>>58691375

Does that mean Ryzen can be my waifu?

Other urls found in this thread:

hothardware.com/news/qualcomm-developing-neural-processing-unit-or-npu-to-mimic-human-perception-and-cognition
en.m.wikipedia.org/wiki/Hardware_description_language
jilp.org/cbp2014/paper/DanielJimenez.pdf
theregister.co.uk/2016/08/22/samsung_m1_core/
twitter.com/NSFWRedditVideo

It'll give you brain. 8 cores!

Confirmed flop, buy a 6700k if you're smart.

>An initial batch of Skylake CPUs (6600K and 6700K) was announced for immediate availability during the Gamescom on August 5, 2015

please buy outdated tech goy

>He bought a computer without an artificial intelligence.

>Calls the company Intel
>Has not neural net intelligence in its product

>more buzzwords
>still no price

But.. Will it have an iGPU?

Lol

Buy a intel 8 core with a igpu then.

They're hoping to charge an extra hundred dollars for each buzzword they can generate between now and the product launch.

Nope, but they will release Zen+Vega/Polaris APUs later this year.

Was hoping for gpu passthru without having 2 graphic cards by using Ryzen APU, now i have to wait even longer then for Ryzen in Q1.

Or do some motherboards have a built in Video card or something? Since i see some of the new AM4 motherboards have DVI and HDMI out.

Which wouldn't work if the Ryzen CPU wouldn't have any video capabilities at all.

Or am i thinking about this the wrong way?

4 core Ryzeen with integrated GPU and 4GB of HMB2 L4 cache if the GPU doesn't need it.

>hey guys lets put useless meme shit in here that nobody asked for because what we put out will be shit compared to intel. Our defense force will have something to crow on about and pretend that it actually has some use that they need it desperately for.

I can't wait.

AM4 is also used by older APUs (Bristol Ridge) released last year.

The same motherboards will support APUs.
Consumer APUs will be 4 cores.
8 core chips are for high end desktops and servers, where iGPUs are irrelevant.

Still going to be cheaper than intel

>more buzzwords

Crazy ass adaptive branch prediction that not only predicts branching but other instructions as well. Actively watches what instructions/patterns of instructions are used and learns where to place things into cache for lower latency.

hothardware.com/news/qualcomm-developing-neural-processing-unit-or-npu-to-mimic-human-perception-and-cognition
So an NPU core? It will basically provide optimization for programs compatible with it. It's like a neural net but the nodes are in silicon.

Holy shit I just read through it. So basically the CPU will build of model of decisions made by the software running on it then optimize itself by determining the best routes for it to go through the processor, similar to how people build neural pathways to let them complete a task without thinking like typing. It could mean WAAAAY faster boot times, launching of programs, etc...

Additionally, it will also analyze how software utilizes the CPU cache then prefetch that data into it the next time you run a it.

Its just a branch predictor. Bobcat and Jaguar both have a more simple implementation of it.
Samsung's custom ARM cores utilize the same IP.

Its nothing special. Don't be fooled by marketing spin.

>tfw 8 cores with a decent single core performance + R7 360 tier iGPU

But if they're promising a "true" neural network on every Zen processor, it must be using something like an NPU for it so that it instead of branch prediction it'd be like knowing the branches before the program even launches instead of having to rely on prediction.

NO IT'S A SUPER COMPUTER HIDDEN INSIDE EVERY CHIP JUST LIKE THE CELL STOP LYING INTEL JEWS REEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE

I always wonder how do you wire that into the silicon

from wiki:
>Improved branch prediction using a hashed perceptron system similar to the Bobcat microarchitecture,[15] something that has been compared to a neural network by AMD engineer Mike Clark[16]
it's not a true neural network. it is a learning system so it's better than previous branch predictor implementations. A neural network needs tons of training data to be effective.

>So basically the CPU will build of model of decisions made by the software running on it then optimize itself by determining the best routes for it to go through the processor
that's what branch prediction is for, we had this for decades you fucking shill

en.m.wikipedia.org/wiki/Hardware_description_language

Yes

>25Mhz boost steps

Zen has .25x multiplier steps? That's something we've never seen in an x86 CPU.

>wikipedia as a source

then read the references idiot
jilp.org/cbp2014/paper/DanielJimenez.pdf
theregister.co.uk/2016/08/22/samsung_m1_core/

Intel chips since Merom have had similar ARM processor on the die, which contains the ME hardware backdoor.

I swear I'm not paid by AMD. I'm just reading through the features. I'm not even saying they'll deliver, it's just based on the description.
The difference between this and branch prediction is that it's learning conditional jumps from programs already ran on the processor so that instead of relying on branch prediction the next time it runs the program it will already know the conditional statements of its decisions as you run it. It'd do so using something like an NPU, except instead of programming it yourself it ships with the algorithms baked into the silicon. This is all my conjecture, don't buy Ryzen unless it actually delivers.

A Perceptron style branch predictor is essentially what we can compare AMD's terminology to, and the perceptron style of design has existed (on paper) for about twenty years.

"Neural net on a chip" or any words likewise are just marketing buzz. All modern branch predictors have an ability to "self learn" through the use of the chip's lowest level caches, and as a poster above stated Bobcat and Jaguar used a similar predictor type, just not as advanced.

The real meat of Zen is the XFR functionality. We're all waiting for hands on with high end coolers to show us what the design is really capable of.

Yeah perceptron sounds more accurate. Most likely that.

It's potentially branch prediction that doesn't suck. Which we've needed for a long time.

It's not really prefetching data into cache from from RAM, just keeping often used/about to be need data in cache and moving it into a lower cache level right before it needs to be there. And more importantly placing it as close as possible to the register/instruction about to be used. AKA don't just move it to L1, move it to the left side of L1.

Unless this shit can actually prefectch from RAM and somehow verify the memory addresses are the same as they were last time. Or instead verify the data(or first/last bits of it) is the same as last time/matches a pattern that was used last time.

>It could mean WAAAAY faster boot times
Holy shit this neo-Sup Forums cancer.

>Bobcat and Jaguar used a similar predictor type, just not as advanced.
That's the whole point. Insects have neural nets too, that's the old Bobcat level AI. But this new RYZEN chips has human level AI.

I don't think it's going to be anymore advanced than a dumb computer trying to finish a video game via aimless trial and error.

It would be funny... "6 month old pre taught cpu"
"Educated University of AMD zen processor, at least 1.33 times smarter than the average Zen and with 94.9% knowledge of all programs. Acts as an anti virus too."

Yeah I just can't wait for my CPU to fail 10 million times to find just 1 solution only to fail another 10 million times. When it finally learns the route... An update comes along adding 1 tiny new thing and the moronic AI can't adapt and goes back to 0.

Knowledge will never beat intelligence and sentient subhumans at AMD know this. Sapience is what sets us apart from the morons.

The RYZEN cpu literally gets better with time. It learns and get smarter. You'll never have to upgrade an RYZEN cpu, you just teach it more advanced algorithms.

AMD FineWine

wanna know how I know you're fat?

So are we still thinking February 27th launch or has something suggested it's been pushed back even further?

I can't even afford an upgrade right now I'm just sick of all this teasing and hype.

>mfw your actually fatter than me

That's post of the new power IP that's hyping everyone

they said this in their first release of zen information last month..

Late february should be lanch. You'll probably start seeing serious reviews and benchmarks in 2-3 weeks time

or in the wrong hands it just gets dumber and slower