http://www.gamersnexus.net/guides/3 032-vega-56-cost-of-hbm2

File: you_have_to_go_back.jpg (212KB, 443x688px) Image search: [Google]

212KB, 443x688px

>>62087778
>take GaymursNexus word as a fact
>>>/reddit/

>>62087778
Why not just raise the prices? It will still sell out irregardless.

>>62087840

why would someone recreate this image. it's low image quality and poor design are what helped make it funny.

>>62087778
Again, the city of amd

>>62087778
Shut the fuck up you retarded shitposter.
It looks like AMD pulled a Larrabee that works.
FUCKING FINALLY!
(ALMOST) FULLY FUCKING PROGRAMMABLE GPU.

Looks like the fucking thing has no strict shader definitions baked into hardware.
Like holy fuck.
The madmen.
Read the whitepaper carefully.

File: Comb23082017023226.jpg (255KB, 1280x1440px) Image search: [Google]

255KB, 1280x1440px

RIP in peace AMD, we barely knew ye.

FUCKING
PIPELINE
API
HAHAHAHA WHAT THE FUCK

Makes me even more want to buy Navi and Zen+. You get what you paid for rather than extreme markup.

I mean that sounds like sheer insanity, BUT
>general-purpose shaders
That surely sounds like it.
Like how would you emulate per-vertex and per-primitive shader if you merged them.

Please be it.
PLEASE BE IT.
It's the fucking raddest idea in GPUs since the inception of Larrabee, but it WORKS!
Like holy fuck that would be some insane engineering.

>>62087996
Sweet. I waited for that game. With my RX 470.

If they have really generalized non-compute shaders that's some A+ grade sheer mad engineering.
But i doubt.

>>62088043
>tfw still just waiting™ with my 290x

Honestly Vega will probably be great by the time black Friday rolls around with driver updates and solid after market cards.

File: UMA_MACACO_STRIKE_AGAIN.png (98KB, 361x453px) Image search: [Google]

98KB, 361x453px

>>62087778
>finished
>bankrupt

BRAZILIAN MEME

>>62087840
this.
>hurr amd fag
actually i hated gamers nexus long before their amd shitting. i started hating them over stuff they said negative about nvidia and sli.

>>62089069
Gamers nexus has openly hated sli and crossfire since its inception.
what rock have you been under?

>>62087996
AMD is doing great

>>62087778
They'll literally re-release the same Vega with "for Miners" on the box at 1.5x cost and they will lap it up. Mark my words.

>>62088162

Huh? What would be insane?

>>62087996

That's...that's actually great performance for initial drivers from AMD.

>>62087778
>100W for GDDR5, massive footprint, have to cut down GPU to use less power and die space, less bandwidth
>120W for GDDR5X, massive footprint, have to cut down GPU even further to use even less power and die space, similar bandwidth
>20W for HBM, smaller footprint, higher bandwidth
HMMM, I WONDER WHY THEY WENT WITH HBM

>>62088282
whitepaper says they did, for the most part

what I don't get, how would it help my toys? basically engine can be slapped with anything you want to do, API don't matter, but can it be universal? I mean if engine has the pipeline for "vega" or something, for one game can it work for the other?

>>62087778
WHERE THE FUCK IS DIRECT LINK TO WHITEPAPER
FUCKING ARCHIVE IT EVEN
LARRABEE 2.0 WHERE

>>62093045
https://forum.beyond3d.com/threads/amd-vega-hardware-reviews.60246/page-60

just read this, I don't know if AMD driver team can even pull it off in reasonable time

>>62092962
In-engine pipeline manipulations would be ASS. I'd leave them to drivers.
But fuck me that sounds silly.
That's a pretty massive change, to say the least.
Generalizing non-compute shaders is bretty insane.

>>62087996
fuck thats good

waiting for the MSI Vega 56

The joke being that AMD's cards weren't anywhere close to being bottlenecked by memory bandwidth.

AMD just spend a fuckton of money improving a segment of their card's performance that wasn't a limiting factor in card performance

>>62093045
The only whitepaper available is a bretty general description of Vega uarch.
But I need a separate one about primshader.
I really do.
>>62093097
Not Sapphire?
U gay?
>>62093122
That why they spent fuckton of money and time to make a working fucking Larrabee in Vega.

>>62093092
>In-engine pipeline manipulations would be ASS
here is what I see why they did it now
1. next gen consoles coming, it will give them massive gains
2. it is in engine in gpu pipeline = drivers do not matter,OS barely matters, directx you can shove it up MS ass
if game runs on console it will run on PC gpu of same capability 10 times out of 10 with identical performance +- ALU count

this is pretty insane if I understand it correctly

>>62093097
>>62093160
MSI aftermarkets tend to be hit-or-miss when it comes to AMD, last time the RX480 MSI version was superior to Sapphire, with the RX580 the MSI went back to being shit in comparison.

>>62093187
so it makes porting games from console to PC easier when the next gen of consoles come around? but how do we know Sony or Microsoft won't jump ship from AMD? will next gen even be using Zen?

>>62093194
asus was shit for several generations, they finally realized amd needs 2.5 slot cooler
first time in a while it looks good

>>62093203
well, where to will they jump? nvidia? to make 3 steps back one ahead?

>>62093234
nintendo is already on nvidia, so why not
why would Sony or Microsoft stay with AMD really?

>>62093203
AMD have those strategic partnerships sewed up. They are offering performance no one else in the market can match and are willing to sell the silicon at nearly cost, on condition that the consoles buy shit loads of chips to offset the massive discount they are getting. Consoles and other embedded systems kept AMD afloat for a while, fun fact, lots of VLC machines use AMD silicon.

>>62093252
VLT not VLC, my bad.

>>62093246
because "whole package"
they would need to create new unified memory, find CPU supplier, make nvidia hw scheduler under good tdp and price, and also make it all work together AGAIN
nintendo had no choice, well they had choice they could've used arm , no idea why they didn't that's nintendo for you

>>62093187
Well yes, it's about as close as you can get to hardware.
Pipeline-level API.
The fuckers made the first working more or less software defined GPU.
Raja boy sorry for doubting your madness.

>>62093292
So Vega is essentially a beta-test of what will become Navi.

>>62093246
>Sony
>going back to nVidia after PS3 shitfest
>ever
>Microdicks
>going anywhere from AMD after Xenos
>ever
Uh nah.

>>62093311
They explicitly said more or less that several times before.
Now they need to merge pixel and primitive shader just for fun.

>>62093311
That's the "easy" way to make MCM work.
The important question right now, if they can make it automatic for vega driver or not at least for geometry, on b3d it explained as one heck of a clever driver.

File: 1466387468724.gif (433KB, 630x206px) Image search: [Google]

433KB, 630x206px

>>62093344
>Now they need to merge pixel and primitive shader just for fun.
nani

sourcing digitimes
even trump has more real news than digitimes

>>62092546
that is perfomance from NO drivers
the thing is cmaa is msaa that pass via the compute path
games nowdays are becoming more heavy on the compute path mainly because it can free resources for other things

>>62093460
>that is perfomance from NO drivers
WHAT THE FUCK
AMD IS MAKING THE MEME REAL
NO DRIVERS?

>>62093350
The driver is actually retarded.
See parameters - do things.
Its not even supposed to even recognize separate shaders as shaders.

The problem with all this is vega does not seem suitable for low end to mid end gpus. HBM2 seems to price vega out of low end alone, then the heavy compute reliance isn't necessary for low end and will just add heat. It's not well balanced for high end work, it's even worse for low end it seems.

File: aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9SLzgvNjkzMDQ0L29yaWdpbmFsLzAzLUNyZW8tU2NvcmUucG5n.png (37KB, 711x533px) Image search: [Google]

37KB, 711x533px

>>62093350
ofc they found a way to do it automaticly its working on the pro drivers

so, hold on, /g/ told me Vega was an absolute flop, but every once in a while I'm seeing benchmarks with it doing okay, and apparently has no drivers? the fuck, why didn't they pour some extra money on the driver team and rape nvidia on launch instead of have it half assed?

>>62093518
it doesn't, AMD said that prim shaders are not enabled on ALL public drivers
it just demoes what new core can do if not choked

>>62093527
We're seeing the same situation as the r9 200s repeated today.
This would age well if it weren't for the fucking power usage. Where are those undervolting perf/gain experiments?

>>62093527
RTG making a pro debut, pro drivers were more important

>>62093542
this is the pro driver in work on vega fe we know they are enabled on them since DAY 1
there is only one way to cull the shader edges and that is through back face culling something prim shaders do

>>62093360
Something along the lines of seeing pixel as primitive.
Sounds fun, ain't it?

>>62093562
power is fine, it's still under 300w 100% load no breaks
powersaver setting it loses 6% performance and runs at 200w

>>62093568
No, everything does backface culling.
It's just primshader is faster, earlier, and has lower overhead.

>>62087996
This is like the only game where Vega is doing inexplicably well, and only with certain specific AA settings. Just fuck off.

>>62087996
>dirt 4
kys

>>62093246
so they can pretend people actually DO want backwards compatibility now

>>62093589
prim shaders on amd do all 4 of them culling tech together depending on the fov along with discard why you think amd is trying to make them automatic instead of letting devs taking control of them and fucking up amd as they did before on project cars and crysis 2?

>m-muh gayms when what is basically a second coming of GPU Jesus is coming
BUT WHY

>>62093641
I am not talking about primshader. Fixed function culling does the same job, just slower.
And no, it will be really hard for AMD to expose it to devs.
Really.

>>62093313
MS was pretty dumb in the way they handled their partnership with Nvidia for the OG XB, but all the bullshit soured them for good anyway.

>>62093527
amd till 2016 had to feed apple sony and microsoft with chips and all of three of them wanted top binned as always so amd has 2 lines currently

1)regular line that gives them top and low bins
2) a hovis line that all the chips are made equally

problem is nowdays amd has also to feed their new opening on the renderfarms and their workstation cards..

consumers are left with low tier chips thats why they bump the voltage on 1.275v mainly because they want to leave some headroom for OC on reviewers but in reality since vega is fiji basicly making it at 1000 or 1100v is ideal since you will get a massive perfomance upgrade while consuming a lot less power and having less heat in general

>>62093641
>Quick note on primitive shaders from my end: I had a chat with AMD PR a bit ago to clear up the earlier confusion. Primitive shaders are definitely, absolutely, 100% not enabled in any current public drivers.

>>62093676
Yes, Xenos made 360 superb.

>>62093313
>going back to nVidia after PS3 shitfest
t. underage b&
Read about what PS3 was supposed to be and how sony fucked up and got cucked by IBM

>>62093658
no it wont be hard amd chose not to give access atm to any developer

https://mobile.twitter.com/bobvodka

>>62093696
Yeah, it was the right call at the right time. If only Sony had done the same back then. Fucking Nvidia and their "no, unified shader processors are totally not the way forward guize".

>>62093703
And they got cucked by NVIDIA soon thereafter.

>>62093703
Weren't Sony intent on having another CELLLLL unit function as a GPU? Sounded like a pretty goddamn retarded idea.

>>62093758
It was, it was retarded, since using CELL as an actual GPU was as retarded as pretending Larrabee was one.

>>62093718
because it's fucking nightmare to explain
do you understand how it works on hardware? to write for it you need to understand how it works on hardware

>>62093736
>make retarded design 2 cell AYY LMAO
>wow it's shit
>we gotta release ps3 in 2 months
>go to nvidia
>wow it's nvidia's fault sony engineers are retraded
Nvidia had to salvage already broken design.

nvidia hasnt done a single good mobile chip lol
even switch 99% of their problems are related to tegra lol it even bends because of the heat

>>62093788
Well nvidia is only good for software.
Like their biggest achievement in CPU space is housefire in-order core.
But their software is good.

>>62093777
its not a nightmare thats why they released their profiler beforehand

>>62093810
The profiler is for low-level APIs.
This is EVEN LOWER.
You're directly touching the hardware.
>>62093780
By not giving a G80-based design, right?

>>62093810
gpu profiler compared to this thing is like dx9 to vulkan

>>62093808
>>good software
>goyforce experience
>drivers bricking cards
Have you been living under a rock since 2008?

>>62093840
>By not giving a G80-based design
This takes time, retard. The shitty part of the PS3 is the cell compard to 360 CPU. Cell is single thread with 7 meme vector units, 360 is 3c6t.

>>62093840
>By not giving a G80-based design, right?
but the first to unified shader, anon, nvidia the best!

>>62093868
when do you think g80 taped out not to mention finished design stage?

>>62093840
>>62093857
just dont
https://vulkan.lunarg.com/doc/view/1.0.30.0/linux/vkspec.chunked/ch19s02.html

also i dont think you understand the usage of a profiler..

>>62093882
This is usual per-primitive shading.
It's far and away from what AMD calls a "primitive shader".

>>62093864
No, they killed my 8800GTS but their software is generally decent. And CUDA is nice to work with.
It's just recently AMD started pulling ahead with Crimson.

>>62093896
this is literally part of the documentation on the whitepaper kid there is no other refrence
also no its not PRE its literally prim shading

>>62093878
g80 also required good cooling. Cut down design to match the PS3 power envelope probably would be slower than older design which had specialized shaders.

>>62093905
>hey killed my 8800GTS but their software is generally decent
They killed your GPU and yet you say they're still bretty gud. You need to seek medical attention anon, you're not healthy.

>>62093911
Eh no.
Not really.
No.
Fat PS3 had some good cooling.
Good enough to handle cutdown G80.

>>62093918
All companies have shit releases dude. It's not fucking exclusive to nVidia.
Initial Crimson launch was also a fucking shitfest.

>>62093909
Its literally usual per-primitive shading.
You invoke a vertex shader for each primitive in the fucking batch.
It's not what AMD calls "primitive shader".

>>62093061
Why the fuck ain't they pumping everything into driver's? Surely this would benefit commercial and workstation compute massively

>>62094036
They are doing exactly that.
Just doing what's implied is kinda sorta very difficult.

File: 4340718431340.jpg (103KB, 904x1200px) Image search: [Google]

103KB, 904x1200px

>>62089069
>i started hating them over stuff they said negative about nvidia and sli
i don't give a crap on this, """"tech-youtuber"""" are just walking ads nowadays with little to no good or interesting videos. Fuck those fags

>>62094036
if Rys to be trusted it works, sort of kinda in internal builds, but my guess it works only for limited number of cases so what they are doing now is increasing this limited number, it's hard.

>>62094066
Well there's people like buildzoid that provide good PCB analysis.
That's about it.

>>62094074
It's working in their internal testing branch (17.320 according to whitepaper).
And there's no "limited number of cases", it simply supposed to identify pattern and invoke primitive shader because of that.
And there's the tricky part: it needs to do it reliably.
Basically tons of testing.

>>62094095
>And there's the tricky part: it needs to do it reliably.
>Basically tons of testing.
well, yeah, that's what i mean by limited cases
sometimes it works sometimes it doesn't, so they make sure it does
how much will it help current vega,still no clue

>>62094139
Oh, it will help it a lot.
Also I'd expect continuous performance gains for Vega in the future.
Current pipeline is supposed to only cull, but it can do more, according to AMD themselves.

>AMD is still trying to figure out how to expose the feature to developers in a sensible way. Even more so than DX12, I get the impression that it's very guru-y. One of AMD's engineers compared it to doing inline assembly. You have to be able to outsmart the driver (and the driver needs to be taking a less than highly efficient path) to gain anything from manual control.
That pretty much implies pipeline assembly.
Well sorta.
If that's the case we truly have a working fucking Larrabee, this time with actual shaders instead of shitty in-order cores running HUEG vectors.

>>62087917
>irregardless
That's not a word

File: Screenshot_20170822_at_9.55.12_AM.png (71KB, 607x615px) Image search: [Google]

71KB, 607x615px

>>62093596

>>62095684
So vega is thermally limited as opposed to bandwidth limited? Is there a similar graph, but with cmaa/noaa?

File: oiK8Bb9.jpg (906KB, 793x877px) Image search: [Google]

906KB, 793x877px

>>62096266
Vega is presently front end bottlenecked because primitive shaders haven't been enabled in drivers and this means Vega is still currently limited to 4 triangles per clock of culled geometry throughput, which is not enough to keep the rest of its rendering pipeline fed. This is why Vega 56 is 100% as fast as Vega 64 at the same clocks at the moment.

>>62096368
That's the theory, but it's pretty clear from that uv ss, that thermals are playing a role, so unless that's a one off manipulative graph, I don't see how thermals don't matter.

>>62096430
Thermals aren't the reason its benefiting from undevolting, it's just the power limit causing it to throttle. A +50% power offset will get Vega to stop throttling all by itself, but you might as well undervolt as well if you are going to tinker with those settings anyway.

And primitive shaders will help with that in two ways: Firstly because without that bottleneck Vega will perform much better in power saving mode, and secondly because it directly saves the power that was being used to work on primitives that would have been subsequently culled anyway.

>>62096479
I see. It's a shame hbm is so expensive, as it seems vega would be pretty good in the lower end market - 560-570 type range - where it doesn't to push such a high amount of compute to the voltage limit on top of everything else in the pipeline.

>>62096584
stop repeating curlyhair dogmas, nobody actually knows how much HBM costs

>>62096829
I would take what the op person said with a grain of salt, but that guy knows his shit.

I wouldn't be shocked if amd went an at cost route with gaming gpus for a cycle or two, just to get market share back up there or at least try to stay in the game.

>>62096368
>primitive shaders haven't been enabled in drivers
Any idea why?
It sounds like a way to circumvent broken hardware desu.

>>62097157
Oh boy.
Let's just say these are more complicated than you ever think.
You don't simply toggle them on.

File: 1397381923814.jpg (32KB, 396x381px) Image search: [Google]

32KB, 396x381px

>GDDR5 at about $6.50 for an 8Gb module
>8 gigs of DDR4 is like $80

>>62097216
>GDDR5 8Gb module
aren't they 2Gb max?

>>62097382
They are, it's just Steve is as retarded as ever.
And no, a pair of 4-Hi stacks does not cost $150.

>>62097403
I don't bother reading GN, knowing someone will point out inconsistency, but this bad, huh