Beware of QAFF CPU on X79 platform

This article is probably not very relevant anymore, considering the age of the platform and rarity of the CPU, but nevertheless, it can be useful for historical purposes or if someone happens to have the CPU mentioned below. Back in 2012 I bought X79-based Sandy Bridge-EP platform (based on ASRock X79 Extreme4-M motherboard) together with the engineering sample CPU having a QAFF S-Spec (supposedly 4-way Xeon E5 which doesn't have corresponding production item). Though the CPU was recognized by the motherboard and system initialized successfully, over time however, it proved to be quite unstable and stressful to use. Two main issues were:

• Initial system boot and even reboot process could randomly freeze with a different number of BIOS beeps or motherboard boot status codes (Extreme-4M has a specific LED indicator for them). Probably the most common final code before freeze was 0x67 - CPU DXE initialization (CPU module specific). I believe it could have been RAM detection/initialization issues. Unfortunately, I couldn't find any pattern to reproduce the failure consistently.
• PCI-E graphics cards with v3.0 specification (possibly v2.0 as well) lead to the failure of the OS boot process into graphical environment (Windows 7, Gnome on Manjaro Linux, etc). Reducing PCI-E mode to version 1 in BIOS settings was partially solving the issue by stabilizing the boot process. In this case however, the system had an occasional performance degradation and/or random lockouts. I was using Radeon RX460 graphics card, but I found a MSI forum thread confirming the similar issue with different graphics cards (including AMD and Nvidia models).

The first issue is hard to investigate because of its inconsistent behaviour, however, the system usually worked better with two RAM modules instead of four, in my experience. Shuffling RAM modules between slots occasionally was solving the issue, but temporarily only.

The second problem was easier to identify since the problems started soon after I replaced older Radeon 2400PRO based graphics card with more modern Radeon RX460 one. Windows boot process locked out immediately on first reboot after AMD drivers were installed. Linux boot process froze as well on loading display server (X.Org). In both cases, the only way to continue was to reset the system. By default, BIOS was selecting PCI-E v2 for the graphics card. Reducing PCI-E to version 1 helped to load the desktop environment in Linux but system performance was unstable up to complete system lock out. Windows was still crashing regardless of the PCI-E mode set in BIOS. Explicitly setting PCI-E v3.0 in BIOS was rendering system unbootable until CMOS was cleared. It seems v3 was either not supported by CPU at all or it didn't work in combination with my motherboard.

X79-based systems are already considered legacy since long ago and it is not very likely that someone will decide to build it these days, especially with a quite rare engineering sample CPU. In this case though, my recommendation is to avoid QAFF S-Spec CPU and choose many other options supported by the motherboard.

EPIA-M900 power usage and performance

This is the last article from my series about my EPIA-M900 system (first, second). This time I'll focus on power usage numbers, some small performance tests and overall feel.

I used  Tunex multifunction power meter to measure a power usage. All tests were made using Windows 7 64-bit edition. The whole system (including eH1 graphics card and HDD) without monitor uses 44-46W on idle. 44W is more constant than higher values. I tested system on BIOS without graphics card too and it showed 29-31W values. So graphics card adds about 15W on idle. Later I used AIDA64 stress tests and GPU-Z render test to simulate system's maximum power usage with and without HDD stress test. On full load (CPU and GPU) power meter showed 75-77W and 73-74W with and without HDD stress test respectively.  That was the maximum power usage numbers I managed to get on this system. GPU render test raised power usage to 55-56W. CPU load only increased power usage more than GPU load to 68-69W. Finally, while playing Starcraft II this range was between 70-71W. So it is difficult to say how much power usage increased because of GPU or CPU as other components takes some too and overall power usage doesn't increase on the same level as on CPU or GPU load separately.

While power usage numbers are quite good we should look into performance. As of 26 February, 2010 1.6GHz (1.73GHz on turbo) is the maximum frequency you can get from dual core VIA CPUs. VIA Nano X2 L4350E TDP is 27.5W. So you can compare it to low usage mobile CPU (if I am correct, current standard mobile CPU's TDP goes from 25W-45W typically) by TDP. Unfortunately, performance is on the lower range and you need to compare them to Intel Atom and AMD APU from bobcat core series. This is the area where VIA CPU is really competitive. But this article won't make any summary and won't make big comparisons . I will provide only some raw numbers for your investigation and reference. 


Firstly I used y-cruncher multithreaded pi benchmark application. I used 50,000,000 decimal digits test and Linux 0.5.5 Build 9187 version. Total result was 126.278 seconds for this test. Single core VIA Nano L2200@1.6GHz completed test in 299.358 seconds. U2300@997MHz needed 497.043 seconds. The same test was slower on Windows for unknown for me reason (about 20 seconds in total). If you want to compare this result to other CPUs, it is comparable to Intel Celeron T3100 result.

I also ran AIDA64 tests (v2.00.1700). I didn't tested L2200 but AIDA64 had results in their table. Memory read using DDR3-1066 SDRAM 7-7-7-20 CR2 was 3747 MB/s (Nano L2200 with DDR2 3352  MB/s). While memory write test was much faster than L2200, 7138 MB/s against 3157 MB/s. Memory copy 4944 MB/s vs 2759 MB/s. Memory latency was the same 117.3 ns on both systems. CPU Queen - 5890 vs 2580. This test was a little bit faster than AMD E-350 (5169). CPU PhotoWorxx - 5083 vs 2493. For reference Celeron 420 had almost the same result, E-350 had quite lower result at 4375. CPU Zlib - 34.2 MB/s, L2200 - 14.7 MB/s, E-350 31.0 MB/s, Core i5-650 HT - 101.1 MB/s. Padlock module helps VIA Nano CPUs to shine on CPU AES test where L4350E got 86803 points and lost only to CPUs with Intel AES support (though considerably) but others without it lost considerably (even 12xOpteron 2431 got 78761 points), L2200 - 40002 points. CPU Hash test gained 1054MB/s vs 549MB/s (E-350 achieved 326MB/s only). FPU was always quite a weak place for Centaur CPUs but Nano architecture improved FPU performance. FPU V8 436 vs 318, This test was bad for VIA CPUs, even E-350 got a little bit more here. On the hand it showed better results at FPU Julia and Mandel tests 1862vs892 and 832vs428 respectively (E-350 got results in between closer to L2200). Finally FPU SinJulia was a weak place too: 302vs132 while E-350 - 506. This is all tests I made. Tests showed quite mixed results but I believe that L4350 is comparable to first generation dual core AMD K8 CPUs or last generation Pentium 4 CPUs. Somewhere they will be quite faster, elsewhere much slower. They are faster than dual core HT Intel Atom CPUs (at least first generation). The last thing I could test was folding@home project. Unfortunately, this CPU is not suitable for SMP client. It is just to slow (dual core phenom ii@3.2GHz will do 1% 12-15 minutes than this CPU will need 2.5 hours). I also added Windows 7 performance index last time it is here once again (the CPU got minimum 3.9 result):

Windows 7 performance index

In general, it feels that CPU is much slower than Phenom II X2. Active processes like antivirus, flash, program loading grabs quite much from CPU. It is expected result of course. You can't compare apples and oranges this time. However, if you want to feel comfortable at your desktop, this system can be a little to slow. Though it is not that bad after you get into it. I want to change HDD to SSD some day. This should solve slower loading times and I guess should help to feel more responsive and fast while using the system. The only program which used hardware GPU acceleration in videos was Windows Media Explorer (MPC classic unfortunately took only CPU). I will try to investigate this in the future.

I didn't do any GPU tests. I tried to run some DX10/DX11 benchmarks but eH1 was unable to run them at normal FPS. However I played Starcraft II at 1280x1024 without any noticeable problems and with good graphics settings. While playing GPU temperature didn't raised more than 6-7°C from starting point (never reached 50°C).  I believe you can play most modern games with such configuration and not very big resolutions without problems.


So this is it about this system. I hope you enjoyed these articles and found them helpful. Thank you for reading.

EPIA-M900 system setup and Windows 7 start up

After I tested some open source operating systems I've decided not to play to long and started to assemble the main configuration for the desktop PC. It consists of EPIA-M900 motherboard, eH-1 graphics card, 2xKingston KVR1333D3S9/4G RAM (I am not really that Kingston fan but lately all RAM comes from them and I can't complain, they have very good compatibility with most boards), Samsung HD161HJ 160GB hard drive and In Win BM-639 case. Initially I planned to add CD-ROM too but some modifications disallowed that (later on that). RAM works at 1066MHz effective speed because it is the maximum VX900 supports.

The main purpose was to replace that annoying small CPU fan. So I perforated one side of the case above the CPU/chipset heatsink and installed 120mm fan. This "project" went ok and I was able to remove default fan. CPU cores temperature goes from 43°C on idle to 63-66°C max on load (speedfan show 5 degrees less than AIDA64 at the same time). CPU temperature is ~55°C by AIDA64. System temperature never rises above 31°C. The downside of this solution was that the fan blocked ability to add CD-ROM as original case design placed construction for 5.25'' drive just a little bit below that side of the case. GPU temperature range starts from 41.3°C on idle to 46-48°C on load. I firstly made one mistake during assembling because one wire stuck between GPU heatsink preventing fan from rotating. This lead to temperature from 81°C on idle to more than 126°C on load. You can understand from this that eH1 can work without fan :). Graphics card fan is surprisingly silent. I've read a lot of angry comments about active cooling solution in this card but it seems that they aren't reasonable this time (I know that fans tends to fail after some time but on the other hand you need to keep good airflow with passive cooling solution and need fans somewhere anyway). Unfortunately, the PCI-E riser was only 1.1 compatible so I needed to turn on the switch on graphics card to support slower PCI-E version. I tested that motherboard supports 2.0 version actually (at least graphics card worked switch in off mode). Finally, PCI-E slot is x16 physically only. x8 is effective speed of that slot.

AIDA64 CPUID

The other task was to fasten graphics card because I bought not suitable riser (I needed right angle one but I bought straight one. It lead to the problem that riser raised graphics card much above case panel for cards bracket fastening). The solution was simple. I inserted graphics card upside down (fan is on the top now). I needed to cut the bracket a little bit and make one hole in the it and the case for the screw but this solution needed minimal effort and almost no sacrifice for the future use of the case or graphics card (yes, the original bracket will not be suitable for normal insertion any more but I have replacements and you probably will find them at various IT stores too). Unfortunately my camera is broken now because of that I couldn't to make photos of these modifications but I'll try to do them in the future.

GPU-Z (some information inaccurate)

The last thing I wanted to do before Windows installation was BIOS update. I found the newer version on viaembedded web page and I made an upgrade. Firstly, it seemed that everything went ok. However, after I loaded system defaults, motherboard just didn't boot anymore. It was very unpleasant moment because I though that motherboard just died because of unsuitable BIOS version. Fortunately it booted after I changed RAM to some spares I had had at home. BIOS showed incorrect checksum error on boot but I was very happy nevertheless as motherboard was alive at least. I downgraded BIOS back to old version (fortunately I made a backup of it because you couldn't download it from viaembedded web page). I don't know why VIA decided to add this BIOS version without any release notes and description but they will gain lots of unsatisfied users with such irresponsible move. These days BIOS upgrades are simple and painless process usually, because of this users will try to do this from time to time. Such surprises should be avoided and BIOS purpose and suitability for particular model stated clearly. Anyway I recommend to use original BIOS versions with VIA motherboards :)...


After all preparations I was able to try Windows 7 (64-bit) finally. I wasn't installing them cleanly, just tried to boot into installed system from previous computer. This process went smoothly... almost. Every time I tried to connect to the network, Windows was hanging up and rebooting with good, old BSOD in the end. Solution was simple yet again, I needed to install network drivers from CD or web page. It seems that Windows update provided faulty drivers for some reason. I also installed the newest 500 series Chrome drivers from s3graphics web page instead of the ones from CD or viaembedded eH1 web page. All other devices were recognized by Windows and successfully installed into the system. Due to system resources requirement for physical memory addresses Windows sees 7.25GB of RAM instead of 8GB. I will discuss some performance results and power usage in the next article.

Windows 7 rating

Shift back to mini-itx (first experience)

So so... As I promised after new year I had some hardware changes. I returned to mini-itx platform once again. This time I had more reasons than just experiment as it had been with VT-310DP. Firstly, I wanted to lower power usage (I don't play much and don't do big computational projects, because of this Intel system is just absolutely unused power beast for me. Yes, I thought about it when I had bought it, however, my expectations didn't materialized and objectives changed a little bit). Secondly, I expected somehow better compatibility with various operating systems and easier environment configurations. New Intel platform couldn't provide that. What is more, I hoped to set up everything with as little time as possible ("perfecting" work place for a half or one year is just to much). Lastly I needed more space for my legs :) (there is not much space under the table)...

Purchase

Unfortunately, purchase process didn't go as well as I had planned initially. I ordered MSI AM-780E mini-itx motherboard with dual-core Athlon Neo CPU in January, however, it appeared that it hadn't been released yet. I was told to wait until mid of March. And I waited... but then I got an answer that I should wait one more month and it might be not released at all. I decided to change it as I didn't want waste time anymore and I ordered the most similar motherboard which e-shop had (from Germany). That was Jetway JNC-81 with quite similar configuration and the same chipset. The main reason I was a little bit disappointed because I didn't like mini-itx motherboards with CPU sockets. I see true mini-itx motherboards only with soldered very low power  cpus. However, I didn't have a choice (I wanted dual core CPU and, of course, Intel Atom wasn't an option). I chose Athlon X2 235e CPU for it (as it was an only 45W dual core in stock in Lithuania. Quad core 600e was available too but it was considerably more expensive). I had some problems in acquiring RAM too. I wanted to get Kingston CL5 but they had become out of stock just after my order. I was forced to get CL6. MSI AM-780E was released eventually but I prepared my computer for work by that time.

The other parts included Akiwa GHB-B05-5 enclosure, Seasonic SS-250SU 250W PSU and Jetway AM2 Low Profile heatsink for JNC81. Hard drives were taken from old computer as always.

Setup

Current configuration is:

Athlon X2 235e 2.7 GHz 45W TDP
Jetway JNC81 
Jetway AM2 Low Profile heatsink for JNC81 (replaced with default AMD cooler)
2x2GB Kingston KVR800D2S6/2G RAM SO-DIMM DDR2 800
Seasonic SS-250SU 250W PSU
Akiwa GHB-B05-5 mini-itx enclosure
(old ones)
Samsung HD160JJ 160GB
WD CAVIAR GP 750GB (WD7500AACS)

Setup was quite smooth except that I installed one RAM module incorrectly at first (I needed to push it more firmly). First problem was loud fans and hot operating temperature without case fan (it was included). I resolved this problem by changing Jetway fan with fan from AMD stock heatsink. It was a little bit higher but with controllable speed depending on load (four pins header) and lower noise. However, case fan was the bigger problem as it was just very loud. I slowed it with fan resistor. After that noise level and temperatures became acceptable. I found one BIOS problem. If you turn off quick power on self test, post tests RAM continuously. You must to press esc to process further (or just to set quick power on self test to enable which is default). Integrated video takes 128 MB from your RAM by default (+128MB sideport memory soldered to motherboard). There is an issue that I didn't resolve. USB ports are working in slow 1.0 mode from USB headers (aka front USB). USB ports in the back panel works correctly in ehci (2.0) mode. I might resolve this in the future.

Windows, NetBSD, Linux, Haiku

I didn't need to reinstall Windows 7 (more accurately Windows Embedded 7 CTP1). It recognized drivers correctly and everything worked out of the box except that I needed to reinstall graphic driver  (ATI catalyst). 1GBit network download/upload speed was slower from my NAS server with one big (0.5-4GB) file than in Intel platform (~30-40MB/s instead of 50MB/s) but it appears to be a usual speed from Realtek NICs. I had an issue with my old network cable (which came with router). Then I inserted it network worked only at 100MBit/s. Solution was to change a cable...

My other long time dream was to make NetBSD as primary desktop system. This time I tried to make it seriously. As I had some experience with it after some years playing with it and using it on VT310-DP NAS server it helped me a lot this time. I decided to compile most packages from pkgsrc (not downloading binaries). It might be slower but you just need to start compiling (make install clean) and all dependencies will be fetched correctly and without interruptions. I don't know why but it is not the case when you set PKG_PATH variable and download binaries. You will always find stalled downloads this way (especially if there are a lot of dependencies). It took me four-five days and two complete reinstall to setup everything but in the end I got the most usable desktop NetBSD system I have ever had. I doesn't mean that there are no problems but I have Gnome 2.28.1, OpenOffice 3.1, Firefox 3.5.8 with flash plug-in (the biggest headache was to make sound work in flash), developing enviroment (gcc, opejdk7, gtk, qt, some IDE, RAD tools, shell, etc), video, music, samba, ntfs-3g, radeon-hd, correct shutdown and so on. Unfortunately, cpu scaling doesn't work and it seems a problem with ACPI driver to manage CPU p-states. There is no proper driver in NetBSD currently. I need to restart pulseaudio every time after restart to make sound work. Videos doesn't work full screen and even with bigger resolutions. I recompiled kernel with newer release 5-0 maintenance kernel (2010-04-17). It solved problem with detecting USB devices (I needed to reattach mouse after first boot often with generic 5.0.2 kernel).

NetBSD 5.0.2 amd64

I didn't test Linux much but Fedora Linux 13 Beta1 installed without any problems. CPU scaling, network and sound works as it should. I don't think that it will much difference than with earlier systems after some configuration. Linux shows good support for a common hardware currently.

I even tried Haiku current (2010-04-17 or 2010-04-18). It installed and booted without problems too. Network worked from the start (I don't know if 1Gbit/s works). I played with several applications but it was limited to that.

That's all for now.