This has genuinely triggered me to write this post. RedGamingTech discussed Intel calling AMD's EPYC totally inadequate for data centers for being basically 4 glued RYZENs 1800X. And then one of all the Intel's propaganda flashed through showed this thing (and still keeps coming back in later videos from time to time):
Concerning the EPYC being "glued", well guess what, with 2 of these "glued" CPUs in dual-socket MB, you have 8 CPUs. Unfortunately unlike Intel there's no support for quad-socket to have 16 CPUs. You can't make 1 big CPU die with many cores cheaply, as well as you can't scale a single-core indefinitely, which Intel learned with Pentium 4 and started attempts like Pentium D. AMD will get definitely better at this as they improve the inter-die Infinity Fabric. Building a datacenter is about being cost-effective, not necessarily reliable or responsive, as in the redundancy has to pay off some time, or as in the data losses should not exceed the profits. In this regard, Intel is acting like Apple, offering their wonderful streamlined ecosystem and reliability sane people don't care about enough to pay all those extras for it. In fact, AMD has always affirmed their CPUs are toys for geeks by not locking the multiplier or making binning of cores or VRAM undoable (flashing BIOS or buying an expensive MB). Now we move to the outrageous underestimetion of Intel's own CPUs, when even an OC'd i5-2500K doesn't really need to be upgraded.
Intel has made a mess with all those *lakes. 1st there was a trip while transitioning to 14nm process, resulting in pure Broadwell being present only in notebooks and a Haswell Refresh (Devil's Canyon) taking place. X99 got Broadwell-E, which wasn't as overclockable as Haswell, on which even an unlocked anniversary Pentium G3258 was released. Then came revolutionary architectural improvements known as Skylake. Some people considered upgrading their Sandy Bridges, Nehalems, Yorkfields, and Wolfdales. Those not early enough were detered by RAM price increase and Ethereum rush. Because Intel realized they can't keep up with their previous tick-tock, they overclocked Skylake, added H.265 acceleration and called it Kaby Lake. The supposed dieshrink Cannon Lake has been taking so long AMD released RYZEN. So they took Kaby Lake, added 2 more cores and called it Coffee Lake for desktops and Kaby Lake-R for laptops and also put the U in "Intel UHD Graphics", even though no improvements over Kaby Lake were made to the iGPU. Meanwhile Cannonlake looks like it will be the next Broadwell, and instead of a new ending for major architecture improvements, Intel announces Ice Lake and Tiger Lake. To make things more complicated, the Atom dies got also some *lake names. After Bay Trail and Braswell (nice portmaneau of Broadwell and Haswell) the *mont architectures (Silvermont, Airmont, Goldmont, Goldmont+ (not Skymont, that became Cannonlake)) spawn Apollo Lake (Goldmont, closer to Skylake), Gemini Lake (Goldmont+, Kabylakeish optimization) and Mercury Lake (Cannonlakeish dieshrink, maybe Goldmont++).
https://en.wikipedia.org/wiki/List_of_Intel_CPU_microarchitectures
INTRODUCTION
Never have I seen so inaccurate chart or table or whatever. There are better "solutions" than just buying an X-series CPU if you want to game or edit in 4K, as well as mix more than 16 tracks in whatever format takes up 32 GB. Intel says only X-series is so potent to do all this stuff so they can get people to pay them their entire salary or whatever remains after taxation, which ancaps know is theft.
OECD ranks Czechia very high in labor taxation, because social and health insurance is counted in. These are mandatory to pay and are defined in %, so they can be counted as just anonther tax. The same with Obamacare - in the eyes of many, it esentially stole a piece of hard-earned salary, telling people what to do with it. Then there's a double standard. For wages below 4x average, which is currently 112928 Kč, or about $5.2k, the limit (in a mathematical sense) of taxation is 48.6 %, but because of "tax sales", it doesn't exceed 47.25 % before it reaches those 113 kKč. Then anything above is taxed independently with a limit rate of 33.8 %, therefore the effective taxation curve approaches that number from above. And this "insurance" thing only applies for employment, therefore doing an actual hard work, not for any business stuff, where there's only 15 % for individuals and 19 % for companies. Therefore the Czech government indirectly encourages people to be lazy and not to work. And the working class hates that, but keeps voting parties in favor of such taxation via some supposed "insurace". There will be no money for rents, why do they care, if they want and have a real estate, they can take reverse mortgage. The Czech Pirate Party, whom I thought to be too computer-literate to be real back in 2009, reduces working and middle class taxation by 1.6 % with willingness to gradually decrease that further. The linked page below may be defunct after elections, so I copied the chart here. The suggested reform adds the most money to those making 112929 Kč (2407 Kč). If you make more than 129595 Kč, it's disadvantageous to vote for them. But the stupid working class once capable of electing Babiš's ANO (which brought us framework for Internet censorship and Chroatian on-line cashdesks) will vote for horrendous overtaxation (~65 %) of the middle class just for being successful at work, which is being pushed by ČSSD (Czech Social Democratic Party). ODS (Citizens' Democratic Party) wants more tax breaks for the rich, but this party is practically dead since it raised VAT percentage to 15 and 21 from 10 and 20, having called itself right-wing.
https://www.pirati.cz/kalkulacka/
So in socialist shitholes such as Czechia, in addition to salaries being miserable despite being almost eaten by Germany, everything is taxed, therefore people just can't afford to pay $999 (Intel tax included) + sales tax, shipping, shipping tax, margins, margins tax, and ecological tax for some CPU that can reportedly satisfy all their needs, when in fact even usually overkill AIO cooling can't satisfy this CPU. So just buy 2 RYZENs or a Threadripper instead (EPYC (Naples) is 4 glued RYZENs (Zeppelin) and Theadripper are 2 of them working, as Intel says, having done this with Pentium Ds) and let Intel bathe in their shit until they come with the 10nm or 7nm process, be it by getting sillicon to cooperate or combining it with germanium or going Indium-Gallium-Arsenide. Good old X99 and Haswell-E and Broadwell-E, not being prevented to run Windows 7, with only few but fairly decent CPUs to choose from. Yet, the god-like nature of i7-6950X is crushed by Ryzen 1800X, which offers 90 % of performance for 30 % of the price. The performace will exceed the i7-6950X as usually Intel-biased compilers make better optimizations and generally accustomate to the brand new Zen architecture, which is a bigger leap than the small Intel incremets, that go like:
SBE IBE HwE BwE SKLX+KBLX CFLX?*** CNLX? ILX TGLX
Wemr SanB IvyB Hasw Brow SKL KBL CFL**CFLR?**CNL IceL TigL ...
tick-tock-tick-tock-*-ticktock-tack-tuck-teck?-tick-tock-tack-tuck-teck-..
|..32nm..||...22nm...||.........14nm.........||...........10nm...........|
* Little tack (Haswell Refresh), mainstream desktop Broadwell cancelled.
** 8th generation "Covfefe Lake": Cofee Lake for desktops, Kaby Lake-R for notebooks, Cannonlake for ultrabooks; leaks suggest 8 core Coffee Lake-R.
*** There's possibility of Coffee Lake-X refinement of Skylake-X (22 cores?), hopefully with soldered IHS this time.
We will run out of vowels for the 10nm process. Before Intel investigates how they will do 7nm or 5nm, they are going to spit so many generations. I hope they'll find better names than *lakes. Now there's so much water I don't think these CPU's even work with air cooling. Tick, tock, tack, tuck, teck, tyck, täck, töck, tück, tëck, tïck, tÿck, tâck, têck, tîck, tôck, tûck, ...
Another thing Intel needs to resolve is the naming after the 9th generation. There are several variants for new i7 mainstream flagships:
i7-10700K, i7-11700K, i7-12700K ... - too long to pronounce
i7-1700K - Here goes the missing 1 from i7-700. Only for 10th generation.
i7-1070K, i7-1170K, i7-1270K, ... - Shift everything to the right. Enough up to 19th generation.
i7-1007K, i7-1017K, i7-1027K, ... - Only 9 CPUs per generation, but enough up to 99 + 19 = 118th generation.
Then there could be a new series scheme, finally convenient:
Quark, Atom, Celeron, Pentium - toys for children and for embedded systems
Core i4 - 4 cores (was i3) for legacy 4-threaded apps
Core i6 - 6 cores (was i5 and i7)
Core i8 - 8 cores (was i9)
Core iX - 10 to 32 cores (was i9)
Xeon Phi - 60 and more Atom-like cores
Only Core series is affected. Hyperthreading is to be denoted with + or HT or P/R/A or something like that, like i8+2700K or i8-2700K+, and/or it's absence can be denoted by - or an unambiguous letter, like i4-2100 or i4-2100-.
There also exists a possibility to devote the digit after generation to refresh numbers and leave only 2 digits for the "model number", as Intel is getting really on with all these *Lake refreshes. Alternatively, if they opt for yearly CPU releases, they could use those 1st 2 digits for the last 2 digits of the year, which isn't Y2.1K compliant on the other hand. The model number could then be based on the number of GHz it runs on all cores and on single core. So something like i6+2345K is a 3rd refresh of the 2nd generation that is 6-core with HT, running at 4 GHz with a 5 GHz single-core turbo, with unlocked multiplier, although there is no reason to lock the multiplier for consumers, because if they fry it, they will have to buy another CPU, which means more profit and more consumption, because some people are apparently that stupid.
http://cpu.userbenchmark.com/Compare/Intel-Core-i7-6950X-vs-AMD-Ryzen-7-1800X/3604vs3916
GAMING
Anything beyond triple 1920x1200 is irrelevant for me, but I guess the load is at the 4K levels, given the extra 120 lines were not optimized for well. How come all the stuff be 1080p when that's not divisible by 16? Nevertheless I believe even 12K gaming is possible even on the almighty G4560 once you have a card with enough VRAM for that. There are only a handful of games requiring 4 actual cores to have smooth framerates, and all I know of are Minecraft, GTA5, Battlefield 1 and other Frostbite games, PUBG and other UE4 games, DayZ/ArmA, AoS (SupCom is too old), and Planetside 2. The resolution isn't dependent on the CPU. GPUs are becoming more and more independent. It's not 1993 when you had to render each frame on the CPU. There are different factors that prevent 12K gaming go mainstream. As of now only the newest DisplayPort can get this going at 60 Hz, and even 1080 Ti struggles with 4K in some AAAs. Still I guess some OpenArena matches will be fine though. The 12K as I learned, means 3 4K monitors, not 11520x6480. /* Then who really 12K on a desktop monitor? Monitors are intended to have 96 DPI and you gotta view that from at least a feet (30.5 cm), preferably 2 (61 cm). */ And there's also a limited space on your table. It's safe to say that anything "better" than a single 40" 4K (110 DPI) has diminishing returns. Barnacules needed a gigantic piece of table for his "12K" triple 50" 4K (88 DPI). If you need scaling, put on your glasses, as you are just wasting pixels, which are to be seen distictly on IBM PCs (pixelart). Even my new 1080p 15.6" notebook came with default scaling of 125 %, so I can say no notebook of this size and below needs 4K as it'd just suck out the battery. And even 1080p in smartphones is more than Retina overkill.
https://designcompaniesranked.com/resources/is-this-retina/
Live streaming depends on your internet more than your CPU (GPU if it has acceleration for that specific coding), unless you're on symmetric 100Mbit fibre. In my place the best Internet connection available is 300/30, just enough for Youtube 4K (22 Mbps), and it's horrendously expensive, though I could unsub landline, throw out radios and TVs and stop paying for them (Britain isn't the only place), and finally stop receiving Russian gas and go full electric. These many cores do come handy for 4K, but you have to consider your audience. The most common desktop resolution is 768p, then 1080p, then 1280x800 and 900p. Inbetween there are some weird mobile things like rotated 360p, 375x667, or rotated 720p. So it's safe to assume there's no major demand for anthing better than 1080p60, which is usually streamed at around 8 Mbps, the speed of my ex-internet. Now I have 22/2, so I can watch some multisampled 4K at 30 FPS if I let it prebuffer a little. You won't appease many people if you record and upload daily videos in 8K, let alone 12K. CGI in movies stays at 4K 24 FPS.
http://gs.statcounter.com/screen-resolution-stats
http://www.rapidtables.com/web/dev/screen-resolution-statistics.htm
Then there's a completely different realm of external capturing, which poses no performance penalty for the PC running the game. Actually it's unprofessional to both record and play things on the same harddrive as well as the same PC and major streamers usually have a double PC setup. i7 to i5 or R5 to R3 will do just fine.
CONTENT CREATION
Similarly to streaming, Youtubers also benefit from double PCs. Render and upload overnight on the less power-hungry one (Ryzen is good for this), and that's how you make a video every single day. Such quantity compromises quality, but that's a different story of enthusiasm and ad revenue.
Here's a PrintScreened Bandit Radio score from Musescore (I didn't want to sign up just to download a damn MIDI) stitched in M$ Paint on supersampled 3072x2048 with 150 % scaling, where I ran into 8K-ish canvas limits:
When I printed it out, I realized I need an A2 printer, as the notes were too small and the DPI of the HP MFC176n so low. I shall mention again that this was done on only 1 1920x1200 monitor.
360° or 2pi video is a gimmick. You won't usually spit 2pi videos each day. There are some people that get motion sickness and their eyes hurt or they feel their head may explode way sooner than others, as it has been with 3D. So nothing for getting massively spread, therefore editing doesn't need to be butter smooth and render can take longer. Same goes for VR headsets and VR gaming - no long sessions -> no long videos -> less computing power needed.
MUSIC CREATION
Did I remind you I made a 4K 6.5 and 15 FPS desktop capture with Hypercam 2 on Windows XP with Core 2 Q9550, 4 GB DDR2 667 MHz RAM, GTX 660 2GB and 24" 1920x1200 monitor, mixing a 42 and 24 channel music module respectively in realtime? And here Intels writes the most outrageous thing of all - that without X-series, we gotta stick to 16 tracks for just "creating" music. You know what has at least 16 tracks? These DOS "apps", as you call them now:
* Scream Tracker 3 (1994, along with 9 Ad-Lib ones, 25 total, 32 possible)
* Farandole Composer (1993, quite obscure)
* Multitracker (1993, 32 channels, KFMF's tracker of choice until FT2)
* Fasttracker 2 (1994, 32 channels)
* Impulsetracker (1996, 64 real channels and 256 virtual ones).
BassMIDI driver does 5000 voices. Intel, the Black MIDI Team may have stoned you for this! The performace is enough to mix 65536 channels, yet you write it can do only 16. Didn't you mean 2^16?
https://www.youtube.com/watch?v=h3DLoq52nUw (Eshm)
https://www.youtube.com/watch?v=0hbjC0Y10pc (OrSaSa)
Also, who the fuck uses such long samples to need 32 GB of RAM to mix more than 16 of them? They could mean entire undo buffers for audio editing, mixing (not the software playback routine), post-processing, mastering, and such things, but this isn't the stage of making the music itself, and special analog consoles exist for that. Also 16 CD-quality 4:20 long tracks take 366 912 000 B of memory. If you need 32F/176.4, it's 2 935 296 000 B, or simply around 3 GB. Nowhere near 32, more like 3.2 GB. To even reach 32 GB, you need to work with around 160. And if you exceed this count even for something as simple as a pop song or EDM track, you should rethink your production process, sure it must be tedious to manage them all. This bullshit has now been successfully debunked. Again an error made when someone is not concentrated.
3D MODELLING AND ANIMATION
Now I am done with the most triggering part of this Intel propaganda, that got me to write this post. In this field I don't have too much to say. But if I used to learn to animate on notebook i3s on an animation and photo editing course in 2011 (where we played CS 1.6 all the afternoon and I saw Minecraft for the 1st time), as well as made a house in SketchUp on an Atom with ION, I can be certain that for the modeling part you don't need to buy X-series CPUs. As for rendering, use GPUs and see the heading about gaming being GPU-driven. RYZEN will serve you better, because AMD was always about permormance for price.
Concerning CADing things for 3D printing it may help you to use fewer round edges and assemble the whole thing only 3D printed, not in the computer. Use the computer to only check if adjacent parts sit together. Also the nVidia Quadro cards appear to be too expensive compared to the capabilities of "consumer" GeForce GPUs, but maybe they are better at actual math like OpenCL or CUDA.
SUMMARY
If you know what you do and don't compensate for your certain organs, you can do everything exclusive to X-series according to the chart above without the X-series. One time even the performance of i9-9980XE will be laughable and this CPU will be sold for fractions of its price as it is now with Core 2 Q9xxx. I have X99 from just pre-RYZEN-hype times and I am planning to upgrade from 5820K to 6950X once it drops significantly below the launch price of 5820K, which was $389. Before that happens, GTX [currentgen]70, 3rd monitor, full RAM occupancy (8x8 GB), M.2 drive, and Blu-Ray drive will come. It's no longer time of buying only the top-tier electronics, but rather going for the cheapest satisfying or price to performance ratio. I wanted to move on from 4 cores too early, so X99 and X79 were the only options aside from the nuclear reactor known as Vishera, of which the 8 cores on 4 modules performed like Intel's 4 threads on 2 cores. AMD witnessed that Intel is keeping cores for Xeons and worked on the best workstation CPUs for the price. And as AMD announced originally unplanned Threadripper (16 cores), Intel bagan panicking (that they struggle with 10nm and Global Foundries are going for 7nm), which resulted in X299 being a messy platform with 4 to 12 and later 2 to 18 cores (modified from Xeons) and 16 to 44 PCI lanes, while 6 cores on LGA 1511 are now available too since Covfefe Lake (Pentiums: old i3 but nerfed, i3: old i5 but nerfed, i5,i7: 2 more cores). The Kaby Lake-X doesn't belong in LGA 2066 and X299 at all. Maybe Coffee Lake-X with even more cores, possibly 34. And AMD counters the 2 surplus cores with Naples (EPYC), now up to 32 cores, which exceeds Intel's 28-core Xeons. Only remaining in Intel's arsenal is the 72-core Xeon Phi with Silvermont cores enriched with 512-bit vectors and 2 more threads, for which there must be specific optimized code. For this generation, Intel is indeed REKT and must stop milking their customer's for money.
https://en.wikipedia.org/wiki/List_of_Intel_Core_i7_microprocessors
https://en.wikipedia.org/wiki/List_of_Intel_Core_i9_microprocessors
