ATI Radeon HD 4870 512MB [Archive]

phoronix

07-03-2008, 08:50 AM

Phoronix: ATI Radeon HD 4870 512MB

A week ago we looked at the brand-new ATI Radeon HD 4850 graphics card under Linux. This graphics card launch was unlike any in ATI's history where with the introduction of a brand new product generation, Linux users were greeted by same-day Linux support -- both through their proprietary fglrx driver and with the open-source xf86-video-ati driver. In addition, some of the board partners are opting to put Tux on their product packaging and shipping the Linux drivers on their product CDs. As we had also exclusively shared, AMD will soon be approaching a feature parity between the Windows and Linux drivers. Today we're publishing our complete review of the new ATI Radeon HD 4870 512MB as well as delivering additional benchmarks from the Radeon HD 4850 under Linux, of course.

http://www.phoronix.com/vr.php?view=12558

d2kx

07-03-2008, 10:29 AM

Damn nice card, but your Diamond 4850 looked better ;)

Melcar

07-03-2008, 10:51 AM

I so want one of those cards. Damn lack of money :mad:.

mtippett

07-03-2008, 11:07 AM

Damn nice card, but your Diamond 4850 looked better ;)

So d2kx, did you like the AA numbers (from page 7 and on?)

Regards,

Matthew

Kano

07-03-2008, 11:38 AM

Then the Linux support is so good with fglrx then show a wine screenshot playing Trackmania Nations Forever...

bridgman

07-03-2008, 11:41 AM

Ahh, you mean the Windows support with fglrx :D

d2kx

07-03-2008, 11:50 AM

@Matthew: of course I do ;) I was talking about the design of the card, just in case you've misunderstood me.

@Kano: I've played Trackmania United: Forever when the Catalyst 8.5 was released, haven't tried it with any driver newer than that one (8.6), because I prefer playing the few Windows-native games under that other OS to get the best performance possible, but I'll try that out again for you in a few days if you like.

Spanner

07-03-2008, 02:41 PM

A very nice review, it was good to see the newer cards distancing themselves a bit more.

Why were there no results for X-plane at higher resolutions and AA/AF settings? (This was what I was most interested in).

Also there's no discussion about X-video playback - from what I hear in these forums, it's not even possible to play back X-video in a window and use compiz at the same time. And what about tearing?

My GeForce 6800GT is hours away from death and I want to hear that as a Linux user I can finally live with an ATI card! I'm hugely encouraged (especially by some of the AMD guys on these forums) but not yet quite convinced...

Yfrwlf

07-03-2008, 02:46 PM

I'm still waiting for this Tux on the box and Linux drivers on the CD thing. I still have yet to find any Radeon 4000 series cards which have either of these. So far there still appears to be a total lack of advertising for Linux. That's great that Linux is supported "internally" better, but without actual exposure no one else will know or care except those who are already using it. Actual advertising for Linux is far more important than a few more features added to their Linux drivers IMO, but it all still helps of course and ATI has been great with Linux support. I know it's up to these vendors to mention Linux, but you'd think they would at least say *something* about Linux now, even if they don't put a penguin on the front, at least list it in the system requirements on the side or mention it on the back or something??

Has anyone seen or read anything about Linux at all anywhere on any of these new graphics cards??

I mean, if we are told that Tux is going to be on the box and drivers on the CD, and then it doesn't happen, anywhere, then, that was a pointless falsehood, so why was it said? Their sources were obviously overzealous before they had the facts I guess? I mean, why say it otherwise?

It's like saying "THIS JUST IN, companies could put a huge penguin on the front of the box if they wanted to!" Well, of course they can, but the real question is will they, and who will be the first? When one of them actually does it, that will be real actual news. Here's to hoping it'll happen eventually though.

cruiseoveride

07-03-2008, 04:46 PM

In addition, some of the board partners are opting to put Tux on their product packaging and shipping the Linux drivers on their product CDs

Which "partners"? I want to buy from a Linux supported partner.

Yfrwlf

07-03-2008, 05:33 PM

Which "partners"? I want to buy from a Linux supported partner.

Exactly, where are they, which ones? Any Phoronix editors wish to clarify? So far, I checked out the 4850 from Diamond, and there is NO mention of anything Linux anywhere. Completely typical box, like Linux didn't exist. In the system requirements, several versions of Windows are mentioned as being required. Why would a company not want to list it being Linux compatible if it is to help sell their product? I don't know what kind of perks these companies are getting for only listing Microsoft products, but I'm going to support the ones that are honest and list the operating systems that are actually supported by AMD not to mention open source developers. If there are any that do, that is.

Melcar

07-03-2008, 06:28 PM

A guy from another forum got a HD4870 and he said the disk came with both Windows and Linux drivers, but the box itself said nothing about Linux. I think it was a Visiontek.

mtippett

07-03-2008, 09:01 PM

To clarify slightly, there are always a number of sub-SKUs within a product family. The second and third sub-SKUs will have definately have the driver (at least the CD image that AMD releases), the AIB may do something different.

The driver on the CD is the same as the 8.501 release included in CATALYST 8.6.

Regards,

Matthew

Laughing1

07-03-2008, 11:16 PM

When delivering our Radeon HD 4850 results, we had found the NVIDIA GeForce 9800GTX to outperform the Radeon HD 4850 in a majority of the tests and there wasn't much of a difference between the Radeon HD 3870 and HD 4850. We were not hitting a CPU bottleneck either, as the Core 2 Duo "Wolfdale" was running at 4GHz, but instead we were hitting a bottleneck within the fglrx driver.
http://www.phoronix.com/scan.php?page=article&item=ati_radeonhd_4870&num=3

I would love to see the benchmarks when that bottleneck is fixed in fglrx... ;)

mtippett

07-03-2008, 11:44 PM

http://www.phoronix.com/scan.php?page=article&item=ati_radeonhd_4870&num=3

I would love to see the benchmarks when that bottleneck is fixed in fglrx... ;)

If the HW is made the limiting factor ( http://www.phoronix.com/scan.php?page=article&item=ati_radeonhd_4870&num=7 ) by ramping up image quality (AA, AF) , there is a vast improvement in the relative performance of ATI Radeon products against the competitive (?) NV product.

Even with the high image quality settings, the rates are all very playable 80-120 FPS at 1920x1200.

Regards,

Matthew

bridgman

07-03-2008, 11:49 PM

Laughing1, strictly speaking it probably was a CPU bottleneck, ie driver bottleneck usually = CPU bottleneck. As mtippett said, "something" has to be the bottleneck and if it isn't the GPU hardware then it's going to be the CPU or the bus.

The bottleneck could be a lack of overlap between GPU and CPU function but that's not the case here.

A 4 GHz CPU is mighty fast but it's no match for a modern high end GPU ;)

deanjo

07-04-2008, 12:16 AM

Well good for AMD, it's about time they had something go in their favor. Hopefully this is a sign of things finally turning around for them. We can only hope this new thirst for regaining top dog status carries on to their processors and chipsets (weakest link still being their southbridge).
http://valid.x86-secret.com/cache/banner/384357.png (http://valid.x86-secret.com/show_oc.php?id=384357)

SavageX

07-04-2008, 02:36 AM

bacatta

07-04-2008, 08:00 AM

9/10 for a card that didn't work at all with wine. That's just 75% of my linux gaming... phoronix ? are you here ?

Michael

07-04-2008, 08:03 AM

However, the timedemos coming with Nexuiz predate rendering features like mirrors, reflection/refraction in water etc. etc. and there's not one scene showing theses effects

Well then, how about some updated Nexuiz time demos with the next release :D (Ideally before Phoronix Test Suite 1.2)

mtippett

07-04-2008, 08:11 AM

It's good to see that ATI is giving rather nice performance with Nexuiz in these tests, given that basically all engine development is happening on Nvidia hardware (I do have a Radeon 3850 to serve as driver guinea pig to make sure things at least *work* on ATI, but I'm not involved with engine work).

Hopefully the improvements in the driver are helping change that :).

However, the timedemos coming with Nexuiz predate rendering features like mirrors, reflection/refraction in water etc. etc. and there's not one scene showing theses effects, which is in this case lucky for ATI. My 3850 is taking a rather steep hit on e.g. the maps "starship" and "ruiner" with reflections activated and using the Windows GPU PerfStudio I often see the GPU being rather idle (like 30-40% load) and *downclocking* despite the framerate hitting the floor. So there obviously are some situations in which the driver apparently cannot (yet?) make proper use of the hardware resources available.

Are you saying that there is a common bottleneck between Linux and Windows regarding this behaviour? If so, that makes it a lot easier to locate and resolve a problem when it hits the right point in our priority queue.

Can you work with Michael to create a Nexuiz profile that does trigger this path?

Upside of this is that I can force AA and AF on without a noticable speed hit ;)

The 4800 series has the leading AA performance (as indicated by the results against other 4 digit cards). We will see how it stacks up against the GTX200 series when they arrive for Michael.

Regards,

Matthew

Kano

07-04-2008, 08:11 AM

When do you add some GTX 260/280 benchmarks? As the 260 is only a tiny bit more expensive. The 9800 GTX however is much cheaper.

SavageX

07-04-2008, 08:30 AM

Are you saying that there is a common bottleneck between Linux and Windows regarding this behaviour? If so, that makes it a lot easier to locate and resolve a problem when it hits the right point in our priority queue.

Behaviour for Nexuiz has been wonderfully consistent across platforms with the new Catalyst OpenGL part (you can take that as a compliment). Any performance-issues I saw on Linux were also there on Windows, at least around march/april when I last checked.

Can you work with Michael to create a Nexuiz profile that does trigger this path?

I'll try to record a timedemo demonstrating any problems (most likely a bot match where one can see what a lousy player I actually am).

mtippett

07-04-2008, 09:35 AM

I'll try to record a timedemo demonstrating any problems (most likely a bot match where one can see what a lousy player I actually am).

Can't be worse than me... Why do you think I am so into timedemos :).

Regards,

Matthew

SavageX

07-04-2008, 01:35 PM

Here we go: http://savagex.planetnexuiz.de/demos.zip

Extract into ~/.nexuiz/data and in the Nexuiz console (ESC + SHIFT) a simple "timedemo ruiner" and/or "timedemo starship" will show me battling easy bots (to make me look better).

On High details framerate on my 3850 will drop well within the twenties - anyway, those timedemos are more demanding than the usual demos/demo1 or demos/demo2.

Michael, do you feel like doing a run on the 4870 vs. the Geforce? If the former happens to not beat the latter then I guess we have a rather nice testcase for the ATI driver team - if not then I made much noise about nothing ;-)

bridgman

07-04-2008, 02:20 PM

Michael, do you feel like doing a run on the 4870 vs. the Geforce? If the former happens to not beat the latter then I guess we have a rather nice testcase for the ATI driver team - if not then I made much noise about nothing ;-)

Either way this is good. Thanks !!

SavageX

07-04-2008, 03:06 PM

Either way this is good. Thanks !!

Oh, you're welcome. Nexuiz has a rather satisfying history of collaborating with GPU vendors: When trying to get information about OpenGL support on XGI Volari cards their response was mostly "so, how many cards do you need for testing?" (too bad they went out of business before any cards arrived - anyway, if there are any Ex-XGI people over at AMD/ATI give them a nice hug for having had a really responsive support) and Nvidia even contributed a patch to make Nexuiz work properly with the GeForce 8. If we're doing stupid things in the renderer impacting on ATI performance it would certainly be cool to fix that (and if the driver happens to have a bottleneck I assume AMD/ATI customers like me would appreciate fixes there as well ;) )

Melcar

07-04-2008, 03:30 PM

date 2008-07-04 12:22:01 | enginedate 10:00:06 May 11 2008 | demo ruiner.dem | commandline /home/principe/.nexuiz/nexuiz-linux-x86_64-glx | result 2673 frames 46.2424362 seconds 57.8040480 fps, one-second fps min/avg/max: 20 72 155 (128 seconds)

date 2008-07-04 12:22:39 | enginedate 10:00:06 May 11 2008 | demo starship.dem | commandline /home/principe/.nexuiz/nexuiz-linux-x86_64-glx | result 2478 frames 22.6356981 seconds 109.4730983 fps, one-second fps min/avg/max: 36 131 454 (119 seconds)First one is Ruiner, second one Starship. High details. It's with an overclocked HD3850 though (800/1000).

Nexuiz is one of the few games that runs rather well with ATI cards (no major graphical glitches either); it's even playable on a 9800pro. Others, like Alien Arena, hell, even my older 7900GS runs it better.

bridgman

07-04-2008, 05:18 PM

Am I correct that the timedemo sets quality preset but not resolution ? Anyways, at 1680x1050 high on a stock 3870 with stock Phenom (forget the model) I get :

ruiner : 25 85 178

starship : 38 146 465

Running Hardy and whichever fglrx driver ships with Hardy - 8.3 I think.

For some reason the playback seems a lot jumpier at 1024x768 than at 1680x1050. Forgot to write down the values (since jumpy playback should have a lower min frame rate) but will try to get some more info over the weekend.

Also noticed that the numbers seemed to go up a bit each time I ran the timedemo (again need to write down the numbers this time to confirm); I think the lowest fps on ruiner was closer to 20 first time -- in fact all of my initial numbers were lower than Melcar's IIRC.

That is a VERY nicely done game btw.

Kano

07-04-2008, 05:40 PM

Funny, you forgot about

grep -e "cpu MHz" -e "model name" /proc/cpuinfo

Melcar

07-04-2008, 06:07 PM

My above runs are done at 1680x1050 with a 3GHz X2, just in case. And yes, it's an extremely well done game. Thumbs up for the devs :).

bridgman

07-04-2008, 06:25 PM

Funny, you forgot about grep -e "cpu MHz" -e "model name" /proc/cpuinfo

It's true. I did... although being Friday night and not caring much 'cause I wanted to go home might have been a contributing factor...

OK, it's a 9850 :D

highlandsun

07-04-2008, 07:41 PM

Would love to hear more about what the UVD2 support will provide. What app-level library changes will be needed for e.g. mplayer, ffmpeg, vlc to take advantage of it? Or, alternatively, when will we see XvMC support in any of these radeon drivers?

Michael

07-04-2008, 07:51 PM

bridgman

07-04-2008, 08:51 PM

highlandsun: as Michael said UVD and UVD2 support is still up in the air right now, particularly for open source. I want to get the 6xx 3d engine support taken care of first since that enables 2d, 3d and Xv acceleration on 6xx and, if the accel code is designed right, on 7xx as well without many changes.

There is enough info out today to start an XvMC implementation on 5xx and earlier parts, since we use the 3d engine for MC even in Avivo and all that info is available today. Next step would be adding IDCT to the XvMC implementation; that needs info we have not released yet, on the IDCT block which is present in all Radeon parts up to and including R6xx.

UVD is in 6xx alongside the IDCT/MC capabilities. Main benefit is that it offloads a bit more of the decoding work and uses less power to do the same decoding work. We don't need it for XvMC, just if we want to go the next step beyond XvMC.

Starting in 7xx, UVD and the legacy IDCT block are replaced by UVD2, so that's the first place where you need UVD for full XvMC, unless the IDCT work is done in shaders.

There is a bit of a race, in the sense that there is a GSOC project going on (mentored by one of the Nouveau devs, IIRC) to write a decoding stack to run on shaders via Gallium, which may provide useful results before we are able to find a solution for safely releasing UVD info. If that happens, my guess is that everyone will jump on the shader-based implementation and forget about UVD for a while, giving me time to figure out what to do with it ;)

SavageX

07-05-2008, 04:45 AM

Thanks for the nice comments. Nexuiz for sure isn't as nicely crafted like e.g. the latest Unreal Tournament, but I think that games like Warsow, Alien Arena, Tremolous, Cube 2, Digital Paintball 2 and Nexuiz (I most likely forgot some other games developed with basically a zero $ budget) are each delivering unique values for people who look for gameplay you won't find in streamlined commercial titles.

Timedemos neither do set a quality preset nor do they set resolution. They're mostly a recording of the packets received from the server.

highlandsun

07-05-2008, 04:58 AM

Currently there is no UVD2 support.

Right, understood. But from this quote

The Unified Video Decoder 2 may be supported at some point in the future under Linux, but for now you will be limited to using X-Video for playback.
http://www.phoronix.com/scan.php?page=article&item=ati_radeonhd_4870&num=3

I was just asking what impact the addition of UVD2 support would have.

It seems to me that there's a much larger population of users who watch movies on their PCs than there are users who care about bleeding edge gaming performance. As such, video decoding acceleration seems like the most obvious biggest bang for the buck in terms of accelerating adoption of Radeon on Linux.

SavageX

07-05-2008, 05:53 AM

@mtippett, @bridgman: I (re-)discovered a rather nasty testcase where the Catalyst VBO handling apparently is problematic:

http://savagex.planetnexuiz.de/reborn-timedemo.zip (again, extract into ~/.nexuiz/data)

Nexuiz has several VBO modes:

"gl_vbo: make use of GL_ARB_vertex_buffer_object extension to store static geometry in video memory for faster rendering, 0 disables VBO allocation or use, 1 enables VBOs for vertex and triangle data, 2 only for vertex data, 3 for vertex data and triangle data of simple meshes (ones with only one surface)"

In the Nexuiz console (ESC + SHIFT):

gl_vbo 1
timedemo reborn

This gives me an average framerate well below 20. gl_vbo 1 used to be the default VBO mode for Nexuiz and usually is giving best performance, but we changed it to 3 after we discovered that we can get eight times the performance on my 3850 with this map using less aggressive VBO usage.

mtippett

07-05-2008, 09:44 AM

...

This gives me an average framerate well below 20. gl_vbo 1 used to be the default VBO mode for Nexuiz and usually is giving best performance, but we changed it to 3 after we discovered that we can get eight times the performance on my 3850 with this map using less aggressive VBO usage.

We'll possibly discuss this further offline. Please PM your email details.

FWIW, I have hacked glc (http://nullkey.ath.cx/projects/glc/) to dump framerate every second. Using a T43p with radeon driver (what I happend to have), I have generated this graph using the ruiner benchmark.

http://img73.imageshack.us/img73/781/nexuizruinerhh3.png

I have sent the patches to the glc maintainer, hopefully he will roll those changes in. This paired with GPU utilization will probably give you the smoking guns that you need :).

Regards,

Matthew

Rich43

07-07-2008, 03:46 PM

I have read this entire thread and I cannot find info on the release of NDA free technical documents for this card. When will this be released?
Perhaps bridgman can help here.
Does the creation of cards now ATI/AMD are merged make it easier to get the legal side of documents done?

bridgman

07-07-2008, 05:14 PM

We're trying to get 6xx (HD2xxx/HD3xxx) 3d documentation out now -- don't have an exact date but we're getting close.

The driver differences from 6xx to 7xx (HD4xxx) seem to be relatively small so I expect we will just release a "here are the differences" document, in the same way that the 5xx acceleration doc covered most of the 3xx/4xx/5xx parts fairly well.

Josko

07-07-2008, 05:51 PM

9/10 for card that is sometimes outperformed by one year old 8800GT (wich is also cheaper) and with non working Wine ... that just can't be real :/ !!!
I think that many people plays games in Wine, because now it is not problem to play most of DX9 or OpenGL games for windows for example Half-Life2 EP2, Dark Messiah of Might and Magic (All Valve Source Engine based Games), World of Warcraft, Oblivion, CIV4 and even Crysis and many others.. and when some graphics card is not supporting this it is useless for me in Linux and I've got a feeling I'm not the only one...

btw. I'm using 8800GT and playing all games...

I hope that ATI devs will soon make their drivers fully working with wine.

Rich43

07-08-2008, 10:13 AM

COOL! Thanks for the info bridgman!

mrt181

07-09-2008, 09:07 AM

Yfrwlf

07-09-2008, 01:09 PM

i can't find any reference in the article that this gpu does not support WINE only that NV is regarded as having better support for it than ATI.

Dear phoronix, please make a wine benchmark report and let us know if this card is also as good for dx games as it is for ogl ones.

That would be a good test to add on, since it does certainly effect Linux gaming. Until there are more native Linux clients, that is. *hope* Hurry up Steam Games / UT3! :P

The Linux graphics card market will really take off once there are more great games out there taking advantage of it to drive it...not that desktop gaming is very hot now days, but who knows how that will change, I think there's room for growth if they just do it right.

Dragonlord

07-22-2008, 10:55 AM

storma

07-22-2008, 04:21 PM

Works fine here, none of those symptons.

Debian kernel: 2.6.25-2-686
ati driver: 8.51.3 (8.7)
xorg-server: 2:1.4.2-1
arch: i386

Yfrwlf

07-23-2008, 12:23 PM

Somebody else has utter troubles with this card? It's randomnly hanging the system ( move cursors over window => freeze, shut down X => freeze, start up X => freeze ).

kernel: 2.6.25
ati-driver: 501
xorg-server: 1.3.0.0-r6
arch: amd64 ( athlon64 )

Those with working cards, what systems/soft-versions do you have? I'd like to find out what is up since I would hate loosing such a formidable system to Windows due to driver problems :(

Sounds like you have a bad video card if it seems to be definitely graphics related. Could also try a live CD to rule out it being a software issue with the particular software/driver you're using, or try installing a different driver.

Dragonlord

07-24-2008, 04:27 PM

I tried it with Gentoo/Ubuntu 32/64 bit and it does the same everywhere.

Yfrwlf

07-24-2008, 11:24 PM

I tried it with Gentoo/Ubuntu 32/64 bit and it does the same everywhere.

I'd re-seat and blow off the card and slot, and if it keeps up try a new card and RMA that one if you can. ^^

Dragonlord

07-25-2008, 07:51 AM

rbmorse

07-25-2008, 09:20 AM

Could the problem be heat related? When started from a cold state (after it's been off for a couple of hours) does it work for a little while before the problems manifest?

remm

07-25-2008, 10:02 AM

There is a bit of a race, in the sense that there is a GSOC project going on (mentored by one of the Nouveau devs, IIRC) to write a decoding stack to run on shaders via Gallium, which may provide useful results before we are able to find a solution for safely releasing UVD info. If that happens, my guess is that everyone will jump on the shader-based implementation and forget about UVD for a while, giving me time to figure out what to do with it ;)

In practice, what would be the benefit of UVD(2) over the generic solution (which is supposed to work with all cards - even no card at all actually, the person is using the software impl of Gallium to test at the moment) ?

In other news for Fedora 9 users, the latest build of radeonhd (only available in koji right now: http://koji.fedoraproject.org/koji/buildinfo?buildID=57366) works fine on my 4850 for basic 2D.

Yfrwlf

07-27-2008, 04:40 AM

I had to re-seat it already. It's a fight about millimeters to get the card in ( the thermaltake case I obtained has a... well... little... design flaw ( or two to be precise ) so it has been a tight fit ). Also if I disable acceleration altogether ( NoAccel=true, NoDRI=true ) then it seems to work ( just slow that is ).

That's not good, bending the card could cause cracks in the solder joints, I'd RMA or return it for a replacement if you can. Some places may even allow you to take your computer there and pop a new card in to test it. If it works, then you can go ahead and buy it from them since you would know what the problem is.

The the mobo and case not being lined up right though is still not good and you may want to consider doing something to fix it or getting a new case. I bought one for $30 once and it was actually a pretty good case lol.

Dragonlord

07-28-2008, 08:03 AM

@rbmorse: I don't think it's a heat problem. The temperature is all fine and inside acceptable boundaries. Also the problem happens directly after booting so it's not like the card heated up and then failed.

@Yfrwlf: Checking out plug situation. I would not like to change the case if possible since it's by itself a good case and getting all those components out again would be annoying :D

bridgman

07-28-2008, 09:13 AM

Dragonlord, I hate even suggesting surgery when we're still working on the diagnosis :D

Is there any chance the HDD carrier could be temporarily removed with a few screws to give clearance just for testing ? It's obviously not a long term solution but it would help us to figure out the problem.

Dragonlord

07-28-2008, 05:53 PM

Unfortunately it doesn't change a thing. Any BIOS options known to cause troubles with this card? I left everything on default ( auto ) to avoid troubles due to overzealous options.

Yfrwlf

07-30-2008, 01:15 AM

@rbmorse: I don't think it's a heat problem. The temperature is all fine and inside acceptable boundaries. Also the problem happens directly after booting so it's not like the card heated up and then failed.

Then it's 99% likely that you need a new video card. Buy one, and if it doesn't fix the problem, take it back.

It's possible that heat or the wonky mobo position has helped to cause this problem though, so try to make the new video card have a more pleasant experience than the last one did if you can.