Originally Posted by orlfman When one is on a budget where even saving $30 - $40 matters
because that extra money can go to a better cooler to allow overclocking or something else. I went with my 8350 over a 4670k because, pairing it with a similar motherboard, I saved $70... which allowed me to get a better gpu. If had gone with the 4670K I would of been stuck on stock cooling and a weaker gpu. Not all of else ooze money like crazy and can't always save every single penny we get because of the situations where in.
Getting that CPU that will play everything out there at max settings with acceptable frame rates and saving money is well worth it.
The problem is that the FX-8350 doesn't save $30-40 all things considered.
Your build is a perfect example of why the FX-8350 isn't a good value CPU for gaming, lets explore why:
The FX-8350 has the same binning quality as an FX-6300 (I can explain this if you're interested), so once the overclocking starts, there's no advantage in terms of overclocking potential on average. The only advantage is the additional module. The FX-8350 costs 66% more than the FX-6300, yet only delivers 33% more computing resources.
The i7-4770K, is normally priced ~40% higher than the i5-4670K, and offers about the same ~33% improvement in parallelism over the i5-4670K that the FX-8350 does over the FX-6300. Common practice is to recommend the i5 over the i7 because the additional ~33% parallelism afforded by the i7 does not translate to much gaming performance since most game engines don't scale into the additional parallelism very well (if at all). If good practice is to recommend AGAINST spending an additional 40% on the CPU for 33% more performance-through-parallelism on the Intel side when value is a high priority, why would it be good practice to recommend spending 66% more for 33% more performance-through-parallelism on the AMD side? The answer, is that the FX-8350 represents an even poorer value over the FX-6300, than the i7 does over the i5 for gaming, yet for some reason, people consider the FX-8350 to be a good value gaming chip. This baffles me.
Clock for clock, a non-hyperthreaded haswell core delivers the same computing performance as an entire piledriver module combined in most workloads. By consolidating that performance into half as many cores, the i5-4670K offers substantially higher per-thread performance, which, unlike additional performance-through-parallelism, scales proportionally in CPU bound real-time workloads. The FX-8350 costs 15% less than the i5-4670K (typically, though right now it's only $10 cheaper), yet performs up to 40% worse in CPU bound real-time workloads. How does that make the FX-8350 a better value? It doesn't.
When switching from the FX-8350 to the i5-4670K, there are numerous Z87 boards to choose from in the same price class as the value conscious enthusiast 990X EVO, 990FX killer, and 990FX UD3. Motherboard price is a wash between these platforms. For example, the popular Z87 Killer, Extreme4, Z87-A, Z87-G55, Z87X-D3H, Z87X-HD3, all comparable in many ways. I would argue that some of those Z87 boards are actually MORE feature rich for the money than the 990X EVO.
When switching from an overclocked FX-8350, to an overclocked i5-4670K, we can subtract up to
~150W (or more) from the power supply sizing all other things being equal and maintain the same power available for the rest of the system. This may be hard to believe so I'm going to back this up with some evidence: http://www.xbitlabs.com/articles/cpu/display/core-i5-4670k-4670-4570-4430_7.html
Note, the result there was measured after the PSU, so the 990FX+FX-8350 system under load at stock clocks, dissipates 95W more than the Z87+i5-4670K. This is partly because the 990FX chip-set has a higher TDP than the Z87 chip-set (by about 15W), but the main difference comes from the CPUs themselves and the differences in VRM efficiency at these significantly different loads. The real-world gap in CPU power dissipation is much wider than the TDP ratings
might suggest. In fact, the i5-4670K dissipates closer to 70W at-the-chip
under full load at stock clocks, while the FX-8350 dissipates closer to 140W at-the-chip
under full load. When we overclock, the gap opens up much wider. The FX-8350 soars well over 200W when overclocked. (~220W @ 4.7ghz is common, and 250W+ if we push for 5GHZ+). CPU VRMs on the overclocked 990FX platform are forced to deliver up to ~15-20A per phase, which forces their efficiency downward to ~80-85%. On the Z87 platform, many overclocking friendly boards are designed such that they achieve peak efficiency at overclocked power levels of ~90% (~10A per phase). The i5-4670K is usually limited to ~125W@ ~4.5ghz by the IHS, so this is the practical stopping point for overclocking.
Power consumptions breakdown:
Z87 (5W) + i5-4570K @ ~4.5GHZ (~125W) + VRM losses (15W) = 145W
990FX (20W) + FX-8350 @ ~4.7GHZ (~220W) + VRM losses (50W) = 290W
This difference in power dissipation can have an impact on the required investment in CPU cooling and PSU size, such that the difference in the price of the CPUs themselves, is generally washed out, if not effectively reversed. The FX-8350 requires ~240mm radiators or large double tower heatpipe cooler to hit an overclock that will have any
potential to give it any edge over the i5 (unfortunately, any edge it gains won't be particularly useful for real-time workloads like games so this may prove to be a moot effort regardless). Remember, the performance of a PD module is about the same as a non-hyper-threaded haswell core, so if we don't overclock the 8350 higher than the i5, then we've given up the only performance advantage the 8350 can ever have in those unusually parallel workloads. Meanwhile, the i5-4670K isn't going to get much past ~4.5ghz whether we spend $30 or $100 on a CPU cooler due to the limits of the IHS, and since the dissipation is so much lower, a basic single tower direct touch heatpipe cooler, or value oriented 120mm rad, is all that is required to run at ~4.5ghz. If we put a budget (<$50) CPU cooler on the FX-8350, it won't go past ~4.5ghz anyway, which basically sets it up to only ever match the i5 in non-real-time workloads that can saturate the execution resources absolutely. In all other workloads, the 4670K performs better. Then there's the issue of PSU sizing. The 100-200W larger PSU size required to implement the overclocked FX-8350, easily eclipses the difference in the price of these CPUs by itself. The i5-4670K + 212 EVO + 650W PSU will cost less than the FX-8350 + H100i + 850W PSU, offers very similar available power headroom for GPUs, and performs better in most games, with the "worst" performance of the i5 matching the 8350's best.
The same differences in compute efficiency that make Piledriver based opterons obsolete in a world of modern 22nm Xeons in the professional world due to implementation costs, come back to bite for value oriented performance tuners as well. The only way for AMD to be competitive, is to offer chips at prices that leave enough budget in place to pay for the higher implementation costs associated with the higher power dissipation. The FX-6300 in the $110-120 range accomplishes this very well, as it has enough unused "budget" to go toe to toe (+/-20% depending on workload) with locked i5 chips in the ~$180 price class. If the FX-8320 were dropped to ~$140 MSRP, and the FX-8350 were dropped to ~$160 MSRP, then it would actually be possible to implement these chips at price points that would be more legitimate in terms of their performance in real world conditions. Paying $200 for an FX-8350 in a world where we can get an FX-6300 for $120, would be like paying $400 for the i7-4770K in a world where we can buy the i5-4670K for $240.
EricEdited by mdocod - 4/24/14 at 12:15am