Natural Opinions

An Archive Of My Thoughts

The FX-4100

Price:  $125, priced the same as a low end phenom II X4 (like a 960T) or core i3

Heat:  Similar to a phenom II X3

Power consumption:  Similar to a phenom II X3

Overclocking:  Can be clocked at 4.5GHz while maintaining surprisingly low power consumption/heat

Single/dual threaded performance:  Utterly destroyed by a phenom II, athlon II, core i3, or pentium (sandy bridge)

Triple threaded performance:  Similar performance to an athlon II X4

Quad threaded performance:  Beats an athlon II X4 but loses to a phenom II X4

Conclusion:  The phenom II X4 960T is better in every way for the same price, core i3 is faster in all but the most heavily multithreaded applications, consumes far less power, lower heat, includes an IGP, etc.

 

The FX-6100

Price:  $190, priced higher than most phenom II X6 and same as an i5 2400

Heat:  Similar to a phenom II X3

Power Consumption:  Similar to a phenom II X4

Overclocking:  Can be clocked at 4.5GHz, power consumption and heat at that clock rate are similar to the competition

Single/dual threaded performance:  Utterly destroyed by a phenom II, athlon II, core i3, or pentium (sandy bridge)

Triple threaded performance:  Similar to phenom II X4, beats an i3 but loses to an i5

Quad threaded performance:  Beats an athlon X4 substantially but still loses slightly to a phenom II X4

Hexa threaded performance:  Similar to a phenom II X4, beats an i3 but loses to an i5

Conclusion:  Phenom II X6 cpus are cheaper and better in every way.  Core i5s are the same price and phenominally better in every way.

 

 

The FX-8120

Not enough data availible, most sites didn't bother to benchmark it

 

The FX-8150

Price:  $245, priced higher than any core i5 or phenom II X6, but lower than any core i7

Heat:  Similar to a phenom II X4

Power Consumption:  Higher than a phenom II X6 or core i7

Overclocking:  Can be overclocked up to 4.5GHz but will consume an absurd amount of power and generate an absurd amount of heat at that clock rate

Single/dual threaded performance:  Utterly destroyed by a phenom II, athlon II, core i3, or pentium (sandy bridge)

Triple threaded performance:  Similar to a phenom II X4, beats an i3 but loses to an i5

Quad threaded performance:  Similar to a phenom II X4, beats an i3 but loses to an i5

Hexa threaded performance:  Beats a phenom II X4 but loses to a phenom II X6, similar to an i5 2300

Octa threaded performace:  Beats a phenom II X6, similar to an i5 2500

Conclusion:  An i5 2500K is cheaper, faster in every test, consumes less power, produces less heat, overclocks higher, has an IGP, and is massively faster in single/dual threaded tests, and most applications are still dual threaded.  Not only can I not recommend this cpu over what the competition offers but I can't even recommend this over AMDs previous generation of cpus.  A phenom II X6 is faster in all but the most heavily multithreaded applications while consuming less power and requiring less cooling.

 

Overall conclusion of bulldozer

Oh god why!  All 4 of these cpus are complete shit at their respective prices.  These cpus are not only inferior to the competition but they are actually inferior to the phenom II cpus that they are supposed to be replacing (which AMD will be shutting down production for soon).  These things will not sell without a massive price drop.

The FX-4100 would need to be priced at around $75 and competing with pentium (sandy bridge) and athlon II to remain competitive.

The FX-6100 would need to be priced around $125 to be competitive since upper end phenom II X4 cpus are clearly better options.

The FX-8150 would need to be priced at around $160 to be competitive since phenom II X6 and core i5 cpus are clearly better options.

The only benefit of this new architecture is that is that FX-4100 and FX-6100 do have slightly better performance per watt in heavily multithreaded applications  than the phenom II cpus that they are replacing.


I must point out that this situation is identical to the situation that the original phenom cpus were in when they were launched in 2008 to replace the third generation athlon X2s.  The top of the line phenom 9700 (2.4GHz) was slower than the 3.2GHz athlon X2 brisbane it was supposed to replace in everything except heavily multithreaded software (which was, and still is, rare).  It also consumed substantially more power than either the athlon X2s it was replacing or the core 2 duo/quads it was competing against and ran much hotter.  It did not overclock very well and had absurd levels of power consumption and heat when overclocked.  The memory controllers, hypertransport bus, caches, and core frontend were all improved extensively, but none of this made much difference to the IPC and the additional heat/power consumption of having 4 cores forced them to lower the clock rate which reduces single/dual threaded performance.  Triple threaded software ran at similar speeds on both the 3.2 GHz athlon X2 and 2.4GHz phenom.  Only when running software with 4 well balanced threads could the phenom X4 claim its performance crown over athlon X2.  Of course even though the situation is the same the reasons for the situation were very different.  Phenom did actually improve the IPC over its predecessor, but not enough to offset the reduced clock rates.  Bulldozer on the other hand has increased clock rates over its predecessor, but not to enough to offset the reduced IPC.  Both architectures were forced to make cutbacks to accommodate having additional cores (which raises heat/power consumption).  They went in different directions (one cut back clock rate, the other cut back on functional units which reduced IPC) but ended up with the same result, reduced single/dual threaded performance over their predecessor.  Phenom II improved both IPC (mainly due to a larger/faster cache) and clock rate and brought single/dual threaded performance above athlon X2 levels (though still slightly lower than core 2 levels).  We can only hope that piledriver will do for bulldozer what deneb did for agena (which would bring bulldozers single threaded performance above nehalem levels but still below sandy bridge levels).  CMT is a really good idea but bulldozers IPC suffers as a result of numerous other unrelated issues and cutbacks.

Bulldozers named instruction set extensions:

FMA4 (introduced with bulldozer 2011)

XOP (introduced with bulldozer 2011)

CVT16 (introduced with bulldozer 2011)

AVX (introduced with sandy bridge and bulldozer, 2011)

AES-NI (introduced with westmere, 2010, and bulldozer 2011)

SSE4.2 (introduced with Nehalem, 2009, and bulldozer 2011)

SSE4a (introduced with K10, 2008)

SSE4.1 (introduced with core 2 duo, 2008, and bulldozer, 2011)

SSSE3 (introduced with core 2 duo, 2006, and bulldozer, 2011)

AMD-V (introduced with athlon 64, 2006)

SSE3 (introduced with pentium IV Prescott, 2004, and athlon 64, 2005)

X86-64 (introduced with athlon 64, 2003, and Pentium IV Prescott, 2004)

SSE2 (introduced with penium IV, 2000, and athlon 64, 2003)

SSE (introduced with Pentium III, 1999, and athlon XP, 2001)

3DNOW!+ (introduced with K7, 1999, sometimes called 3DNOW!2)

3DNOW! (introduced with K6-II, 1998)

MMX+ (introduced with K7, 1999, sometimes called MMX2)

MMX (introduced with Pentium MMX, 1996 and K7, 1999)

PAE (introduced with Pentium pro, 1995, and athlon XP, 2001)

 

Bulldozer has the most complete version of the x86 instruction set to date.  The only extension that intel cpus have that bulldozer does not is VT-x, which is intels version of AMD-V.  All x86 cpus up to Pentium P5 (1993) included new instructions but did not provide a name for their instruction set extension.  All modern intel/AMD cpus support these old extensions.