Vega already in some reviewer hands, we're probably looking at a early May NDA and Computex launch.
>>59989109
THIS CAN'T BE HAPPENING
Maybe I should have started by posting made up benchmarks
no, no more hype, fuck off
>>59989272
But AMD has shown you nothing but architecture details about Vega, the only ones building hype are wccftech.
>>59989109
he's a big guy
>>59989109
I don't remember Ryan being that fat.
>>59989221
I want to donate money to the author of this image
>>59989109
he sure not the subtle type
>>59989358
>#notsubtle
>he's sure not subtle
Were you born as parrot or were you gifted that?
I guess the hype is starting, kingloo himself here.
>mfw vega is gunning for volta and thats why Nvidia is hurridly getting as much out of pascal before it gets annihilated
>>59989643
I know AMD's CPU division could pull a rabbit out of hat with Ryzen, but can their GPU division do the same?
>>59989643
Given how under funded AMD is, Vega is designed to compete with Pascal, and Volta is going to destroy it, especially if Volta has proper async compute and stuff. I don't know why you think they're rushing Volta either, the only rumors I can see about Volta say that it's still on track for 2018, while professional grade Volta for super computers and stuff will come out in late 2017.
>>59989752
just like kabycoffee lake will destroy ryzen right?
>>59989643
Vega has a lot of changes that Nvidia made to Maxwell and Pascal (but mostly Maxwell) that gave them their huge efficiency boost.
ROP L2 cache
Tiled rasterizer
Much faster geometry engines
On top of that this is coupled with much higher clockspeeds, but putting it as Vega competitor is pushing it.
>>59989772
It sure will, you just have to waitâ„¢
>>59989791
Given how well GCN does without those features a new arch with them is going to be very, very powerful (unless of course there are regressions in some areas).
>>59989791
>but putting it as Volta competitor is pushing it.
Ah, forgot to mention, it actually might compete with GV104, but I doubt it's GV102 comparison unless we're talking games that really take advantage of AMD's arch.
Furthermore we really have no idea what Volta is, so comparing it is nonsense as it's not coming out until like April 2018 for consumers.
>>59989387
were you born in USA where guys like him normal?
>>59989819
Oh god please give us Fermi tier housefires again I want to relive the glory days.
>>59989842
It's normal everywhere in 1st world countries, it's just more "normal" in USA because they don't know food that's not junkfood exists.
Also I don't care about his body fat, but his GPU reviews which are still top notch.
>>59989685
Yes, they are making the same changes to GCN NVIDIA did to kekler resulting in Maxwell.
Vega will go above 1080 but fail to get ahead of 1080Ti and all that at a horrible powerdraw!
Then Volta rolls along to annihilate their own top cards once again!
>>59989791
Vega is the first AMD design since Hawaii and Tahiti that wasn't bottlenecked in some nasty way, and we know those two chips are still beasts considering how old they are.
>Hype train
>It's literally a rebrand
???
>>59989935
Can you read?
>>59989928
this
they basically fixed fury's horrible bottleneck and are adding maxwell/pascal efficiency stuff on top
vega is going to be superb
>>59989939
Why bother asking stupid questions?
>>59989924
Then Huang puts hardware scheduler into Volta and we all witness Thermi 2.0: electric boogaloo.
>>59989928
>mfw my friend is still on a 1360MHz 7970
That thing is fucking sick.
>>59989986
It's really weird how GCN1 designs could overclock like 30% but everything after that was pretty uneventful.
>>59989986
>1360MHz
Holy shit
How fast has he got the memory running?
POOR VOLTA
>>59990043
He couldn't hit 7GHz, unfortunately.
>>59990025
Hawaii is a stupidly dense chip (along with fiji) which really limits overclocking. Besides GCN as a whole is meant to ideally clock in the 800-900mhz range for peak effeciency vs performance. Power draw rocketed despite the improvements when they clocked hawaii so high to edge out the titan. That said hawaii is a relatively "balanced" chip and is still a brute force solution to every problem.
>>59990154
I'll never understand why AMD never banked on low power/high perf SKUs, such a waste, the Nano is their only such design.
>>59989772
Nvidia and Intel are different companies, you know. Intel being shitty and only giving 5% performance gains every generation doesn't say anything about Nvidia at all.
>>59990191
AMD's management was pretty retarded before Su
How did AMD manage to make Ryzen very power efficient, but Polaris is a house fire?
>>59990316
Polaris is half-Fiji, it's a very dense design. Its not intended for such clocks.
>>59990316
Polaris is a very efficient card when undervolted.
>>59990316
Ryzen is run at its optimal voltage/clock curve.
Polaris (desktop) is so far above it it's not even funny.
Look at Polaris 11 in the MBP, it's 35W
>>59990316
>>59990336
The nano is a goddamn marvel of the 28nm node - iirc it is the most effecient (performance vs power draw) 28nm card ever built.
>>59990389
>Ryzen is run at its optimal voltage/clock curve.
lolno - ryzen is a fair bit above it for the most part. The 3ghz get close though.
https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/
>>59990407
I meant the lower clocked ones, but for comparisons the 1700X/1800X aren't exactly not power inefficient compared to the Broadwell-E competition.
>>59990431
Intel was never that efficient besides low clocked gimped mobile dies.
>>59990437
Xeon-D's are pretty efficient.
>>59990469
We'll see how efficient it is when Snowy Owl lands.
>>59990437
The desktop chips are the mobile/xeon rejects for the most part. The lower core count xeons (lol monolithic die yields) are very effecient. Ryzen is just all around hardcore as its a design that seemingly only has one true weakness: it doesn't clock really high. Given the effeciency it shows at though at even 3.3ghz on high core coutn chips on the desktop platform is AMD saves the best binnings for the server platform we could be looking at bending Intel over a barrel in a lot of areas.
>>59990495
>seemingly only has one true weakness: it doesn't clock really high
That's not design flaw. Fab it on high power process and you'll get your clocks but lose some efficiency.
>>59990523
Perhaps, but at the end of the day effeciency is still king in chip design. Do more with less.
>>59990523
There's a small chance that AMD never really intended to have a high clocking pipeline in the early stages, so even a HP process might not help all that much.
Small chance as I said, I've seen a 1500X OCd to 4.3GHz a few days ago that neatly questions that chance.
>>59990543
Yea. And they may fab Pinnacle Ridge on 14nm LPU which is another 10% clocks increase.
>>59990555
Nah, Zen is all-around design intended to substitute both construction and cat cores. It's certainly intended for high clocks.
sold my housefire 1070 for this shit
better be good raja
>>59990755
It's basically fucking Maxwell from AMD.
>>59989986
I'm on a gtx660...
>>59990789
You do know that AMD actually managed to get a similar perf/watt with Fiji like Maxwell had without all the fancy shit like tiled raterization, much weaker geometry units and L2 <->ROP association, right?
>>59990820
Yes.
>>59990820
Wasn;t most of that down to power efficiencies in hbm though?
>>59990820
But he's right, the tricks that Vega will feature just brings AMD's architecture up to par with Maxwell
Just announce it AMD.
I am literally one step away from doing crossfire Furys instead.
>>59990873
You're looking at tops some 20-30W
>>59990891
Computex.
>>59989685
well, vega also has infinity fabric, so i'd imagine that they can pull some nice stuff out, even more if paired with ryzen..
and also, isn't hbm2 supposed to be extremely fucking fast?
>>59990925
No, it's wide not fast.
>>59990981
In technical terms it's wide, but in relative terms it's fast.
>>59991015
It's very fast and efficient. I wonder how long it'll take for novideo to adopt it for their consumer cards.
>>59989109
How the fuck do you get Vega from that?
>>59991034
Volta - after they have had basically a year of fucking about with it on the super pascal chips. Fiji only has HBM because AMD wanted to get a headstart on learning the nuances of the technology as HBM solves a problem that GDDR5X merely delays - that ram is becoming a major problem when you want more of it on a gpu. The chips aren't dense enough, they are physically very big and suck down a fair amount of juice. HBM solves problems 1 and 2 and takes a good shot at problem 3.
This sort of 3d stacking isn't new - Nvidia originally backed HBC as a GDDR replacement but JEDEC didn't so the tech sort of fizzled out.
>>59991034
I guess it depends on how you mean fast, but it has a larger transfer rate but it's lower frequency than GDDR5
>>59990755
>1070
>housefire
Maybe if you have one of the single fan ITX cards in a case with shitty airflow.