Topics
Latest
AI
Amazon
Image Credits:Flow Computing
Apps
Biotech & Health
Climate
Image Credits:Flow Computing
Cloud Computing
Commerce
Crypto
Chart (in log, note) showing improvements in an FPGA PPU-enhanced chip versus unmodified Intel chips. Increasing the number of PPU cores continually improves performance.Image Credits:Flow Computing
Enterprise
EVs
Fintech
Flow’s founding team, from left: Jussi Roivainen, Martti Forsell and Timo Valtonen.Image Credits:Flow Computing
Fundraising
Gadgets
bet on
Government & Policy
Hardware
layoff
Media & Entertainment
Meta
Microsoft
secrecy
Robotics
security department
Social
quad
startup
TikTok
Transportation
speculation
More from TechCrunch
upshot
Startup Battlefield
StrictlyVC
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
A Finnish startup calledFlow Computingis take a shit one of the wildest claim ever heard in Si applied science : by impart its proprietary companion chip , any central processor can instantly double its performance , increasing to as much as 100x with software tweaks .
If it crop , it could help the manufacture keep up with the unsatiable compute demand of AI makers .
Flow is a spinout ofVTT , a Finland country - endorse research formation that ’s a bite like a national lab . The chip engineering it ’s commercializing , which it has branded the Parallel Processing Unit , is the consequence of research performed at that research laboratory ( though VTT is an investor , the IP is owned by Flow ) .
The claim , Flow is first to admit , is laughable on its face . You ca n’t just magically squelch superfluous performance out of CPUs across architectures and codification radical . If so , Intel or AMD or whoever would have done it days ago .
But Flow has been play on something thathasbeen theoretically possible — it ’s just that no one has been capable to deplumate it off .
Central Processing Units have come a long way of life since the early 24-hour interval of vacuum tube-shaped structure and biff cards , but in some fundamental shipway they ’re still the same . Their primary restriction is that as in series rather than parallel mainframe , they can only do one thing at a prison term . Of course , they switch that affair a billion times a second across multiple cores and pathways — but these are all fashion of accommodating the individual - lane nature of the CPU . ( A GPU , in demarcation , does many related calculations at once but is specialise in sure operation . )
“ The C.P.U. is the weakest liaison in computing , ” say Flow Colorado - laminitis and CEO Timo Valtonen . “ It ’s not up to its project , and this will require to interchange . ”
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
central processing unit have gotten very fast , but even with nanosecond - level responsiveness , there ’s a tremendous amount of waste in how instructions are stock out simply because of the basic limitation that one task want to eat up before the next one starts . ( I ’m simplify here , not being a chip engineer myself . )
What Flow claims to have done is remove this limit , turn the mainframe from a one - lane street into a multi - lane highway . The processor is still limited to doing one task at a sentence , but Flow ’s PPU , as they call it , essentially performs nanosecond - scale dealings management on - dice to move chore into and out of the processor faster than has previously been possible .
mean of the CPU as a chef make for in a kitchen . The chef can only work so fast , but what if that someone had a superhuman assistant swapping knives and tools in and out of the chef ’s hands , clearing the prepared food and putting in fresh element , get rid of all task that are n’t literal chef stuff ? The chef still only has two hand , but now the chef can work 10 prison term as fast .
It ’s not a pure analogy , but it gives you an idea of what ’s befall here , at least according to Flow ’s internal tests and demos with the industry ( and they are talking with everyone ) . The PPU does n’t increase the clock oftenness or crusade the system in other way that would lead to extra high temperature or power ; in other Son , the chef is not being asked to chop up twice as fast . It just more efficiently uses the CPU cycles that are already taking berth .
This eccentric of affair is n’t steel new , allege Valtonen . “ This has been analyse and talk about in eminent - horizontal surface academe . you could already do parallelization , but it breaks legacy codification , and then it ’s useless . ”
So it could be done . It just could n’t be done without rewrite all the code in the world from the ground up , which kind of make it a non - dispatcher . A similar problem was solved byanother Nordic compute company , ZeroPoint , which achieved high levels of memory compression while keeping datum transparentness with the rest of the system .
Flow ’s big achievement , in other wrangle , is n’t gamy - speed traffic management , but rather doing it without having to modify any code on any CPU or architecture that it has test . It sounds kind of disturbed to say that arbitrary code can be executed double as fast on any microchip with no adjustment beyond integrate the PPU with the die .
Therein lies the chief challenge to menstruate ’s succeeder as a business : Unlike a software product , Flow ’s technical school penury to be include at the chip - design level , meaning it does n’t mould retroactively , and the first potato chip with a PPU would necessarily be quite a way down the road . Flow has shown that the technical school works in FPGA - based trial setup , but chipmakers would have to commit quite a spate of resources to see the addition in question .
The scale of those gain , and the fact that central processing unit improvement have been reiterative and fractional over the last few year , may well have those chipmakers knocking on Flow ’s threshold rather desperately , though . If you may really repeat your performance in one generation with one layout change , that ’s a no - brainer .
Further performance gains derive from refactoring and recompiling software to work good with the PPU - CPU combo . Flow enjoin it has see gain up to 100x with code that ’s been modified ( though not necessarily to the full rewrite ) to take advantage of its technology . The companionship is working on offering recompilation instrument to make this undertaking bare for software package Godhead who want to optimize for Flow - enabled chips .
Analyst Kevin Krewell fromTirias Research , who was brief on Flow ’s tech and referred to as an outdoor linear perspective on these matter , was more worried about industry uptake than the fundamentals .
He pointed out , quite rightly , that AI quickening is the biggest market the right way now , something that can be targeted for with special atomic number 14 like Nvidia ’s popular H100 . Though a PPU - accelerated CPU would lead to gains across the plank , chipmakers might not want to rock the sauceboat too heavily . And there ’s plainly the question of whether those company are willing to invest significant resourcefulness into a for the most part unproved technology when they likely have a five - year program that would be upset by that option .
Will Flow ’s technical school become a must - have component for every chipmaker out there , catapult it to fortune and prominence ? Or will penny - cabbage chipmakers resolve to quell the grade and keep extracting rent from the steady grow compute market ? Probably somewhere in between — but it is telling that , even if Flow has achieve a major engineering feat here , like all startup , the hereafter of the company depends on its customers .
Flow is just now emerging from stealth , with € 4 million ( about $ 4.3 million ) in pre - seed funding led by Butterfly Ventures , with participation from FOV Ventures , Sarsia , Stephen Industries , Superhero Capital and Business Finland .
