AMD A8-3850 With Radeon HD 6550D

[Qaridarium]

Regarding memory scaling, I have tested Llano A8-3850, 8GB: 2x DDR3-1866 G.Skil RipJaws-X, Asus F1A75M-Pro, seeting the memory to 1333 and 1866 speed respectively, Results for Unigine and results for free games. All at 1280x1024 (my 2nd TFT). I also did encoding tests to show thet ram speed doesn't matter there. And compared to my old Phenom 9350e (Phenom 1st generation, 4x 2.0 GHz). Also reduced Llano to 2.0 Hz for comparison.

I can barely see visited links in this forum so I set them to bold.

I used (K)Ubuntu 11.10 Alpha which runs fine with the catalyst driver, I even got decent screen output with the Radeon driver delivered by Ubuntu, yet had big problems with setting the screen res there. The driver sees a VGA and a DVI screen while only DVI was connected.

Power usage for me is idle: 30W idle with Catalyst, 35W with Radeon. Only CPU fan plus Notebook HDD attached, P/S: Enermax Pro82+ 385W. Full load under Unigine Tropics: upto 115W. Prime95: 110W. When I set CPU voltage to 1.25 (prime stable) I get max load closely under 100W. Btw: accordingt to lanet3dnow.de their Llano had CPU Voltage-ID at 1.275V, the board set close to 1.4V though. I think there's somethin wrong with a BIOS lib, MSI told them this would be fixed with the next BIOS.

My board sets 1.41V by default ...

there are tests out there proofed the 2500mhz ddr3 speed up.

sure you need to overclock it to 5ghz to get the effect.

[edgar_wibeau]

there are tests out there proofed the 2500mhz ddr3 speed up.

sure you need to overclock it to 5ghz to get the effect.

Have you followed my links? I forgot to write down my conclusions for the games, see below, but you could have found them yourself from the benches. For the games, I too have a speedup. Not for the encoders though, just about one percent which is within measurement errors. When you need to overclock to insane clock speeds, there pratically is no benefit for normal users and even on-air overclockers.

Unigine benches normalized against the 1333 run: 6%, 10% and 13% speedup.
Free games normalized against the 1333 run: 6% .. 31% speedup
Encoding normalized against the 1333 (A8) run: 0% .. 2% speedup.

31% speedup from 40% memory speedup, all other staying flat, is a very decent achievement. To me at least. But clearly not in all application types.

[agd5f]

Any chance you can try the Displayport (DP) connector? Non-DP connectors use DP bridge chips (DP-to-VGA/TMDS/LVDS) which are still a little tweaky on some systems.

[Michael]

Any chance you can try the Displayport (DP) connector? Non-DP connectors use DP bridge chips (DP-to-VGA/TMDS/LVDS) which are still a little tweaky on some systems.

Yeah I have a DP display but I think I tested that already with Llano... Will try it again today if I have time while finishing packing, otherwise I'll try it again when I get back with the latest code on the DP monitor.

[Kivada]

there are tests out there proofed the 2500mhz ddr3 speed up.

sure you need to overclock it to 5ghz to get the effect.

So you know nothing of overclocking or hardware in general I see... These chips are designed to use DDR# 1866Mhz as stock memory, you can easily find DDR3 rated t run at 2.4Ghz as it's stock speed at the DDR3 voltage spec of 1.65v, the only thing you need is a mobo with the options in BIOS to change the clock rate and timings past the memory controller's default.

Considering you can do this completely independently of the CPU, GPU, FSB and HT Link speeds and the fact that the ram is within $10-20 USD of DDR3 1600Mhz if you shop around theres no real reason not to max the ram to get as much punch ot of the GPU as possible.

These chips can also be crossfired with the HD6670 or paired with any PCIe GPU you like and used as a dedicated OpenCL coprocessor whenever that becomes a major part of OSS software.

Lets also not forget that AMD has stated there will be mobos for these chips using Coreboot instead of BIOS or UEFI as default. Makes it an easy pick for the would be RMS around here that want something that is as OSS as possible.

[Luke_Wolf]

Lets also not forget that AMD has stated there will be mobos for these chips using Coreboot instead of BIOS or UEFI as default. Makes it an easy pick for the would be RMS around here that want something that is as OSS as possible.

I personally will be so glad to have that. And such boards I will definitely favor, Here's hoping for Gigabyte to be The One or one of many anyway...

[edgar_wibeau]

Guess I missed the virtual "ironic" tags in Qaridarium's post, well how funny. I agree with Kivada, it's just some extra bucks vs. DDR3-1333 (+€20 for 8GB, my 1866s are at 1.5V btw.) and you get extra value for it. It's NOT overclocking by any means.

[DarkFoss]

Don't forget if you populate all 4 dimm slots you can enable simultaneous reads and writes but only up to 1600MHz, 1866 or higher and you'll lose that function.

[edgar_wibeau]

Hi,

I also tried 11.04 (natty) with xorg-edgers which serves the latest git xorg stuff, an I had the same results as Michael shows on page 2 in his review. I run an HP LP2465 via DVI, my board is an Asus F1A75-M Pro. FGLRX 11.7 didn't work well for me as it repeatedly crashes when I watch VDR (DVB-S) via kaffeine/xinelib and switch channels or choose another video source inside VDR (like AVIs from hard disk). After re-login kded4 utilizes one CPU core at 100%, systray gets slooooooooooow. I had the same phenomenon (X crashes when switching channels with VDR/kaffeine) with FGLRX years ago when radeon wasn't stable with my older HD 3300 IGP. Still unsolved it seems.

So I threw out FGLRX and edgers, went back to the original xorg and radeon packages and upgraded to 11.10 (oneiric) alpha 2, current state.

And - surprise surprise - Radeon works!

One major drawback though: the driver and KMS "see" a VGA monitor (on "VGA-0") which isn't there and it's configured to 1024x768. That one is the primary monitor, thus the real monitor on DVI ("DVI-0") is set to mirror the VGA and also runs at XGA res. There is at least one work-around though: in KDE System Settings -> Display and Monitor, which calls xrandr at login time. A little tricky though as X hangs when fooling around too much with the settings. The way it worked: I first set 1920x1200 using krandrtray (a systray tool), started systems settings/display and disabled the ghost VGA monitor, set the whole thing as standard. Kwin now calls xrandr at login time to apply these settings. The result in ~/.kde/share/config/krandrrc is:

[Display]
ApplyOnStartup=true
StartupCommands=xrandr --output VGA-0 --pos 0x0 --mode 1024x768 --refresh 60.0038\nxrandr --output DVI-0 --pos 0x0 --mode 1920x1200 --refresh 59.9502\nxrandr --noprimary

Problem: while booting, when KMS kicks in, my screen switches off. And comes back after KDE-login so I have to use auto-login. Which I do on my home machine anyways. It can be set to lock the screen after login for the scared amongst us

I'm sure this can be done in xorg.conf too, but I want to stay with the standard graphical tools until radeon/kms are fixed. Also kernel parameters might do the trick, I saw older documentation and the switches didn't work for me. I think I tried "video=DVD-0:e video=VGA-0:d". Maybe it's the syntax with two times "video=", didn't try.

So, everything works for me now, idle power was 5 watts higher when I last measured so I guess the GPU power management lacks behind FGLRX's, but 3D works, 3D desktop works (after first reboot with new settings), but is slower than FGLRX of course. But rock stable. Btw, V-drift is faster than FGLRX only because many effects (like reflections) are inactive when running it with radeon.

So, that setup works fine for me now. And if Oneiric breaks temporarily (only, hopefully) on it's way to release, I have a backup and also a running Natty copy with FGLRX, my data sits on a RAID and my mails on IMAP

Oh and DarkFoss: Llano has two memory channels of 64 bits, populating four DIMM slots only hurts power draw (at least), latency, max freq.

[chithanh]

I think I tried "video=DVD-0:e video=VGA-0:d". Maybe it's the syntax with two times "video=", didn't try.

I think you got the output names wrong. You need to use the names from /sys/class/drm/card0, not the names shown by xrandr.