Well after the price dropped I decided to buy a cheap Phenom 9650 as a drop-in replacement for my aging Athlon X2. By selling the old CPU, the upgrade cost was pretty minimal. My next upgrade will probably be a DDR3/45nm Phenom or Nehalem (depending on relative pricing and performance), but this CPU should hold me over until those become a little cheaper and more mainstream (spring/summer 2009 time frame).
I'm using an older 690 board, probably the cheapest AM2 board on the market as it's a special, simplified version of the already simplistic M2A-VM budget m-ATX board. This means no HyperTransport3, but the difference is said to be negligible. CnQ 2.0 is also non-functional which means slightly higher idle temp/power draw. Other than that, the ability to drop these CPUs into pretty much any AM2 board is a pretty big advantage and the board, while only certified for 95W CPUs (which means the X4 9650 and X3 8750 are as high as it will go, at least until Deneb), has been rock solid so far.
Out with the old CPU, the 2.5 GHz Athlon64 X2 4800+ Brisbane at 2.9 Ghz = 5600+ speed
In with the new, Phenom X4 9650 at 2.3 Ghz
Ready for test run switching CPUs is one of the easiest upgrades you can do these days, thanks to the excellent design of the coolers and retention mechanism. Just align it and swing the small lever around. No chipped cores and no risk of running a freaking screwdriver through the mobo....
POST'ed without problems. BIOS Setup confirms that the CPU is detected correctly.
Cooler comparison X2 cooler to the left, Phenom X4 cooler to the right. They are very similar, very light aluminium heatsinks. The Phenom puts out nearly 50% more heat than my 65W X2, but even though the cooler looks almost the same, it doesn't seem to have any problems at stock speed.
Idle: 45C (using dynamic fan speed so cooler is nearly silent).
Full load: 55C - CPU fan reaches 3100 RPM which is not its maximum possible speed. That is either 3300 or 3500 RPM. You can hear the fan at this point, but it's not very loud. The PSU fan also spins slightly faster than before because of the increased load.
These temps are almost identical to my overclocked X2 temps as it would also max out at 55C with the CPU fan at about 90% but slightly higher than the the X2 at stock 4800+ speeds, full load at 52C.
Now for some benchmark results:
3DMark 06
Athlon X2 5600+ 2.9 Ghz
Total: 10042
SM2.0: 4425
SM3.0: 5197
CPU: 2076
Phenom 9650 2.3 Ghz:
Total: 10658
SM2.0: 4191
SM3.0: 4935
CPU: 3108
Comments: The Phenom is actually slightly slower in all game tests, but thanks to the higher synthetic CPU score, it still scores a higher total 3DMark result. Still, at 2.3 Ghz it's awfully close to the 2.9 Ghz X2 in every test 3DMark 06 does not take much advantage of multiple cores, except for the synthetic CPU tests.
3DMark Vantage
Athlon64 X2 5600+ 2.9 Ghz:
Total: 5789
GPU: 6564
CPU: 4269
Phenom X4 9650 2.3 Ghz:
Total: 6753
GPU: 6531
CPU: 7518
Comments: 3DMark Vantage is obviously better at taking advantage of the extra cores. It also does a good job of isolating the GPU for its GPU subscore it's virtually identical even when using two very different CPUs.
UT3 ShangriLa Flyby HOC Benchmark, High with tweaks no AA/AF, 1680x1050
Athlon X2 5600+ 2.9 Ghz: 56FPS
Phenom 9650 2.3 Ghz: 89 FPS
With 4X AA/16X AF:
Athlon X2 5600+ 2.9 GHz: 55 FPS
Phenom 9650 2.3 GHz: 55 FPS
Comments: Wow, the 2.3 Ghz Phenom is almost 60% faster than the 2.9 Ghz X2 and this is a real game I'm testing. Obviously it's GPU bound with FSAA/AF enabled. However this is just a flyby demo. In a botmatch or during online play, when the CPU is doing mre work, the Phenom would probably maintain a higher framerate even when there's lots of action going on while the X2 would struggle.
Call of Juarez DX10 Rolling Demo, 1680x1050
Athlon X2 5600+ 2.9 GHz:
Min: 17.8
Max: 75.5
Avg. 35.9
Phenom 9650 2.3 GHz:
Min: 20
Max: 80.2
Avg: 36.1
With 4X AA
Athlon X2 5600+ 2.9GHz:
Min:10.6
Max: 54.6
Avg: 23
Phenom 9650 2.3GHz:
Min: 11.1
Max: 53.5
Avg: 23.1
Comments: Another GPU bound benchmark with no AI or physics active. Still, the 2.3 Ghz Phenom manages to outperform the 2.9 Ghz X2.
Oblivion, 1680x1050 4xAA Adaptive Q/16XAF, AEVWD, QTP3, Unique Landscapes, Better Cities plus many more mods
AthlonX2 5600+ 2.9 Ghz
Outdoors: 21
Market District: 12
Phenom X4 9650 2.3 Ghz
Outdoors: 20
Market District: 12
Comments: An older game that doesn't scale well with increasing number of cores. I'm using a ton of mods here (actually going to cut down on them to improve stability and performance) which means the framerate is only a fraction of what you get from Vanilla Oblivion. Again, the Phenom at 2.3 Ghz performs roughly the same as an AthlonX2 Brisbane at 2.9 Ghz. When you take into account that the fastest Phenom runs at 2.6 Ghz and Deneb will hopefully go much faster, it doesn't look all that bad. Of course that's without taking any Intel CPUs into account as nothing from AMD can touch any Intel CPU above ~2.5 Ghz.
Crysis 1680x1050 High DX10
Athlon64 X2 5600+ 2.9 Ghz
CPU test: 25.77
GPU test: 22.93
Phenom X4 9650 2.3 Ghz
CPU test: 27.56
GPU test: 26.76
Comments: Ironically, the GPU Test saw the biggest improvement when upgrading the CPU. Crysis is clearly not fully optimized for Quad core. Still, the 2.3 Ghz Phenom outperforms the 2.9 Ghz X2 in both tests. Let's hope Warhead is more optimized for Quad cores so the improvement is more like that observed in UT3.
Flight Simulator X, DX10, 4XAA Boeing Field Runway 13R highest detail settings
Athlon 64 X2 5600+ 2.9 Ghz: 11.9
Phenom 9650 2.3 Ghz: 11.3
Comments. FSX scales extremely poorly both with faster CPUs and with faster videocards (it's actually faster on the 9600GT than the GTX280 according to THG). The engine is supposed to be threaded, but it only uses multiple threads for streaming scenery and texture data faster, not to improve the framerate. The only way to improve framerate in FSX is through brute-force methods like high clockspeeds and high IPC's. The ideal CPU is probably one of those C2D's with crazy amounts of cache, overclocked to 4 Ghz. Still, the 2.9 Ghz X2 is only 5% faster despite the 26% clock speed advantage.
Bioshock Welcome To Rapture 1680x1050 16X AF
Athlon X2 5600+ 2.9 Ghz: 31
Phenom X4 9650 2.3 Ghz: 37
Comments: I decided on a particular location in the level with a lower framerate and saved the game there. I then read out the framerate using FRAPS. Again the Phenom wins the game feels smoother too. The slight mouse lag that was present with the X2 is completely gone the Phenom appears to be producing a more steady and even framerate with the FPS not dipping as much when there's lots of combat going on.
So in games that do not take advantage of multiple cores, the Phenom is about as fast as an X2 at between 2.8 and 2.9 Ghz. As soon as a game starts using the third and fourth core even to a small degree, the Phenom leaves the 5600+ X2 behind, being 60% faster in UT3 despite the 600 Mhz lower clockspeed, possibly even faster in an actual botmatch. Hopefully, the trend I observed with UT3 will be the norm for future games. In such games, the Phenom X3 and X4 can at least compete with Intel's Dual Core chips, allthough their Quad chips will of course blow the Phenom away.
The TLB issue and low clock speed have caused many to stick with their X2 or switch to an Intel platform, and never considered the Phenom as an option at all. The Phenom brand name is still associated with buggy, unstable and slow chips.This is unfortunate. Anyone in the market for a new AM2 CPU should look at the Phenoms rather than the K8 X2's IMO. The higher IPC already makes them faster in almost all apps, and as Quad core CPUs become more common and developer support increases, the Phenom will only increase its lead. If the board can take a 125W+ CPU, the higher clocked Phenoms can be had for not much more than the 9650 and they will outperform even the fastest X2's in virtually everything.
If you already have a fast Athlon X2, then whether a Phenom makes sense or not depends on what you use your computer for or even which games you play if you use it for gaming. Switching from the X2 6000+ to the X4 9650 for example will most likely reduce single-threaded performance slightly. Since I also use my computer for making music, the Phenom made a lot of sense. I can now use almsot twice as many virtual instruments in the DAW before the audio begins to stutter and playback fails.