480 Shilling General

>wait years for Zen
>runs on par with cherrytrail

>>55747644
Hello fellow shiller, Zen will be about 10 percent under intels chip performance. How ever if we at AMD can price Zen just right we maybe able to take a big part of Intels market.

>>55747680
And zen is in q0 stage which means they only just started checking how far the chip can be pushed. After finding the fastest stable clock speed they can decide which chips are failures and which are successful, that is why Intel and and waste so much money on chips

>>55748733
>Intel and and
Does and stand for advanced no drivers?

>>55748760
Lmao you legit turned my mistake into gold

File: GCN vcore-clock.png (37KB, 1171x703px) Image search: [Google]

37KB, 1171x703px

>>55747644
RX 480 is fairly impressive for what it is. AMD's only hiccup is that the internet is full of retard "enthusiasts" who don't know enough to understand simple architectural nuances. Improvement in perf/watt with 4th gen GCN is actually massive, and the general arch improvements alone are astounding. The fact that we have a 36CU 32ROP design anywhere close to a 44/40CU 64ROP design is noteworthy in itself. Couple that with the memory on the RX 480 accounting for 40w~ of its total board power and the picture becomes even more clear.
Compared to the original 32CU Tahiti die, 36CU Ellesmere/Polaris10 is incredible.

The 14nm LPP FinFET node its fabbed on as very impressive characteristics for clocking. Average fmax is higher than any 28nm part for less voltage, and clock scaling is far more linear, with a near perfectly flat clock rate up to 950mhz~ If Polaris 10 was a slightly more complex die with 36CU and 64ROPs it would trounce the R9 390X at lower clocks and lower power.

The Zen core's ability to clock isn't going to come close to Vishera's, the 32nm PD-SOI node despite its fault was intrinsically designed to favor clocks in exchange for high leakage, but it will clock moderately high for almost no increase in voltage.
Summit Ridge will be a decent Haswell-E contender, and undercut the middle weight Broadwell-E.

>>55748733
A RevB ES chip leaked months ago. AMD is pretty far along in tweaking Summit Ridge silicon.

>>55748882
AMD made a huge mistake when they thought relying on specs would improve market share.

Support is the important aspect, they will never learn.

>>55748933
damm tripcoders,

>>55747644
>>/v/

>>55748933
They thought price point and general performance would improve market share, and so far its looking to be the case. The RX 480 cards are selling well.

Everything about the Polaris 10 die was designed around ensuring yields were as high as possible to make profit margins as high as possible. That let them produce a $200~ reference card that on average out performs the R9 290X and GTX 970. Denser logic, increased transistor count, more ROPs, all of these things would have increased die cost. The purpose of Polaris10 was never to be a benchmark king, just to be a decent mid range card at the price point.

>>55748992
but the question is, will they release a RX 490?

Invest
>To spend money, time, or energy on something, especially for some benefit or purpose; used with in.

So what is the benefit/purpose in investing in 480s

>>55748992
We will find out when they release their quarterly won't we dumbass.

Pajeet literally said that specs were important to people raw power for a cheap rice. Whats important to people is having their games work smoothly, even if you do well in most games it isn't enough to stop bad PR, you need to provide better support.

>>55749001
Probably, I'd bet on it being a Vega die.
I don't buy into the rumor of a hypothetical RX 490 being a dual Polaris 10 card. Combining two 110w~ ASICs along with two 8GB of GDDR5 drawing 40w~ each to compete against the 175w~ GTX 1080 is a pointless endeavor. Even more so because tons of high end titles are forgoing Crossfire/SLI support all together. Dual GPU cards were never particularly popular to begin with.

>>55748992
Does it really out perform the 290x? Enough to justify a purchase at euroMSRP?
That's my current card and I recently moved to a much warmer city that puts the blower on when I'm playing some games at 1440p.

>>55749071
maybe they could make a polaris 11 card w HBM Vram on it?

File: perfrel_1920_1080.png (39KB, 500x1130px) Image search: [Google]

39KB, 500x1130px

>>55749084
At 1920x1080 with launch drivers it ends up a few percent above the R9 290X on average.
At 2560x1440 the 290X is 1% faster on average.
At 3840x2160 the 290X is 4% faster on average.

I'm not sure what the price is over in Euroland with VAT and all, but right now I'd buy a RX 480 over a R9 290X.

>>55749137
Polaris 11 is the smaller entry level die.

>>55749153
oh, Polaris 12 then?

>>55749176
For the Polaris designs there are only 10 and 11.
Thats it for their new 14nm 4th gen GCN parts with the Polaris moniker. Their upcoming higher performance designs are Vega 10 and 11.

>>55749025
They already released their quarterly report you fucking tard.

File: Zen_Summit_Ridge_First.jpg (73KB, 369x676px) Image search: [Google]

73KB, 369x676px

>>55748882
Since socket AM4 has finer power delivery than any platform AMD has ever released, even tighter tolerances than intel's sockets, I'm predicting that Summit Ridge has a few tricks up its sleeve.

Boost states varying depending on how many cores are loaded, even gating off a full CCX, and clocking one higher to make full use of available TDP.
Something like:
2.8ghz "base" clock
3ghz nominal clock on all 8 cores
3.2ghz full load on all 8 cores
3.5ghz peak boost on 8 cores
3.7ghz on one CCX
4ghz on one thread with 1 CCX
I'd anticipate a lot of granularity with idle and lower power states as well. They should be able to keep pstate really low in typical every day usage, so average power will be low. Like clocking everything down to 1ghz, with granularity up to 2ghz depending on load. So your average draw while browsing the web may be 10w, peaking up to 25w when loading a heavy page momentarily.
Thats the only logical reason behind such tight VRM tolerances. A chip demanding absurd granularity and efficiency over a wide range of working voltages, and the only use for that range is adaptive clocking.
It appears power gating and thermal management were big considerations in Summit Ridge's core layout.

File: Clipboard01.png (27KB, 673x295px) Image search: [Google]

27KB, 673x295px

>>55750224
How well Zen scales down in clocks will be of great interest as well. Summit Ridge is a 95w TDP chip, and thats only a stone's throw away from the 65w/45w position of parts like intel's Xeon D. If Summit Ridge sips power around the 2.5ghz mark AMD's new X86 arch could be quite competitive vs some of intel's specialized datacenter offerings, particularly if they can accomplish a very aggressive turbo strategy with their CCX model.

What this brings into question though is AMD's now immaterial K12 arch. If K12 is to be even the least bit relevant it'll need to have significantly higher IPC than Zen, or clock significantly higher per watt to make up for the difference. ARM offerings in the datacenter aren't competing well against X86, and one more lack luster hat in the ring won't turn a profit for AMD. Zen might end up stealing the show while K12 never makes it to the main stage.

File: Zen-Architektur+Core+V0.3.2.png (115KB, 640x629px) Image search: [Google]

115KB, 640x629px

Hype

File: Exascale-APU.png (372KB, 1417x909px) Image search: [Google]

372KB, 1417x909px

Reminder that AMD has high performance enterprise APUs coming this year, and will finally realize their decade old ambition of a massive HPC chip next year.

Zen core arch instead of the theorized Bulldozer derivative thought up all those years ago.

>>55747644
If AMD comes out with a card that trades blows with the 1080 (probably a 490 or 490X) at a lower price, I'll buy it. The Fury X, even at XFX's price point, doesn't quite get there.

File: X.jpg (844KB, 5000x2500px) Image search: [Google]

844KB, 5000x2500px

>>55753504
Fury X owner here.

Assuming you get a golden OC card like mine, you'll be blown away by how much more this card can push. While it isn't exactly 1080 performance (its expected of course, this is from a previous gen) its very competitive.

>>55748882
Means shit when you can cook bacon on the card. Why would I buy a card that runs at 82C+ when I can buy one with extra performance that runs below 70C.
I would go for the 480 but there is no way I buy a card that's sitting at 80C from day one, now imagine a hot day + poor case ventilation = dead card in a year or two.

>>55753928
>I don't understand simple physics

The Polaris 10 ASIC is only drawing 110w. The reference cards only run hot because their heat sink is so small, and has no heat pipes.

File: 82416.png (30KB, 650x450px) Image search: [Google]

30KB, 650x450px

>>55754020
Not to mention pic related
80c under load in Furmark isn't out of the ordinary, nor is it a problem. Insinuating in anyway that this operating temp would kill the card is an outright baseless fabrication.

>>55754020
>>55754076
Still no reason to buy it when the 1060 runs at least 10 degrees C cooler and quieter. I really wanted this card to be good but looks like I'm stuck with a 1070 or waiting for Vega.

>>55748760
>drivers
>cpu
are you retarded tripfag?

>>55754110
>all that fallacious logic
The reference cooler being small, and perfectly adequate, has nothing to do with the architecture, or after market cards.

>>55754131
Holy fuck, I'm not going to waste my time. Even Aftermarket cards have higher temperatures compared to the 1060.
1060 pulls less power = less heat to deal with = better temps.

>>55754169
>my fallacious shitposting got called out
>better start overtly shilling for Nvidia since I can't be subtle any more

You sure fell apart quick. You'll never make the big rupees that way.

>>55754110
>Doesnt like rx 480
>Buys 1070
Isnt 1070 2x the price?

>>55748882
Yeah, and its competitor, the 1060, does the same shit, but better, at the same price point.

>>55754908
shill

>>55754908
>telling the truth is shilling

>>55748992
>They thought price point and general performance would improve market share, and so far its looking to be the case.

Nvidias cost the same everywhere except US, because Nvidia cut the prices to negate the import tax.

Also R9 series has seriously dropped and you can have a 390X or a Fury for just a few bucks more for more performance and all the same async/mantle support.

AMD price policy has been a failure and a complete disaster and they aren't selling. The shops seem to be always in stock.

>>55755065
they are all out of stock of everything in europe
and have a hefty mark up, the old stock sell off is the only good side

>>55755289
I just checked 3 major stores, half of 480s are still there, all with 3 star rating complaining about loud fans and shitty cooling.

File: img-thing[2].jpg (19KB, 300x300px) Image search: [Google]

19KB, 300x300px

i want to get the nitro OC edition but i don't know here to find one or when it comes out

>>55753928
Do you not remember the other 480 a few years back?

Aka GeForce 480 gtx

It was a 105°C card under minimal load

Do you know what nvidia fans said?

IT'S MEANT TO RUN THAT HOT SO IT'S OK, IT CAN HANDLE THE HEAT

guess what? The 480 can handle 105C but it runs at 80

which non reference cards are going to be released in the end ? I heard of the sapphire, the asus strix and the gigabyte one. Will there be others ?

>>55755408
Its ok when Nvidia does it.

>>55751927
Can you school me on that FP unit design there. Do they combine to make the suspected 256bit as well? I mean it's just the first time I've seen the fpu laid out in such a way in a diagram, normally it's a single big block with the width on it.

Also is this thing supposed to end up with a shorted pipeline than newer SB+ intels and excavator? Some anon said it based vaguely off jaguar by the look of the die, but I don't know how long that arch is.

>>55755772
128bit FMAC can be addressed together to handle 256bit ops, but its not the same as native 256bit execution, theres always a performance penalty for it. That diagram is just general high level design, you can't really gauge performance or low level design from it. No one knows any of the nuances of the core arch yet. August 23 AMD details it in full at HotChips so we won't know until then.

No idea on pipeline length, though for reference the Jaguar core has a 14 stage pipeline:
http://www.realworldtech.com/jaguar/
A lot of people have speculated that the Zen core would be a larger/wider version of Jaguar going back to when it was originally announced. The Jaguar core actually has amazing IPC for how tiny and narrow it is, so it seems like a good base to work up from, though AMD said Zen is brand new from scratch. Interestingly enough some of the original Bobcat core team moved over to the Zen team, so there might be some similarities they share.
Based on what we've seen from the Summit Ridge die it does look like its cache system is a beefed up version of what Jaguar uses. Jaguar cores come in a quad core cluster with a shared L2 that has an incredibly beefy cache interface connecting them all together. From the looks of things Zen's CCX design uses the same approach, though the L3 is shared while each core has a private L2.

>>55755939
It's a bit worrying if it's competing againt intels 512avx fpu like that then, but am I right in thinking that thing chews up a lot of die space and power?

I'm greatly looking forward to AM4 though, z170 has given me such a ball ache and I'm looking forward to the market being shook up. Wish carrizo am4 hitthe consumer market though.

>>55756074
Broadwell-E doesn't yet support AVX3.1 aka AVX-512. Only their new Xeons are getting that love. Though intel did introduce a new set of rules for AVX execution in this latest i7E line to reduce clocks due to high power consumption and heat output when those datapaths switch on. Some users on Anandtech modeled it and you can actually see the exact moment when the CPU changes state when handling AVX which is pretty neat.

Regardless, Zen isn't going to tackle intel on the FPU front. Intel has way too much of an advantage there, and AMD doesn't have the money necessary to pick up that fight. Maybe successive iterations in the Zen family will make big strides in FPU throughput, but not yet.