No thread about the leaked VEGA slides yet?
http://videocardz.com/65406/exclusive-amd-vega-presentation
They're using fucking NVRAM, people.
I'll post the interesting slides.
>>58335489
>>58335502
Post more
>>58335489
>No thread
>>58334258
>>58335489
not really worth it since the presentation is in two hours.
As expected, it's a compute monster, so
25TFLOP fp16
12TFLOP fp32
6TFLOP fp64
>128bit 32bit ops per clock
>per CU
fucking wot m8
CUs are drastically larger in Vega.
>No prices
>No release dates
>No benchmarks
It's fucking nothing!
For muhgayms this is the most relevant in goyworks titles.
1/2
>>58335550
2/2
Look at the second to last paragraph for the frontend performance.
>>58335502
not sure why I did this.
>>58335513
Another 3.5 in the works for GDDR memory pools?
>>58335535
>12TFLOP fp32
>compute monster
Isn't that basically the same as GP102?
1/2
And this is probably the most interesting image as it directly shows how efficient the arch is.
>>58335580
>>58335579
gp102 can't do fp64 for shit, nor fp16
>>58335579
It's like 11Tflops
but 64bit is 350flops 16bit is 170flops (At least if you trust wiki)
>>58335595
>that floor
>that upper left
Holy shit it's almost like it's not rendered at all.
>>58335627
that is not how it works, every cluster of polygons you see with very little space is wasted polygons (aka Tessellation)
There is really no reason for the floor to be more than 1000tris.
>>58335566
You should visit a doctor or something
HurfdUDRRRR
Nvidia GP100 is 'only' 5.3TFLOP(fp64), and that's the Mezzanine version, the PCIe one is .8TFLOP slower.
>>58335762
That would be much more impressive if Vega's die size is around 400mm2
Time till the death of Nvidia
>>58335489
>NVRAM
what does this marketing term mean? it had better fucking mean HBM2
>>58335832
NVRAM and HBM are entirely different things.
>>58335832
Nope, no HBM
I guess they can't do it on such a large scale in time.
anyone know release date?
>>58335832
it's RAM that doesn't loose it's content when power is off, it's like, very well known thing.
>>58335852
some time this year
>>58335546
Cause you a fucking moron that happy with catalogs reader like Linus rather than knowing each part. Fuck off illiterate
>>58335848
no HBM? no buy!
>>58335848
You're in for a big surprise, FUD poster.
>>58335864
stop posting before it's too late.
>>58335848
What? Vega has 16GB of HMB2, the Instinct cards 32GB
>>58335832
>>58335832
It is non-volatile RAM, memory that persists state when unpowered. MM refers to the main memory here.
>>58335489
>NVRAM
they are bringing the ssg tech to the consumers
they are still using hbm2 its literally the poster boy of hbm2
isn't NVRAM drastically slower...
>>58335489
What the heck are they gonna store on the NVRAM that would persist through poweroff?
>>58335856
wow it's fucking nothing
>>58335579
the nvidia consumer cards are using the cuda at 1/128 rate they cant compute at all
>>58335502
GCN already had a high IPC compared to Nvidia, they should really look at getting the clocks up like Nvidia has, even lowering the IPC by a little for much higher clocks.
The shills are getting desperate.
Its a sign that AMD has a killer line up coming. All the fanboys and $0.02 per post Chinks are coming out of the woodwork to try kill the buzz.
Vega cards use HBM.
"When NDA ends, you will also see the chip itself with two HBM2 stacks (so 16GB HBM2 memory). It’s hard to say how big it is exactly, but it’s definitely comparable to Hawaii. Enjoy."
They'll show off the die and package during the presentation.
>>58335895
I really doubt they'll gonna put M2 slots on consumer Vega cards..
>>58335489
yeah yeah blah blah
let's get to the most important part: the price tag
>>58335929
Nobody cares, poorshit, fuck off.
>>58335923
one can wish
>>58335931
>Nobody cares
Retard
>>58335923
>>58335901
Well, there was this slide so maybe game textures, which would actually do very well on loading times and performance and would be much faster than storage media > ram > VRAM
>>58335931
good goy yes take more loans don't worry you can pay it back
>>58335949
>>58335942
Shut up, niggers.
I didn't come here to talk about something paltry like its price, I came here to talk about the GPU.
>>58335828
Where is that countdown from?
>>58335513
Hahaha, LOL, 3.5 confirmed
>>58335918
Whatever you say, Lisa
>>58335959
Never should've come here
>>58335489
Would you please stop leaking our presentation online and delete the images you've posted? Otherwise we will contact the site administrator and proper authorities.
>>58335979
http://ve.ga
>>58335993
no https no click
>>58335989
Sure.
>>58335993
>The connection has timed out
I bet nvidia shills are DDoSing it.
>>58336000
its a countdown clock lol i really doubt its the nsa trying to spy your chrome history full of midget porn
>>58336021
wtf how did you know my history? I thought I was secure
>>58336029
Of c you are, it was just a lucky guess!
>>58335979
>>58335993
learn2read
NVRAM is off package like DRAM >>58335489 while the MM(HBM2) is part of the interposer.
So it's gonna either be.
1- Vega with M2 slots like their SSG (most likely, optane combination when!?)
2- Motherboards with NVDIMMs (not gonna see it in consumer space, ever)
What use is there? Well, for regular joes like us, there's not that much use unless I'm missing something.
>>58335502
>AMD CPU have lower IPC and lose to Intel
>AMD GPU have higher IPC but lose to Nvidia, at least for the last 2 gens
Really makes you think
>>58336093
WTF?!?!?!?
>>58336093
>>AMD CPU have lower IPC and lose to Intel
Because their CPU have not much higher clockspeed, performance is a function of clock x IPC
>>AMD GPU have higher IPC but lose to Nvidia, at least for the last 2 gens
While Nvidia does have lower IPC, their clockspeed is far, far higher.
But this isn't all fun and games, something wide like a GPU nearing 2GHz clocks have nowhere to go performance-wise, that's not even mentioning the clock-syncing and signal quality issues those clocks entail, Pascal was a one time thing, you can be sure Nvidia next gen will be at a much lower clockspeed.
Will there be a stream?
The preview website is down
>>58336167
ve.ga just loaded for me.
49 mins to go before it starts.
>>58336161
we arent on the 80s anymore...if it was just the clockspeed then amd would have oblitare intel since they dont have any cold bugs and they can push 5-6 ghz on all cores....modern cpus are just risc with x86 emulators basicly..
>>58336167
just search amd on twitch
>no card launch
you had one job amd
>>58335489
any livestream link?
>>58336232
ve.ga
>>58336184
it's combination of both, let's say you have 2IPC and 1Ghz your total IPC is 2000
and now let's compare it to 1IPC and 2Ghz your total IPC is 2000
You see how it works?
>>58336270
its alot more complicated because now we have branch predictions we have commands based on the load of the ssm cache itself
honestly i can name 10 things now that can increase and "increase" the ipc and without even mention the speed its self
at least we know that the cringe fest award for 2017 goes to nvidia
amd we try today to beat them but i seriously doubt that
unless they have another monkey trying to play battlefront again
>tfw I don't understand any of this shit
Fuck you guys and your superior knowledge. WHAT DOES IT ALL MEAN!?
>>58336371
It means the Cubs win the world series.
>>58336340
>at least we know that the cringe fest award for 2017 goes to nvidia
What did they do? I missed out on the stream.
Was it just CARS CARS CARS?
>>58336381
No, it was AI and botnets and paying Nvidia to play stream games with latency.
When is the VEGA stream?
>>58336232while true; do; mpv "https://wwwDOTtwitchDOTtv/amd"; sleep 1m; done
>>58336405
23 mins
>>58336381
>Was it just CARS CARS CARS?
Pretty much that and Jen-Hsun being an autistic fuck the entire time. They had a section where they brought on some guy from Bioware to talk about gaymens, and it was pork shoulders levels of cringe.
>>58336405
Go to ve.ga
>>58335606
>fp64
>>58335535
>6TFLOP fp64
pretty sure that the first Vega is not guaranteed to be 6 TFLOPS of fp64.
like GCN, NCU seems to be implementation dependent, where they can cut out the functionality to save transistors on chips they don't intend to dual-purpose for enterprise.
>>58336433
and the funny part is 90% of the "deals" he said it was deals we knew since feb 2016...
and the good part was when they introduced us the bb8...
for you to disable the autopilot you NEED to SAY "disengage autopilot"
now imagine if you are scottish or some redneck texan with an accent of a crocodile.
>>58336548
There is nothing cut. Just like with their CPUs, AMD produces one lot of dies for consumer and enterprise hardware.
Double precision rate on all consumer cards is limited in firmware only.
I thought amd was going to go balls deep with hbm?? What went wrong?
>>58336371
don't matter iz jus magnets is sheit bro, just buy Nvidia ;)
>>58336564
Can you not read the thread?
Why didn't you bother to actually check the source posted in the OP?
Why are you so dumb?
what the fuck is going on in amd stream on twitch?
>>58335535
> 2x fp16 throughput
yawn. already long-known.
> 4x int8
this is new. they definitely are going after Nvidia's lunch on Deep Learning shit with this, since there's really virtually no use for this in graphics.
>>58336551
>now imagine if you are scottish or some redneck texan with an accent of a crocodile.
So you mean you have to speak like a beta nu-male faggot in order for it to understand you then? Nvidia confirmed degenerates.
NDA ends in 3 minutes by the way.
ve.ga ddosed?
>>58336590
the fact that you need to speak at all is just wrong in so many levels..imagine if this shit doesnt detect an accident(which is nvidia based so..)you will HAVE TO SPEAK in order for it to give you control
>>58336615
Your first countdown?
>>58336622
they usually end outside my timezone
>>58336548
AMD doesn't have specialized FP64 units, Nvidia does.
FP64 is part of their CU, just disabled for power purposes
so what now ? where is the presentation ?
Any other source then ve.ga?
https://www.youtube.com/watch?v=Ku5OMWYVKSs
well
VEGA
>>58336412
>i can't stream the event
>plays some chink rpg
amd fucked up
>>58336690
That's a big chip
>>58336697
Around 400mm2~
>>58336697
4Poo
>>58336697
Compared to Fiji and GP100? It's smalltime
>>58336697
You see small 2 black boxes under the chip? That is HBM2.
It'll be mighty impressive if it's 6TFLOP fp64 on a 400mm chip
website doesn't load, rip lifestream anyone got a mirror?
>>58336752
There's no livestream just few pre-recorded videos on their youtube channel.
>>58336752
no stream, no presentation
chip looks small though, that's promising
>>58336768
that is, disappointing.
>>58336774
>>58336775
It's 6 AM in Las Vegas. Of course it's not a presentation
>>58336775
not really, instead of cringy stream with spaghetti we got somewhat decent videos about the arch
>>58336800
nothing really detailed, website is down, let's hope it will have more info there.
>>58336640
All implementations of GCN can do fp64, but the default rate is something like 1/8th or 1/16th.
Higher throughputs require fractionally more silicon to track more off the parallel pieces of an add/multiply in flight, to the point that only Hawaii so far has actually had 1/2 rate, and only Tahiti had 1/4 rate.
>>58336810
http://www.guru3d.com/articles-pages/editorial-amd-vega-gpu-architecture-preview,1.html
for details
>>58336800
no useful information for us consumer mongos though.
no price point.
no release date.
NOTHING.
>>58336830
>AMD made it clear on the media-event that the enterprise market will be served first.
here goes my hope
>>58336830
>do not expect VEGA to launch for consumers in the form of a Radeon RX 490 anytime sooner than Q2 of 2017
weeeeeew lads
>>58336841
enterprise first, for other details watch instincts presentation
for me it's a good thing, I won't have money till q2 anyway
No wonder Nvidia didn't unveil the 1080 Ti. They knew AMD had nothing
Nvidia wins yet again boys
>>58336942
neither nvidia had nothing lol we barely saw the ti on the shipment listings of hong kong meaning that the card is far from being done yet
straight up went from AMD youtube to ebay to look for used gtx 1070's.
THIRTY SEVEN FRAMES PER SECOND
>>58335949
you already got your orange elected, fuck off with your memes
>>58336830
>The photo shows a working VEGA setup in a demo room running Doom at ultra quality settings in an Ultra HD resolution, this was a demo room at the event. The photo comes from Golem.de and as explained is Doom running at 4K Ultra. The card is outing an FPS of 68 fps, with the "Vulkan" API enabled and what probably is a very early unoptimized engineering sample. That roughly is indicative for the performance of a GeForce GTX 1080.
So the most interesting shit:
>enterprise market first
>8GB HBM2 (pic related - DOOM shows it)
>GTX 1080 performance in DOOM Vulkan
price fucking when
all this bullshit is 100% meaningless until P R I C E
>no price
>no benchmark
>only marketing shit
Well played AMD, well played.
>>58337016
This
I am gonna get a GTX1070 now, even though I won't have G-Sync because I've got a freesync monitor.
But without any cards on the level of a GTX1070, it's a non-argument to buy any AMD cards.
>>58337016
this, i will buy now a Nvidia GTX 1070™ and a Nvidia Shield™ to enjoy all my games how it's meant to be, wherever i want
>>58337127
>device id 687f:c1
There was an earlier leaked benchmark that showed AOTS performance, had the same device id.
If that is true, then its roughly ~30% weaker than GTX 1080. I suspect it would be competitor to 1070, but not more.
>>58338733
>If that is true, then its roughly ~30% weaker than GTX 1080
That's extremely unlikely given its performance in Battlefront
>>58335489
>leaked
Are you fucking retarded?
http://www.anandtech.com/show/11002/the-amd-vega-gpu-architecture-teaser/3
>>58339105
>NDA ended 3 hours ago
>hurr it's not a leak when answering to a 6 hours post
>>58335555
>>58335566
Digits confirm AMD year.