What is Floating-Point Performance?

[@StelznerGaming]

FUN/GEEKY FACT: The new GPUs with over 90 Terraflops are powerful enough to run Cyberdyne's Skynet from the movie Terminator. In the film at one point they say its 90 Terraflops, which when they wrote the script in 1984, was considered mindblowingly insane! But now, we have that much power in our office/bedroom!

[@akimdeadly6247]

in my office? *cries willingly*

[@xwtek3505]

Films are rarely rendered with realtime rendering speed. In films, details are more important than speed. It took Disney 83 days to render the whole Frozen film, not 1.5 hours.

[@obi-wankenobi8023]

@@xwtek3505 this isnt about rendering. Its about the fictional AI in the movie which took as many tflops as modern GPUs have

[@SimoneBellomonte]

@@xwtek3505 What does this have to do with what he said.

[@MattP5000]

which terminator movie?

[@meowgoesthedog_]

2:55 tfw linus is such a pro at dropping stuff, his mere presence infects other people with this superpower

[@user-st5ir8mg3q]

Best comment

[@asdasddas100]

Waiting for Tom Smith to comment

[@EVRLYNMedia]

i want it to happen

[@Envashion]

Tom here but i ain't no smith

[@r.pizzamonkey7379]

I just sent this video to a guy named Tom Smith, no joke. I really hope he does

[@tomspcalt159]

My name is Tom. Does that count?

[@kaungkyaw465]

Toms Pc alt no

[@bensmith7190]

Tom Smith... so close

[@Gurci28]

2:22

[@Gurci28]

2:56

[@iustrenght2762]

Do IPC (instructions per clock) as fast as posible guys. We need that episode

[@KristoKorps]

IuStrenght Yeah, it would be an easy one to make since they mention it so much in other videos.

[@LORENSSIOK]

and floating point for cpu

[@code-dredd]

LORENSSIOK Floating-point ops in CPUs work the same as in GPUs. There may be CPU/GPU differences when it come to _other_ details, but the _representation_ of these numbers and how math is performed with them is the same in both. It's based on the IEEE-754 standard. *EDIT:* It should be noted that the number format described in this video is a _double-precision_ floating point number, which takes 64 bits of space. Graphics pipelines are generally designed to work using _single-precision_ floating point numbers, which only require 32 bits. This is done for performance, since math with 32 bit numbers is faster than on 64 bit numbers.

[@DaveSohan]

That's kinda obvious though 😁

[@fnige]

You can post it onto their fourms, just saying

[@martinenglish6641]

I remember the floating point being a dedicated chip on the mother board and in vary early pc kits, they were up-gradable and some were even flash-able like CMOS BIOS chips. That is still the way on mainframes of about 20 + years back are upgraded.

[@nokkusuu]

you should do a video on error codes

[@reD_Bo0n]

I love my IEEE 754 double precision numbers

[@memeyselfandi]

Thats some quick maths!

[@Funnywargamesman]

VRTX British person located.

[@funbucket09]

Or literally any other country other than USA...... lol. We in AUS say Maths too.

[@Xyzewell]

muffs

[@nairemuh]

meths

[@markoftheland3115]

2 + 2 is 4, minus 1 thats 3

[@sebastiane7556]

What happened to Synergy? I miss it.

[@mcnamaraky]

ROFL. Comment of the day.

[@SupremeMasterr]

The ending pun was about money i wanted to see dollar shave club :(

[@miadhossainrafi810]

Steven Mactavish you made my day. 😂 😂 😂

[@KarrasBastomi]

freshbooks is more appropriate in this manner

[@iPtrck]

Sebastian E I actually bought that synergy and after that synergy 2 went out. I don't even have many PC

[@raptyaxa5771]

2:52 Linus: *exists Stuff: *drops He can drop stuff with just his mind.

[@anishshanbhag6122]

Literally just was coding something and needed info on this topic. Thanks techquickie for your perfect timing!

[@deltaninjadrm]

The Tom Smith that watched this is freaking out by now.

[@mryesnno4301]

I want to know how to calculate the computation of 1 GFLOPS processor would process 1tb of data?? Is it divide 1terabyte into megabytes then times by 1GFLOPS? I just need to know what 1 GFLOPS can process out of the 1tb data per second. Thanks

[@jca-arkitekto-tv]

I'm on my office desk, sipping coffee while watching Linus Videos at my free time until I came to part 2:53.. I spilled my coffee laughing.

[@zelmatrix731]

This has to be the most simplified explanation of Floating Point per second calculations I have ever seen.

[@zackery261]

Floating Lpoint Operations Per Second What a forced acronym

[@TheOJDrinker]

It could be FLOating Point operations per Second, or FLOating point operations Per Second, or FLoating point OPerations per Second... or just FLoating point OPerationS. Could FLOPS per second be redundant like so many other acronyms? The world may never know. (So: FLOP-S, FLO-PS, or FL-OP-S) "Forced acronym" is redundant, an acronym is a type of abbreviation (similar to an an "initialism") that you can pronounce, so pretty much all of them are "forced" unless you're lucky enough to have one that's already pronounceable. (Unlikely.)

[@khaskid]

@@TheOJDrinker brh

[@AurumFaber]

FLoating point Operations Per Second

[@nyquillusdillwad9119]

Thanks for the quickie, Linus!

[@MrRajabMrwass]

two days ago when I was youtubing for floating point I said to myself I hope Linus make a video about it. And you did :D thank you a lot

[@dna129]

For all the Cali peeps... Hellaflops.

[@WeightyLemur]

*HECKA*flops

[@gixxerfixxer4159]

This is a Christian server after all.

[@ranjanbiswas3233]

@@gixxerfixxer4159 And Cali people really give shit to religion 😂😂 They don't even know who is their real father or real mother. 😂😂

[@xxnike0629xx]

@Techquickie Is this related to the whole teraflops talked about with the PlayStation and Xbox?

[@Hobbles_]

Yes and no. In the specific instance of of the XBox One and the PlayStation 4, their GPUs and CPUs have identical architectures (and both are consoles, rather than PCs), and so TFLOPs actually can be used as a rough estimate of how much stronger or slower one is in relation to another. For consoles, unlike PCs, the engines can be made to interface directly with the hardware, rather than through a graphics APIs and drivers. It's not PERFECT still, as optimization still plays a large role in how well a piece of software performs, however, assuming identical levels of optimization, the higher TFLOPs of the PS4 would mean that the PS4 is X% more graphically capable, with some margin for error.

[@tomstech4390]

^This is partly wrong, xbox and ps4 share architectures (cpus and gpu) but otherwise they are just pc's they have an OS, graphics drivers and API's (xbox literally runs a version of windows 10) and the Gflops shouldnt really be compared...but they kinda can be, the Gflops talks about the stream processor throughput however a difference in render outputs..ROPs can have a big factor if they both werent so screwed in terms of cpu horsepower, one can have literally 3x the graphics flops of another but have the same rops which wil limit the maximum graphics performance even before it gets held back by the really low cpu horsepower (jaguar was designed for low end laptops...like atom) so consoles are pc's and the gflops can only be very loosely compared. and real world performance mightbe very different to what gflops would leads you to expect.

[@ABTechMarathi]

Seriously

[@tim3172]

*Triggers violent flashbacks to CompSci 101*

[@docjazz8743]

Did this as I was watching and was able to switch the resolution on this video from 1080 to 4K before the video ended thanks for the tip!

[@jaycraswell5379]

Its still of value to measure the speed of a CPU / GPU in floating point operations. Running high level code or ugh a video game to see how fast it draws frames leaves so much up to the skill (or lack) of the programmer, the compiler and the operating system she or he used. Hardware speed may sometimes give results that are misleading but regardless of someone saying an 8080 ran 0.6 Mega Flops or 0.2 you could see that a CRAY 1 that ran at 200 Mega Flops was a significant improvement for doing floating point. Saying its not a perfect system of measurement is true but if you have that and???? Bash script code versus some tight written assembler code you really have zero information to judge the parts by. The same of course with people running some program written on top of a giant operating system with 7 layers of abstraction versus a program running without any interaction with the rest of the system. Some people get mixed up with Instructions per second and again you can really screw with the numbers by having one measuring how many NOPs run in a second versus another machine moving data from one memory location to the other. MHZ is really not of value because for a time you could get a guesstimate that a 5 MHz CPU (Same Type) was twice as fast as a 2.5 MHz but even then the faster machine might have memory that needed wait states to retain the data that made them a lot closer in speed. Newer machines have more then one core and each core can be running more or less instructions per clock. Even these machines can still measure a system by Mega Flops and have some idea of what your buying. If one core is 100 Mega Flops and you upgrade to a dual core? You probably have a rough doubling of speed. Marketing BS is what makes this a painful debate. One example was our company sold a 16" monitor which was measuring the from bottom left to top right of the viewing area. Other companies suddenly sold "higher resolution" 17" monitors because they decided to measure the distance between the mounting bolts of the exact same display. Take how many flops a CPU/GPU can do with a grain of salt. If you think of it as a rough estimate I think its still of great value. P.S. I want a fat wallet like yours that FLOPS loudly when you drop it.

[@tamaskovacs3034]

omg! I am Tom Smith! HE SAID MY NAME! Haha :D Thx Jon for writing us in the script :)

[@Sulejek]

congrats bro :D

[@_-_-_-_-_-_-_-__-]

Good job man

[@newguy69]

I just wondered how many Tom Smiths gonna freak out :D

[@jjss488]

Nooo, youre Tamas Kovacs

[@usseal922]

JJ SS directly translated from Hungarian, he is Tom Smith

[@matamanthemaster]

Very good vid. Already knew about IEEE floating point from uni but that was really good info about FLOPS. I'll remember that acronym and look at FLOPS now :)

[@kareemmedhat7762]

ERROR at 1:18 the numbers are mixed up

[@J0nDaFr3aK]

3:40 funny how Star Wars IX is just around the corner and I happened to watch this video. Unlimited power!

[@BuzzaB77]

If I may chime in, for anyone making music, floating point performance makes a massive difference to how your DAW handles effects. so if you're trying to decide between an i5 and i7 at similar price points , in almost all audio cases the i7 will do better in audio production.

[@andtpfack8243]

BuzzaB77 yeah doood always jerk off at night ttto my flac files. I also make high quality beats on my phone. But noone buy ;(

[@psssantosh]

Hi Linus!!how are you!?

[@joseruiz6626]

p.s.s. santosh does he ever reply?

[@MrGameteller]

no lol

[@balarab1]

Hey there

[@eesahash3163]

Why would he reply to you? He's obviously arrogant

[@dariusduesentrieb]

hi

[@AlexanderPrussak]

3:11 | i wouldnt say that you can compare the importance of mflops with mp

[@Daehawk]

I remember the mid 90s and my PC in 94. It was a little cheaper because it didn't have a floating point unit in the CPU. It was an SX and not a DX. Though back then it was not a big deal. I still think about FPU and Im not sure what has them now or not lol.

[@abcmole]

They really do need to come up with a better naming system for CPUs and GPUs. When they say a Core I eye matey 42xwing or GeForce Padawan 69-12x edition it literally means NOTHING. I find it extremely confusing. They should have a standard benchmark rating in the name so it's easier to judge what it can do without having to read the small print to find the number of cores and so on. I'm old and remember (for example) the Pentium 2 266 MHz, and you immediately knew the speed compared to other processors. They need to say Intel i7 (standard rating) or Ryzen 7 (standard rating), and I think people would find what they are looking for a LOT easier.which would ultimately lead to increased sales. It would also be advantageous in knowing if a game would run on your system by having something similar as a standard benchmark rating for GPUs.

[@mudgie0205]

I just went over this in CS class

[@TheJuancapella]

You have a Counter-Strike class? tite

[@mudgie0205]

TheJuancapella ahaha

[@oli2.019]

Would be nice if there was a single number that could precisely determine performance or maybe just a benchmark that every product had to go through.

[@navjotsingh2251]

Too many variables do take into account, there won’t just be one number 🥲

[@thatguyalex2835]

I agree with you. An automated benchmark that simulates real world use would be nice. :)

[@ChrisVanMiddelkoop]

you teased me with showing me this on wan show. so happy to finally see it!

[@LittlePenguin1-17]

Linus: 4:00 to 4:05 "Go ahead over to TunnelBear," Me: "I'm better of using Tor Network Bro!" ^_^

[@ayumuchan3541]

Finally they mention Tunnel Bear, it's been a while!

[@H3Vtux]

You should do a video on Fast Inverse Square Root

[@glorytoarstotzka330]

2:28 "can vary quite a *BIT* "

[@Gurci28]

1:01 Appendix D What Every Computer Scientist Should Know About Floating-Point Arithmetic "Floating-point arithmetic is considered an esoteric subject by many people. This is rather surprising because floating-point is ubiquitous in computer systems. Almost every language has a floating-point datatype; computers from PCs to supercomputers have floating-point accelerators; most compilers will be called upon to compile floating-point algorithms from time to time; and virtually every operating system must respond to floating-point exceptions such as overflow." docs.oracle.com/en/

[@Gurci28]

1:20

[@Gurci28]

3:53

[@kngwnk]

TOM SMITH's MIND WAS BLOWN. 2:49

[@superstealth385]

2:54 When you watch too much linus videos - Linus drop tips

[@SSteelification]

here i remember adding floating point co processors to both 68k and x86/x87 to boost performance.

[@tomstech4390]

Yep and people said FX wasnt a real core despite cpu's didnt have FPU units and intel didnt have an IMC for donkeys years after amd but thats easily forgotten cus its intel.

[@Nobody-eg4bi]

Toms Tech FX share floating points like 8 cores FX = 4 cores real and 4 cores virtual because every 2 cores share one floating point.

[@robertlinke2666]

floating point is literraly scientific notation in base 2

[@plamendimitrov5762]

there are co-processors in your processor to make it calculate that - trust me, had a lecture exactly for this in Tuesday

[@jesse-dg8yx]

2:48 who's name is Tom Smith?

[@justincase9z]

cool and good

[@lesterthemolester8293]

Warm and bad

[@redeclipse4ever409]

The Justas Ahh refreshing

[@justincase9z]

are you implying that this video was bad ???? dissapointed

[@HilbertXVI]

The Justas drincc

[@lesterthemolester8293]

dude its a joke

[@xorinzor]

Speaking of flops, go head over to tunnelbear!

[@lonelyrock14]

Suggestion for next Video: What is Dynamic Bass Boost and how it works. :)

[@JasonGroom]

I am a data scientist, and I am very jealous of 110 teraflops

[@nikolalovrencic5026]

All that start with big boom, in year 2005 with creative X-Fi and nvidia and ati start with naming flops and no tools and programs that will run softwares. And later on on forums known as xtremesystems, felas saw this on topic ofc. This may be foture. And then, thank god smart Nvidia create bunch software (no harm for me) that we can run as on graphic cards, to do weather, heal, and there is list of tools and programs using gpu math to do faster (any) jub

[@sayanghosh6996]

looping from 2:54 to 2:57 all day long 😂😂

[@JohnPepp]

I feel old for that was a very big deal when processors came out for the PC in the early 80s.

[@ddz4853]

Please do a video about Aux vs Bluetooth LDAC and Bluetooth on audio transmission.

[@soerenbo]

You can hear Linus being tired of saying the tunnel bear part :D I have never heard him so unmotivated.

[@desertfox1100]

Nice work, Linus! Make more videos like this one, simplifying abstract concepts.

[@WGuitarImprov]

2:53 someone other than Linus dropping something... I THINK NOT! I don't know if you finally hired a witch doctor for an editor or got Corsair to add separate key profiles in Cue2 for Taran "THE MACRO KING" Hemert but I'm onto you. But on a serious note great videos, love the content (all three channels) and remember. When lifting heavy objects always lift with your back in a jerking motion.

[@tressasmith7283]

I don't like how you talk to my brother Tom!

[@ymcha9278]

Can you share your opinion on the low profile mechanical key switches ?

[@TopHatProductions115]

1:19 - Yay signed values!

[@Meowrian]

that was some serious check right there

[@42thgamer80]

This is the content we want!

[@Kooldude183Gaming]

1:48 I only gotta be Linus in math okay

[@quangluuuc5263]

so helpful love it

[@Omlet221]

oh my god linus is so funny hes the greatest comedian ever im literally dying he should do stand up

[@iBlaze1232]

K

[@PCr4zy]

Interesting how nGreedia present their new flag ship as *5x performance* in Double precision, and same week czech student getting awarded for saving electricity and cutting down on operation time by simplifying single and half precision.

[@darius9710]

significand is also known as the Mantissa (I believe)

[@mitradev1]

Man tunnelbear must be having a great business going on

[@Steampunk_Wizard]

So basically FLOPS is just as pointless for comparing gaming performance as clock speed?

[@IamMorpheaus]

Ok Linus .. I am changing over to a GeForce card .. was an AMD fan boy but after watching a lot of your videos and looking at the line ups .. I am making the big plunge .. Keep up the good work ..

[@nathanmead140]

Pause and read @ 0:19 and 1:50

[@Andy-em8xt]

The Exponent is 3 not 4!! 1:21

[@mercury044]

For an idea of what to do for a fast as posible: how voice recognition works

[@PixelBoyMiner]

The one time I'm early for a tech quickie video

[@alsayedjalal]

Everybody do the FLOP!!

[@NoorquackerInd]

1.2*10^3 + 1.6^10^2 = 1.36*10^3

[@EVRLYNMedia]

*to all cuda cores 1000 times per seconds*

[@Wilus0]

2:53 Raja Koduri drop tips

[@mcnamaraky]

A FLoP is the word I would use to describe the case you guys designed in your last LTT video. ZINGGGGGGGGGG

[@Devills_hill]

the last time I flopped something on a counter I got arrested and i'm never allowed to even come near that place again...

[@ROLOGamingOfficial]

What is that video where someone drops something form Radeon?

[@matheusspable]

Anton is a wonderful person. Not a supercomputer.

[@ryanwakebradtelle8682]

Can we get a techquickie on why watching a different video or having a volume control on every possible device can be so annoying and how it could be fixed.

[@ryanwakebradtelle8682]

If that wasn't clear enough sometimes when I switch videos or my Chromecast or Roku restarts the volume is set to maximum, and or one of the devices in the chain passing along the volume percentage is at 10% for some reason so maximum volume is still barely Audible.

[@ryanwakebradtelle8682]

When using my Chromecast on two different televisions there is a third volume control that's on the top of the TV instead of the bottom is extremely weird. Sometimes it goes from no volume to it's so loud at my parents who were sleeping have now come sometimes it goes from no volume to it's so loud at my parents who were sleeping are now yelling at me from upstairs, as well as zero to 100% feels like less than 30% of the TV's actual volume.

[@CodeXBro]

Hey Linus, I could not find the hoodie you're wearing in this video in your store.

[@soullessanger4001]

Just imagine what all the people named "Tom Smith" thought when they were just checking their phones.

[@luceatlux7087]

i finally understand the difference between my Dad's 486sx vs my 486dx back in the day.

[@kinzokushirogane1594]

Can people stop fucking saying the titan V has 110 Tflops? It's not the same kind of benchmark as any of the other GPU's on the market. The actual value is like 12Tflops

[@jonathan643]

Yay! TunnelBear :D

[@kakarroto007]

Linus, I see less commercials on T.V.... that must be why I'm rocking an LTT shirt, and seriously considering signing up for tunnel bear vpn... dang it!

[@Mojobojo]

You missed SIMD, most 64 bit processors now days can do up to 512 (16 floats or 8 doubles) bits of floating point math per instruction.

[@trollichu504]

CPUs vs. GPUs is the TechQuickie which actually got me subscribed Amazing vid btw

[@brandonpoko]

do a video on the machine learning titan V, like how is it different etc

[@restaurantattheendofthegalaxy]

Before FLOPS (usually mainframe) computer performance was expressed in MIPS (millions instructions per second)

[@SlayerOmair]

I'm not tom smith!

[@gauvmusic]

Craft Omair 😂

[@AJD...]

Craft Omair he was talking to me

[@SlayerOmair]

AJD so your name is tom smith?

[@AJD...]

Craft Omair could be. You'll never know

[@SlayerOmair]

AJD OOOOOOOOWOOOOOOOOOOOWOOOOOOOOOOOOOOOOOO!

[@Slimecrazy234]

im confused, if one chip can do 10 floating point operations per second and another, regardless of architecture differences etc... can do 8, wouldnt that mean the first chip is faster? Why is architecture relevant in a measure of final output performance like tflops?

[@CardThrower-rb6eg]

"X=2+Y" the background equations looking very floating point right now

[@mynameiszoro]

no tom smith yet..

[@nicksonomambia382]

tunnel bear must be one of the most resilient advertisers ever...

[@DJDanja1212]

We need a Serato DJ & mix Emergency eternal make up video. Something that help DJs select the right processor and GPU to run these programs efficiently there’s just no information on that stuff for us I think you’d be interested.

[@crystalquark3969]

It´s the same way you may explain sugar, C12H22O11, or sucrose=½glucose+½fructose. The main thing is good frame rate in newer games and happiness 😁, period.

[@unknownentity835]

Thank YOU ! :-)