LeStringMan

Infinity Fabric is based upon, and is an improvement on AMD's Hypertransport Link, HT Link was first introduced in 2003 in AMD cpus, and replaced the traditional Front Side Bus, Intel's QuickPath Interconnect does the same thing and was introduced in 2008. 

The FX series does very well on LN2, 2 volts is fine according to AMD's own overclocking guide (on LN2 obviously) and they can hit well above 8ghz, as for the lower temperatures at higher TDPs, this is the same as the Phenom II series, and as you probably know they do not have onboard temperature sensors of any kind, but rather they use an algorithm to estimate temperature. My guess is that whenever they hit that temperature, (at their pre-defined voltages) they are hitting the TDP limit. But AMD was never actually very clear about this. (AMD for the record uses a harsh maximum TDP meaning when the cpu hits its max TDP it throttles hard, Intel on the other hand uses more of an average TDP rating, meaning in some application it can go above its rating.)

Nobody knows where Vega will land, what we do know is that it has 64 compute units, (4096 Stream processors), its the same gpu used in the MI25 which AMD says has 12.5 teraflops, for it to hit 12.5 teraflops it would need a core clock of about 1525 megahertz, we also know that Vega has a lot of new software tricks such as something similar to tile-based rasterization like Nvidia cards now, so the teraflop value is useless compared to previous gen AMD cards, in theory it should be much closer to a 12.5 tflop Nvidia card now, but nobody can say for sure because nobody except AMD knows how efficient their new software tricks are.

Im not sure about ipc in gpus, i do know that Polaris for instance is about 20% faster on average than previous generations at the same clockspeeds (somebody did a test between 380x and 470 clocked identically) There are far more important things like keeping the rasterizers and rops fed (this is where Vega gets most of its changes), as for teraflops that value is pretty much just stream processor count times two times clockspeed, so the Fury X is (4096)(2)(1050)=8,601,600 flops or 8.6 teraflops.

Nobody knows where Vega will land, what we do know is that it has 64 compute units, (4096 Stream processors), its the same gpu used in the MI25 which AMD says has 12.5 teraflops, for it to hit 12.5 teraflops it would need a core clock of about 1525 megahertz, we also know that Vega has a lot of new software tricks such as something similar to tile-based rasterization like Nvidia cards now, so the teraflop value is useless compared to previous gen AMD cards, in theory it should be much closer to a 12.5 tflop Nvidia card now, but nobody can say for sure because nobody except AMD knows how efficient their new software tricks are.

Nobody knows where Vega will land, what we do know is that it has 64 compute units, (4096 Stream processors), its the same gpu used in the MI25 which AMD says has 12.5 teraflops, for it to hit 12.5 teraflops it would need a core clock of about 1525 megahertz, we also know that Vega has a lot of new software tricks such as something similar to tile-based rasterization like Nvidia cards now, so the teraflop value is useless compared to previous gen AMD cards, in theory it should be much closer to a 12.5 tflop Nvidia card now, but nobody can say for sure because nobody except AMD knows how efficient their new software tricks are.

Nobody knows where Vega will land, what we do know is that it has 64 compute units, (4096 Stream processors), its the same gpu used in the MI25 which AMD says has 12.5 teraflops, for it to hit 12.5 teraflops it would need a core clock of about 1525 megahertz, we also know that Vega has a lot of new software tricks such as something similar to tile-based rasterization like Nvidia cards now, so the teraflop value is useless compared to previous gen AMD cards, in theory it should be much closer to a 12.5 tflop Nvidia card now, but nobody can say for sure because nobody except AMD knows how efficient their new software tricks are.

Nobody knows where Vega will land, what we do know is that it has 64 compute units, (4096 Stream processors), its the same gpu used in the MI25 which AMD says has 12.5 teraflops, for it to hit 12.5 teraflops it would need a core clock of about 1525 megahertz, we also know that Vega has a lot of new software tricks such as something similar to tile-based rasterization like Nvidia cards now, so the teraflop value is useless compared to previous gen AMD cards, in theory it should be much closer to a 12.5 tflop Nvidia card now, but nobody can say for sure because nobody except AMD knows how efficient their new software tricks are.

Im not sure about ipc in gpus, i do know that Polaris for instance is about 20% faster on average than previous generations at the same clockspeeds (somebody did a test between 380x and 470 clocked identically) There are far more important things like keeping the rasterizers and rops fed (this is where Vega gets most of its changes), as for teraflops that value is pretty much just stream processor count times two times clockspeed, so the Fury X is (4096)(2)(1050)=8,601,600 flops or 8.6 teraflops.

It only has two stacks of ram which will be 8gb on a 2048 bit bus.

That's cache, not ram, GCN has 16 kilobytes of L1 cache per CU which is what thats showing, look at any other GCN card like a 260X and it will show the same.
https://compubench.com/device.jsp?benchmark=compu20d&os=Windows&api=cl&D=AMD+Radeon+R7+260X&testgroup=info

That's cache, not ram, GCN has 16 kilobytes of L1 cache per CU which is what thats showing, look at any other GCN card like a 260X and it will show the same.
https://compubench.com/device.jsp?benchmark=compu20d&os=Windows&api=cl&D=AMD+Radeon+R7+260X&testgroup=info

Vega is still GCN at heart and GCN is limited to 64 CU's.

There was a leaked slide from an AMD presentation that showed the R5 1600X as having the same clockspeeds as the 1800X.

It's one of Thermaltakes newer units, here is a review, comes close to platinum efficiency.
http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story&reid=499

Try bumping up your CPU-NB voltage, also make sure that APM is disabled if you are throttling back to 1.4ghz under load, if its just idling at that clock that is fine, no reason to disable Cool'N'Quiet. You can also try lowering the HT Link speed.

I would get a Scythe Fuma or a Thermalright Macho Rev B. over either the H5 or Dark Rock, the Scythe is by far the best ~50 dollar cooler on the market and the Macho Rev B beats the Dark Rock 3 by a degree or two and is just as quiet for a lower price.
Scythe Fuma:
https://www.amazon.com/Scythe-SCFM-1000-Compact-Socket-Cooler/dp/B01B53HG0W/ref=sr_1_1?ie=UTF8&qid=1485459743&sr=8-1&keywords=scythe+fuma
 Thermalright Macho Rev B.
https://www.amazon.com/Thermalright-814256001052-Macho-Rev-B/dp/B00PKJ21LW/ref=sr_1_1?ie=UTF8&qid=1485459771&sr=8-1&keywords=Thermalright+macho+rev+b

The Red Devil's vrms are slightly lower quality than the reference card, because the have a lower temperature tolerance, so a well known AMD overclocker made a youtube video about these vrms saying they were not good, but he was specifically talking about high overcooks (1400 mhz plus clockspeeds, ln2, that kind of stuff.) so a bunch of people saw this video and took it out of context that it is a poorly made pcb, but the truth is that the Red Devil RX480 is perfectly fine as long as you are running the default bios you wont hurt it (Powercolor has several bios available for this card quiet mode, oc mode and unlocked i believe, the unlocked bios is the one that can cause problems when upping voltage, with stock bios you pretty much cant put enough voltage through it to hurt the gpu or vrm.)

This is actually a known issue with 15.7.1 and 16.2.1, there a quite a few post on AMD's forums about this exact topic.

That board has no vrm sensors or northbridge sensor, most AM3+ do not have either of those things as a matter of fact. What type of clockspeed are you trying to push, i would stay under 4.4 ghz or 4.3 ghz if voltage is not too high, obviously a 4+1 phase board is not meant for overclocking but that one is capable of 140-150w and has a thermal shutdown for the vrms, so it should shut itself off before killing itself, just dont push it too hard. I ran with a 6300 at 4.3ghz for a year or so and it never gave me any problems though voltage was only about 1.3125.

What they do is take another older card and clean it up and make it look new, that has an SLI finger which the 750ti does not even support sli and in the description it says 192 cuda cores whereas the 750 ti is (EDIT) 640 cuda cores, so that is probably a 550 ti they are masquerading as a 750 ti. 

The Red Devil's vrms are slightly lower quality than the reference card, because the have a lower temperature tolerance, so a well known AMD overclocker made a youtube video about these vrms saying they were not good, but he was specifically talking about high overcooks (1400 mhz plus clockspeeds, ln2, that kind of stuff.) so a bunch of people saw this video and took it out of context that it is a poorly made pcb, but the truth is that the Red Devil RX480 is perfectly fine as long as you are running the default bios you wont hurt it (Powercolor has several bios available for this card quiet mode, oc mode and unlocked i believe, the unlocked bios is the one that can cause problems when upping voltage, with stock bios you pretty much cant put enough voltage through it to hurt the gpu or vrm.)

The 370 is an odd card, it was very overpriced for its performance, the GTX 950 easily beats it,at well...everything, that being said the 370 is a rebranded HD 7850 (R7 265) and is easily faster than a 750 ti , the R7 360 however matches the 750 ti quite well, digital foundry did a comparison of the two and the 750 ti got higher fps in 4/8 games and the r7 360 got 4/8 games higher fps and two games were a dead tie.