Originally Posted by Warsteiner
You might want to think about an APU build with a 7670 that will crossfire with the APU. This should be attainable in this price range and would give you a newer platform that will have better longevity.
This should be a last ditch resort if you truly can not get a discrete solution in budget OR if the build has absolutely no need for "good" GPU performance. APUs are best leveraged in mixed use extreme budget machines. The GPU performance of an APU is respectable considering the integrated competition and typical performance of integrated GPUs of the past, however, can't compete with discrete solutions except on the very low end. As soon as you hybrid X-fire these, you're just over-paying for a band-aid solution to a problem that won't go away until the switch to a discrete card is made.
Starting off a build as a hybrid X-fired configuration is a HORRIFIC TRADEOFF that should never be made at the onset of a build. The HD6670 gets purchased AFTER YOU HAVE ALREADY MADE THE MISTAKE of buying into an APU thinking it would have enough GPU grunt.
Originally Posted by MAKATTACK
Thanks guys. I would like to stick with quad core machines since dual core is becoming fairly outdated. Now the question is....go with AM3 or FM1....
Having "more cores" is not a straight forward solution to "not being outdated." One could make the same argument for buying into a CPU architecture that is several years old.
A quad core phenom II 955/965:
...can execute on up to 4 integer or 4 FPU units simultaneously, any combination of up to 4 threads scales with perfect linearity. The older traditional architecture and mediocre cache performance results in a still more-than-acceptable per-core performance for modern machines for most peoples needs and very predictable performance scaling with no special conditional requirements to achieve good scaling. Slightly out-performs i3 in highly parallel integer heavy workloads. Beats quad FX and i3 in most parallel FPU workloads. The now hard to find thuban variety was respectably power friendly in it's day at 95W TDP. The older Deneb based 125W chips on the market today represent very poor compute/watt performance. The sub $100 pricing of this chip is competitive, however, keep in mind that it should not be used on a cheap motherboard with a small VRM package. Cost to implement can prove higher than expected when you factor in a quiet cooling solution and a motherboard that can be trusted to handle the chip, this is compounded if you intend to overclock. The high implementation cost to overclock this chip is superseded by a similar price i5 build.
A dual core i3 Sandy/Ivy:
...can execute on it's integer and FPU modules simultaneously per cycle per core. Up to 2 integer and
2 FPU calculations (4 threads) per cycle is made possible though a fetch/decode/scheduler configuration that is scaled up from the traditional dual core approach. Superb cache performance and refined architecture proves highly effective in per cycle/core performance. Performance scaling above 2 threads is conditionally dependent on the type of workload. The i3 beats stock clocked AMD quads in less parallel workloads and trades blows in parallel workloads that are mixed FP/integer (this is common). As low as 47W TDP makes this the most power friendly option in the comparison here, with the highest compute/watt ratio. Known for superb performance in 1-3 threaded games that are highly sensitive to on-chip cache performance (like SC2's (supcom and starcraft). No overclocking possible. The typical $100-145 price range on these chips is competitive for their level of performance and low cost to implement (cheap motherboards are fine and the stock fan can be configured to run quiet without a problem!).
A quad core FX chip:
...contains 4 integer and 4 FPU units, however, has a fetch/decode/scheduler configuration that is scaled down from a traditional quad (pairs of integer cores are sharing fetch and decode units, and pairs of FPUs are sharing a single FPU scheduler) . Turbo core is used to accelerate less threaded tasks and software is now catching up to properly schedule tasks to separate modules first. The result, is conditionally dependent scaling beyond 1-2 threads that is better than an i3 but worse than phenom II. Poor cache performance and a very young architecture translate to poor per-cycle performance (worse than phenom II). Typically beats i3 and Phenom II X4 in parallel integer
workloads. Trades blows with Phenom II and i3 in mixed parallel workloads. Beats Phenom II in most less threaded workloads through raw clock speed. High power consumption means similar cost to implement and cool quietly as phenom II. Pick Vishera to improve the compute/watt performance considerably. Best value with these chips is going to be a mild overclock attainable on stock or cheap coolers. Investing more for a high overclock is again, superseded by i5 as the smarter implementation for the money. The $10 difference between the FX4300 and 6300 basically nullifies the the 4300 as a contender in my eyes. The extra module on the 6300 guarantees a win over the i3 in parallel workloads every time. Cost to implement after factoring in a motherboard you can trust to run the chip is still higher than the i3, but Vishera has certainly helped the argument for an FX build.
One thing that I have found in my research, is that currently, a hyper-threaded Sandy bridge core, has similar compute performance as a bulldozer module. A hyperthreaded Ivy bridge has similar compute performance as a Piledriver module. Clock rates, cache sizes, and workload cause fluctuations in results, but that rule of thumb holds close to true often enough that it is a good "guide" to follow on your path to finding the best value for your build.
The point I am making here is not that the dual core Intel solution, OR the AMD quad solution is better, rather, that there are TRADEOFFS to be considered. Don't dismiss, rather, honestly compare your options.
Nobody asked... What is this machine going to be USED FOR???
EricEdited by mdocod - 11/13/12 at 12:38am