IT'S OVER INTEL IS FINISHED

IT'S OVER INTEL IS FINISHED

Source:
hardwareluxx.de/index.php/news/hardware/prozessoren/41469-amd-zeigt-doom-auf-ryzen-und-vega-cpu-bei-3-6-ghz.html

Translated source:
wccftech.com/ryzen-ces-2017-3-6-ghz-base-clock-f4-stepping-4-0-ghz/

Other urls found in this thread:

twitter.com/CPCHardware/status/817044837358780416
youtube.com/watch?v=M3chULm4glM
twitter.com/SFWRedditImages

>smeared_turd.jpg

Shut the fuck up you stupid germcuck

eehehe memes

Legit hype, even though I'm on the 1151 platform Zen performing well might drive down prices

At least directly link the tweet instead of posting wccftech.
Linking to currytech should be an instant perma ban here.

twitter.com/CPCHardware/status/817044837358780416

>ayymd goes to 3.9 overclocked
>intel does 4.5 stock

HAHAHAHAHAHA

DELETTTTT

>clocks up to 4ghz
>still can't beat the i5 6600 in game performance

not with 8 cores you dumb shit

Are you just retarded or just desperate at this point?

>on a 3GHz French engineering sample with no drivers and a preproduction motherboard

Uhhh, THIS ISN'T REAL

I PUSH MY FINGERS INTO MY EYES

>Prescott Lake babies will defend Intel anyways

Some people noted that the 3.4ghz engineering sampled used in the New Horizon event was over volted. The chip was pulling 93w under load. Its actually not at all unexpected to see clocks increase a bit more.

I'm not sure if they'll keep pushing for higher clocks before the initial release, but I absolutely expect a 4ghz base clock 8c/16t chip to come out eventually. With base clocks this high theres no way that they're radically far outside of the sweet spot in clocks per volt.
This also means that Zen now slaughters Excavator in energy per clock. The Excavator module takes 16w to hit 3.5ghz, 3.15ghz is 10w.
Zen is hitting 3.6ghz at 10w per core.

the performance of these when underclocked/under volted will be crazy. Probably 65w at 3ghz, probably 2.5ghz or higher at 45w.

3.15ghz base clock to be exact, and it was only about 10% behind the i7 6900k in the games that CanardPC tested.
Blase clock is now over 10% higher, and max turbo is nearly 20% higher.

Kek, Broadwell IPC (outside of AVX) at a much higher clock and lower power, unless Skylake/Kabylake-X is a monster Intel just surrendered a market.

>Kaby Lake does 4.5 ghz stock
>Rayjeet Zeen does 3.9 overclocked

Ayymd BTFO again

>Skylake/Kabylake-X

Moar coarz, that's how Intel will compete with these two.
Which is nice because more cheaper cores is a great prospect.

What is this trying to show?

>Blase clock is now over 10% higher, and max turbo is nearly 20% higher.
>still can't beat the i5 6600

Intel legitimately thought AMD could only threaten their low end chips. Kaby Lake's low end options were actually a panic play by Intel. How else do you explain a hyperthreading Pentiums and unlocked i3s?

But they didn't take similar steps at the high end and it's going to burn them.

8 core ryzen chips aren't competing against mainstream intel chips you retard

You realize these are 8 core chips, correct? Intel doesn't have that.

Not like it matters, with those clocks they'll still slaughter them.

They aren't copeting at all with that 3.9 ghz

>How else do you explain a hyperthreading Pentiums and unlocked i3s?
Intel gave you a nice gift for Christmas :=)

Competing

Vega is Huge! Huge!

>reality denying frogposters ITT

Oh I'm laughing my ass off right now.
Is that the best Little Blue bux can do? Greentext and frogposting?

I hope AMD will remove the PSP in this new microarch, then I have one less reason to hate x86.

>yfw the i5 equivalent Zen will have 4 hyperthreaded cores
>with competitive clocks and IPC

>198W
Is this even safe?

that's a big GPU

Top end Broadwell-E only hits 3.8ghz on a single core, and has a 140w TDP.
Poor little intel. Can't even compete with AMD and Samsung/GloFo's "low power" process.

>According to the report from TechReport who got to thoroughly test the Vega sample with 8 GB VRAM in DOOM, this specific sample is somewhere in between the GTX 1070 and GTX 1080 levels of performance. When set to Nightmare settings, the game reported max frame times of 24.8ms. We also recorded some gameplay of DOOM running on the Vega sample while we were at AMD’s CES event.

>not even released
>already lost to 1 year old Nvidia cards

HAHAHAHAHAHAHAHAHAHA

Each stack is 7.75 mm × 11.87 so you can calculate the die size with a ruler.
Should be close to 500mm2

>hyperthreaded
pls do not use intel marketing term

Nice of you to exclude the part where the card was throttling like your mom.

>burns your house down
>still weaker than a 1070

>implying AMD gives a shit about the $1000 GPU market anymore

>ayymad hbm2
>loses to Nvidia gddr5

Good job amd, you did it again

My lovely shill, Vega is a 12tflop card at 225w

The 1070 is a 6.5tflop card at 150w

Do some math.

>Vega is a bigger FLOP
At least we agree

So it will be about twice as good as a 480? Interesting, that was just what I was looking for. With the inevitable driver updates, this card will be a beast. And now that AMD is going full open source on Linux drivers, I can rest assured this will work great on a wide variety of hardware.

>fury x is a 9tflop card
>gets destroyed by 6.5 tflop gtx 1070

ayymad BTFO again. When will they learn?

for loo

>costing $149-$199 [spoiler]please![/spoiler]

Wonderful

>this denial

Ayymad dissapointed again. Sad!

>980ti 5.6tflop
>gets destroyed by a 3.8tflop 1060

Really makes you think, it's almost like tflops are a measure of the peak performance of a card or something!

higher tflop doesn't automatically translate to higher fps

My point exactly. Ayymad got destroyed again in spite of all these tflops!

who is this semen daemon?

Actually, it quite does if the chip is fully utilized and not half sitting there twiddling its thumbs like the Fury X.

>ENG sample
>is better than a 1070 close to a 1080
>will cost less and consume less

The NCU can do up to 128bit single precision instructions per clock. That means 128 ALUs per NCU.
A 64 NCU die would be 8192 ALUs.
AMD stated Vega is targeting higher clocks.
8192 ALUs at even 800mhz would be 13.1TFLOPs

That slide can't be right, unless the ALUs are miraculously processing two instructions per clock.

4.0GHz is impossible for GloFo, it's simply not gonna happen and you're a stupid shill in denail if you believe it

Intel has the most advanced fabs in history, nothing changes that

Oh look, its the guy who said he'd hang himself by his nuts if GloFo could hit even 3.4ghz.
They're at 3.6ghz base now.

When will you livestream your suicide, NEET?

We have no idea what the NCU constitutes, and even the recent Instinct presentation the Vega has 12tflop fp32, I have a hard time believing they're lying.

just like a GTX 960 wich is a cheap as fuck card for office PC beats a monster of a GTX 780

We heard the same about rx480

Ffs

N o

Oh golly the meltdown ITT is delicious.

tell me, do you even recognize the flag that the dude fucking the girl is wearing

>Muh 4GB version
>Muh only 1400 clockz
>Nvidia btfo
And in reality it is about 3 fps (on average) better than 970

They never include sALU performance in those figures, its always strictly ALUs.
So to pull off 128 instructions per clock in an NCU, they either have to have 128 ALUs, or they can miraculously process 2 instructions per clock.
That still leaves the question of the clocks. FLOPs is a simple calculation. ALUs x frequency x ops per clock.

So we have only a few possibilities

1.) The NCU is 128ALUs. Vega 10 is a 32 NCU design.
2.) The NCU is 64ALUs. Vega 10 is 64 NCU design, but can somehow managed literally double instructions per clock.
3.) The NCU is variable size. Vega 10 is a 64 NCU design, but some are drastically larger, and some are smaller lower power. The touted new abilities are all *up to* figures

How the fuck double precision goes 750GFLOPS if the card goes 12GFLOPS on single This is totally fake.
Each stack of HBM 2 goes to 4GB we know that the specs shows up to 8GB but nor samsung nor Hynix were able to produce a 8GB stack for now.
We see that Vega have two stacks, that means 8 gigs, up to 1024 bit each then 2048 bit bus, maybe they go cheap and is 512-bit each.then its 1024-bit total bus.

kill yourselves Sup Forumsbabbys

>implying Zen will be cheap

Do you know why the core clock is lower than Nvidia? It has massively more CU in a smaller die.

>AMDead will be anything more than mediocre

>muh corez
You CAN'T make this shit up

Just reverse search it.

fixed

So much shitposting. Who even knows what's true anymore?

Currytech is still not a legitimate source but it does make me hopeful for Zen.

Currytech isn't the source, did you even read?
Their original articles are trash but they're occasionally handy for pulling news from other sites I wouldn't even know to look.

In short they're good as a news aggregate, shit for content.

watch the vids released yesterday, iirc in one of them some pajeet talked about the new compute units and how they could do more ops per cycle or something like that

sour grapes

>everyone in the target demographic already has 1060/1070/1080
>nvidia BTFOs amd yet again with volta

shader cores are highly relevant for gpus you retard

youtube.com/watch?v=M3chULm4glM
Apparently there is some voodoo magic going on

Geometry throughput is referring to the front end, nothing inside of the CU/NCU.
They define a CU as 64 ALUs, but do not define NCU at all.
I watched all the videos they released, I read through the entire slide deck. There is no explanation provided.

>3.15 Ghz early ES
vs
>3.9 Ghz
~24% clock speed difference, even if the IPC is same, the clock difference should give 6600 a considerable boost in most games.

Can the NCUs be a part of the variable length CUs with a large scalar core patent AMD got some months ago?

I think thats a possibility.
Its got to be one of those three things I listed.

and whose fault is that?

blame devs all you want, you don't build a bird feeder expecting the squirrels to change their mind about stealing your seeds.

Watch again, 1:38-
>"We did the same thing with cu [over 2x shit per clock] ... can actually process more operations per clock cycle"

Obviously it's AMD's fault, but they did well state that Vega is planning to fix that.
Do you honestly expect AMD to not fix a glaring problem like CU under-utilization at some point?

Oh boi, AMD CPU with Nvidia GPU.

What a time to be alive

>What a time to be alive
So you weren't alive 10 years ago?

Tell me your secrets, undeage.

That's exactly what I want.
AMD might be even 10% behind current Intel in single threaded performance which would be somewhere around my 3570k, I don't care. As long as they offer 8 threats for $100 less than Intel I'm absolutely fucking sold. Many people will understand this pretty damn soon - since 2015 more and more games started using more than 4 cores so they use up to 8 cores/threads (for example GTA 5, Watch Dogs 2, Witcher 3 etc. all of them bottleneck my gpu cuz of i5 like a whore). Multi-threaded performance is where it's at in 2016.

Intel hasn't released high end yet, it's still Broadwell.

Considering the massive improvement from Broadwell to Skylake I'm expecting great things from Skylake-X

>target audience already has 1060/70/80

Nope. Lots of people who have a 970, 390, Fury, 980, 980ti, ect wouldn't mind an upgrade, but understand that the gains in the 1000 series aren't worth the retarded pricing. Plus people locked i with freesync monitors.

Autistic babbling aside what's the verdict on AMD's new lineup? Are they back in the game or is it just like their other "moar cores!" releases?

Absolutely back in the game.
Near Broadwell IPC, clocks are good, the mobos show that manufacturers are taking it seriously, they expect strong sales so they aren't skimping on high end boards.
If an 8c/15t CPU is hitting 3.6ghz/4ghz inside of 95w then the 4c/8t chips will easily be clocked higher to take up that free headroom.

>2017
>8 threads
The next CPU with only 8 threads that AMD will release will be those $60 Athlon style igpu disabled one 4c/8t Zen CCX chips from the Raven Ridge lineup in 2H 2017.