzen 7 1700, everything works out of the box with Fedora 26 Alpha (Kernel 4.11).

The purpose of this hardware is to render high resolution graphics at high frame rates. Are there any comparative benchmarks available? There is a lot of discussion of the hardware, but few if any metrics of the resulting performance.

Where are 490, 490X? (if not 490X2)

I came across a blender OpenCL vs CUDA benchmark in which 480 was being compared to GeForce GTX 1060 and a claim was made that OpenCL is finally on par with CUDA. So 480 can't be a flagship model of 400 series. (1060 has 4 ahead of it in lineup: 1070, 1080, Titan X, Titan XP).

I think I read that OpenCL on the 480 finally caught up to OpenCL on the 1060. I believe CUDA still works better.

- Either 480 vs 1060 was a garbage comparison (480 is AMD's flagship while 1060 is nVidia's lightweight GPU) and OpenCL is way way behind CUDA in blender render performance (to the point that I would call that news clip a bullshit clickbait news item).

- or AMD 480 is not AMD's flagship 400 series offering, in which case why they haven't released one?

edit: anyway I just read some of the details of 580 and it makes me sick how pathetic AMD is. They went from 200 to 300 series by pure rebranding and practically zero tech advancement. And now they're doing the same thing going from 400 to 500. (Which means a 580 is really a 380 or 390 in terms of improvements over an 8GB 290X).

On the other hand, my 3 year old R9 290X runs a heavily optimized OpenCL code at the 70% of the speed at which my brand new Nvidia 1080 runs the equivalent code. 1080 costs close to 1000 euro, while R9 could probably be found second hand for a hundred, while 480/580 is faster than 290X and costs 300 eur brand new.

Oh boy. That means it quite possible if an 8GB 290X is benchmarked against a 580, the 290X comes out on top.

I'm not a big fan of nVidia either. I think nVidia and AMD used to compete on GPU technologies; now they compete on shenanigans.

Here's the article: https://wiki.blender.org/index.php/Dev:Source/Render/Cycles/...

Regarding your edit, hasn't Nvidia done that multiple times too? Didn't they skip the 800 series (it was mobile only) and weren't the 100 series rebranded 9000 series and the 300 series rebranded 200 series?

I wouldn't be surprised if they have. Their proprietary CUDA gimmicks are much worse. But with AMD supporting the OpenCL open platform, there is some hope of do-no-evil. But they keep disappointing.

If I'm reading correctly then the only thing changed from previous models are memory and core clocks. Why even bother with such "update"? Difference in performance is negligible, but 30W more power consumption costs you real money

> Why even bother with such "update"?

This is a pretty usual practice for GPUs; one generation has a new architecture and new chips in the mainstream/performance segment, the next generation uses the same chips with some minor tweaks. The lower end often stays on the same architecture for longer. All GPUs get the same marketing number be it "GT 5XX" or "RX 4xx" to have a consistent line-up. It's just marketing, really.

OEMs require yearly refreshes. AMD's slides show comparisons to cards 2 generations ago, so this refresh is to appeal to their own former customers who haven't upgraded for a few years. It's not that appealing to anyone who's upgraded recently.

Reply on news.ycombinator.com