So AMD has their 8350 Black with 8 cores, and Intel has their 4770k with 4 cores. The 8350 runs at 4GHz stock and the 4770k at 3.5, yet I see almost everyone using the 4770k. Why is this? Other than the 125W thermals compared to Intel's 84 I can't see why you would use Intel's. Am I missing something?
You are missing many things... lol
Clock speed doesn't equate to a better CPU. Look at it this way. The Intel Pentium 4 came out in 2004 with a clock speed of 3.8Ghz. Is it faster than a 4770? Of course not.
Just because the clock is higher doesn't mean the CPU itself is faster or performs better. Much is down to IPC. Instructions per clock. IE How much work the CPU can do for every tick of clock speed.
The 4770 has much better IPC and a few other features and lots of other microarchitecture things that are quite complex and difficult to explain succinctly.
The 4770 is faster that the 8350 because it can do more work per Ghz. It is a faster CPU and while in gaming they are usually pretty similar, the i7 is the better part.
That is coming from someone with an 8350.
Although the 8350 has 4 more cores, doesn't that come into play too? I thought it assisted in video encoding and cpu heavy tasks, and it's 150 bucks cheaper from newegg - is it better in price/performance?
Yes that does come into play.
The 8350 does have 4 more cores although as they are split into pairs that share a decoder their performance is a bit nerfed compared to what it could have been.
Yes you are correct in some tasks the 8350 will be better than the i7 or at least about the same. Video encoding is one of them. H.264 encoding on the 8350 is amazing. It is faster than even the Socket 2011 Intel 4820k in that aspect. It also runs VMs like a boss.
The thing is though, the software it is executing has to be specifically coded to take advantage of all of the cores otherwise it doesn't perform as well. In well optimized tasks it is nearly as quick as the i7 but in poorly optimized tasks it is much slower.
In terms of price to performance though you are very much correct. The 8350 and especially the slightly lower clocked 8320 are way better value for money. The 4770k is overpriced in my opinion. You'd be much better going with the AMD or the 4670k i5, or the Xeon 1230 V3 (i7-4770 without the iGPU and $80 off).
"Xeon 1230 V3" only problem with Xeons is the locked multipliers, which with haswell means next to impossible overclocking.
I don't plan on overclocking anyways so no biggie
I like many others use Handbrake for video encoding, would that be on the list for "well optimized tasks"? And the motherboards I looked at only support up to PCIe 2.0 x16, is that a problem? Can I still buy modern cards and throw them in?
Wut. Yes it has a locked multiplier and really is impossible to overclock but that has nothing to do with it being a Haswell part. Any CPU with a locked multiplier is difficult/impossible to OC. You can screw around with the bus but that really isn't recommended.
H.264 encoding with Handbrake is a "well optimized task." Additionally it is a great gaming CPU and will only be slower in poorly optimized titles like those based on the ARMA Engine (DayZ) and WoW. Those like single core performance but even then they will still be perfectly playable.
Yes PCIE is an amazing interface and is backwards, forwards and sideways compatible (okay I made that last one up). You can put a PCIE 3.0 card into a 2.0 slot no problem. A X16 2.0 slot is slower than a X16 3.0 slot but it doesn't really matter because no modern GPU can even fully take advantage of the X16 2.0 bandwidth. You can even run a GPU in a X8 2.0 slot and it will be just fine.
No AM3+ motherboard has support for PCIE 3.0. There was a revision of the 990FX Sabertooth that ASUS lists as having 3.0 capability but I don't believe they make it anymore. it is impossible to find.
Ok thanks!
I can do my best to explain.
AMD are running with a cluster-core technology (what they call modules), which is in basics melting 2 cores together and removing the redundant parts.
So the Fx 8320 have 4 CMT cores.
Intel is running with SMT technology (what they call Hyper-Threading), which lets another thread (Which is in basics a line of code) run simultaneously. This is possible because Intels architecture have alot of execution units in each core, so it is hard to utilize them all by a single thread.
Lets get into the back-end of both architecture; (Back-end is the execution units in the core).
In the backend of the core, it is in basics splitted up to be either Integer (Full numbers. Example 1, 2, 3, 4 and etc etc) or floating point values (decimal numbers. Example = 3.14).
The integer values are handled by the ALUs in the core and the fp (floating point) are handled by the FPU (floating point units, better callled SIMD).
Piledriver (The architecture of the fx 8350 processor);
Per CMT core there is;
4 ALUs (2 per core).
4x128bit SIMD
Haswell (The architecture of the 4770k)
Per haswell core:
4 ALUs
3x256bit SIMD.
So a Haswell core have the same amount of ALUs as a CMT core, but have a superior SIMD over piledriver.
But since a CMT core, split up its backend to two "cores", so a single core (NON-CMT) from piledriver have 2 ALUs and are sharing the 4x128bit SIMD with the other core in the CMT core. (Sounds very misleading, but it is hard to explain. There are 2 cores in a CMT core...)
This is the reason why Intels singlecore performance are so much greater than piledrivers, and the reason why they are tie at higher multi-threaded integer workload.
fsb/strap overclocking :P there is more than one way to Overclock a CPU
In the Piledriver AMD does use a SMT like technology to get more use from the shared FPU in each module http://cdn3.wccftech.com/wp-content/uploads/2013/07/AMD-Steamroller-vs-Bulldozer.jpg , but its kinda sorta not perfect.
It is called CMT.
Its called FlexFPU and its part of CMT since Piledriver.
http://www.xbitlabs.com/news/cpu/display/20101026234515_AMD_Calls_New_FPU_Flex_FP_Defends_Dual_FMAC_Approach.html