Coaxialgamer

Linus actually did an OC guide for X79 way back when the chip was new Basically you turn up the cpu ratio until you get instability , then increase voltage . Rinse and repeat until you reach a voltage or temperature wall . If your cooler can handle it , you'll get to a point ( typically around 4.5-4.6ghz, but may be lower or higher ) where increasing voltage even by fairly significant amounts will not allow you to increase your clocks . at that point you want to find the lowest voltage possible for that frequency and run a stress test for 12-24 hours for validation

The reason faster memory helped was mostly because IF and DRAM shared the same clock domain, so increasing mem speed increased IF speed. Although , if i remember correctly from the article , both these clock domains might be decoupled with ryzen+ , so faster ram might not have the same impact

it's worth remembering that GF 12nm is basically just an updated 14nm , in the vein of intel's 14nm+ or 14nm++. It actually used to be called 14nm+ in roadmaps before TSMC and Samsung pulled some node-let shenanigans . Power scales quadratically with voltage , but it's not necessarily comparable between processes . 12nm might use more power at the same voltage but scale lower , who knows ( and thats why i'm being cautious about blaming GF for the power just yet ).

well , off the top my head , here's what i remember that wasn't already posted : -power consumption is significantly higher : the 2700x uses ~142w under load at stock , 13w more than the 1800x . It also consumes more power at the same voltage , although voltage scaling isn't actually known ( it might go lower in voltage ) . Idle power is lower though. -The cache and IMC improvements bring a ~2.5% increase in performance at the same clock speed -XFR 2 etc allow ryzen+ to sustain boost for longer and on more cores -Memory latency : at ddr4 2666 , latency goes from 85ns on the 1800x to 76ns on the 2700x , a 10% decrease. Bandwith gets a similar boost (44 vs 40GB/s ) Lastly ; while i'm not quite sure what this means , the article seemed to state that many limitations of the original ryzen in terms of speed weren't actually real process limitations. Rather , the article states that these were a result of problems and bugs within the "uncore" , and that ryzen+ can almost be considered a stepping , with most of the performance coming from bug fixes within the cpu. CPC did get things right with OG ryzen though , so i trust them with their results.

I actually bought a physial copy of the magazine last saturday ( i live in france , it's easy to get at the newstand ). I was tempted to post the results myself but was concerned about the legal issues associated with posting parts of a full print magazine . I don't really want to be sued for piracy and distribution of intellectual property ( especially considering the ryzen story is their top story for that issue of the magazine )

Granted , but those things can appear anywhere in the universe , and you have no way of knowing where they appear. I wish for a linus and luke fanfiction

granted but you can never activate the siren . I wish to only get good ideas

X58 boards are usually the most expensive part of an x58 build . You could easily get yourself an x5650 for cheap , which is an overclockable 6 core.

This isn't actually an ASUS board as listed . It's actually an HP board for their Compaq presario CQ1-1000 Series , based on the early single core , Pineview Atoms . This board uses an atom D410. http://www.ascendtech.us/hp-626780-001-intel-atom-motherboard_i_mbhp626780intel.aspx Soldered cpus on boards are fairly common in low power SFF pc's . It's the only package type where you'll get atom's , low power celerons/pentiums and older AMD E series cpus.

The Isa simply does not support it. It's not even a technical limitation. Even if amd wanted to have that full 64 bits, they couldn't without an extension. It's simply by design But it's true that i can't necessarily fault linus if he got his info from anand. Also, Armv8

That's IBM's system/z ISA, it's not x86-64. Limitations are entirely different. Some of those addresses might not be accessible either. The 48/52 bit limitation is one of x64 exclusively

It's fairly hard to get documentation on this but, from intel's own website on x64 assembly https://software.intel.com/en-us/articles/introduction-to-x64-assembly From wikpedia:https://en.m.wikipedia.org/wiki/64-bit_computing The x64 extension has support for a 64 bit flat memory model. This is why addresses are 64 bits long. However, it turns out that the isa itself doesn't support these large adress spaces. Everything past bit 47 of an address is a copy of bit 47, and marked No eXecute by the processor. But even from the perspective of physical memory, it's worth remembering that the word size (64 bit in x64 and 32 bit in x86) actually has no bearing on the address space. As long as you can store those addresses in memory/registers and can express them with instructions, it can be as large or as small as needed. Early 8 bit processors with 16 bit addressing are a great example of that.

it's just plain wrong . the X64 extension only allows for a max address space of 52 bits ( in practice it's 44-48 bits ) . Not 64 . Memory addresses are 64 bits long for compatibility , but the upper bits are all marked NX and not accessible.

yeah , i've actually spread wrong information , especially when it comes to some more complex CS and architecture .They're likely unintentional , and some of them might be because of oversimplification . My "favorite" one is the RAM amounts accessible to processors , where linus claims 32 bit cpus can address 2^32 B of ram and 64 bit cpus can access 2^64B , when in reality no such limitation exists and there is no correlation between processor word size ( "bits" of cpu ) and memory addressability . I mean the pentium 3 could theoretically access 64GB , the 8 bit z80 could access 64K and the 8700k can do ~16TB, so why you keeping this hoax alive linus? the only reason it exists is because MS was segmenting it's OS by disabling PAE outright on consumer versions of an OS Edit: after doing some reading on PAE support in windows , it appears i was only partially right . Although PAE is actually supported by basic versions of 32 bit windows , there are still memory capacity limitations depending on the version (W7 basic only gets 8GB max for example ). PAE also needs to be specifically enabled by the user , so it's mostly blocked behind a skill wall for most users . PAE increases physical addressing capability , but per process it's still limited to 4GB (though i'm not sure if it's a SW or HW limitation). Point still stands

reset doesn't delete your files ; just your programs and settings . I believe windows even confirms that before reset anyway

I was kind of surprised when i saw this tweet a few days ago tbh . It's nice that they be called out for it

in theory a thread like this can go down perfectly fine , but then you remember that in the real world the internet is a thing and so are derailed political threads

this thread will go well no doubt

As long as you got the $$$ it should be totally doable although boards are not yet particularly common https://www.supermicro.com/Aplus/motherboard/EPYC7000/H11DSi.cfm https://www.newegg.com/Product/ProductList.aspx?Submit=ENE&DEPA=0&Order=BESTMATCH&Description=epyc&ignorear=0&N=-1&isNodeId=1

yes

the load isn't solely on the servers , but also on the machines running the game.

userbenchmark is kinda crap tbh. In terms of reliability , anandtech's comparator is better. In terms of pure core/clocks scaling , the stock 3930k should be faster by ~30% due to the 2 extra cores . It loses in single thread, but don't forget it can be overclocked to roughly ~4.6ghz on air ,easily making up that per core difference vs the 3820 and making up most if not all of the delta with the 8700k at stock. you just need to have the cooler and power for it actually completely forgot about the unlocked xeons. OP : if you can find an e5 1680v2 , then you'll have an unlocked 8 core on your hands.

You can actually find 3930k's for a very good price atm . Highest end you can get is a 4960x ( which still cost a pretty penny , as does the 4930k) . Those will be fast 6 cores , and i've seen them do 4.5ghz . Otherwise , you can go for xeon . Honestly , i'd get a cheap 3930k or 4930k and OC , which will be close enough to the newer 6 cores but a lot less expensive because you can keep the board and ram. Also there's ryzen , which is an option . It's worth waiting out for zen+ at this point but just keep your expectations in line : - ryzen + will be ~2.5% faster per clock due to IMC and cache improvements -still slower than intel per thread , but it will reach higher clocks -power consumption will be higher , with the 2700x reaching ~140W ( +20w over the 1800x) -newer XFR and boost tech lets the 2700x reach boost clocks for longer and on more cores than ryzen -