Ok /g/, Vega was a bit of a disappointment. Not to keep the

>>62125597
I should clarify, I know Raja has designed GPUs for AMD in the past. And historically, they've all been good. But I know Navi will be the first one since he came back that will be all his handy work.

>>62125597
you're just a retarded parrot, Vega isn't bad at the right price, a 980ti isn't bad at the right price, etc.

>>62125624
>Vega isn't bad
Be objective

Navi may be a massive improvement but by the time it is actually delivered Nvidia will be waiting with Volta's replacement

>>62125624
Vega is literally DOA, only the fanboys will buy this meme card

>inb4 undervoltage

>>62125649
I feel like Nvidia will wait until closer to Navis release to release Volta. They don't seem like they are in any rush.

>>62125624
Are you out of your mind? Power draw is off the charts, performance is subpar for the price.

Look, it could have been a good card, but they pushed it too hard and charged to much money. HBM cost them big on this release.

And to be clear, I'm an AMD fan. I've been rooting for them since the 6970, oh which I owned two.

But i can still be objective about the shit they've been putting out with their GPU department. The R290X was the last card they made worth buying. Fury was shit, half the shaders never got touched, miners ruined the rx480, and rx580, (They would be good cards at their MSRP, though). and Vega is a hot loud expensive mess.

They need a Ryzen equivalent form RTG, and it needs to be Navi.

old news, Vega is holy grail of GPUs now
get on with times gramps

>>62125597
Design wise, Navi is a huge leap in performance, but designing the software will be a headache and we all know how bad AMD at early driver.

File: undervolt.png (105KB, 1204x1407px) Image search: [Google]

105KB, 1204x1407px

>>62125652
why dont you improve your yields,leather jacket manlet?

>>62125597
>Ok /g/, Vega was a bit of a disappointment.
Pulling a fucking Larrabee is a disappointment?
The software lags behind though.

>>62125597
Considering its a scalable MCM you can expect anything.
>Fast cheap 600w TDP (a few fast dies)
>Faat efficient expensive (many slow dies)

One thing im sure of its gonna be a blast.

But wait a minute m8ty
You probavly are running a ficking 1080p display, you dont need a gpu like that for your shitty resolutionlet mibitor.

Either get a 580 for 1080p@60 or Vega 56 for 1080@high fps+freesynch

>>62125597
>>62125624
>>62125815
If Vega's release drivers weren't such a garbage fire and it wasn't overvolted to all fuck because AMD's trying to push for performance that with these drivers simply isn't there, it'd be a good gaymen card if it also actually fucking retailed for the original launch prices.
At the moment though it's only good for folks that need a lot of compute. And at even its current retail pricing the 64 is an absolute beast at compute, matching even the Titan Xp, which retails for quite a bit more.
Nvidia really needs to start implementing HBM2 in its consumer cards so prices can drop.
And yes, at the end of the everything CAN be blamed on Nvidia being the massive kikes that they are.
As for Navi, who knows, really. I'm hoping Ryzen pulls in the dosh for AMD so they have more to invest in, well, DRIVER SUPPORT, but I honestly wouldn't trust AMD with making Navi a good gaymen card at release, especially considering how revolutionary it will be in terms of design. Unless Nvidia releases their own equivalent to Navi AT THE EXACT SAME TIME, game developers probably aren't going to support the new design with proper software at release, because AMD has virtually no fucking mindshare. And since Nvidia, as I've already mentioned, are massive kikes, they won't release their Navi equivalent whether it's actually ready or not, just to sabotage AMD. Or they'll make it so it's impossible for developers to code for both at the same time.
Of course, AMD could get around this by going to game developers themselves with money in hand and forcing them to optimize for Navi, but that will all depend on Ryzen being a MASSIVE success so AMD can afford to go through that. So if you want Navi to succeed, buy Ryzen.

But at the end of the day none of this matters desu because gaymen is for manchildren and graphics cards are for rendering work and crypto mining. Which both Vega and Navi are and will be excellent at.in all likelihood.
Stay salty PC gaymen mustard race.

>>62128172
>So if you want Navi to succeed, buy Ryzen.
Why should i?
AMD already has a money-primting machine in EPYC.

>>62128181
Fair point. But EPYC is a couple months away and Navi is already in development, and so are all the games that will be released at the same time as Navi. So unless AMD is already going to game developers with money in hand, Navi is DOA.

>>62128215
>So unless AMD is already going to game developers with money in hand, Navi is DOA.
What?

>>62125597
just get an xfx480

>>62128172
>40-60B industry just on PC, add consoles it goes to the skies
>1000% investment returns
>muh rendering for movies that do not make even 200% back is serious business
but gaming is for children
you don't know what is going on
if AMD pulled Larrabee developers will flock to them themselves, it is a way to create engine that will work identically well on any AMD GPU
consoles = PC, save money on optimization, porting, development
no need for API and almost no need for driver
whoever makes such engine first, which is pretty hard, will win the market

>>62128391
Vega's shader model requires no explicit programming.
All is done by the driver.
I mean you can but why.

>>62128119
>literal arab amd shill
can't make this shit up

>>62128434
>arab
?

>>62128412
>All is done by the driver.
not. yet. if ever.

you will do custom pipeline to make your engine work ideally on everything that would support it
no porting costs
no optimization costs
"pc has too many configurations" will not be a thing

>>62128119
Looking at this i can only speculate that AMD fucked up their firmware somehow and got the voltage settings wrong on release cards?

>>62128492
Custom pipeline is only for AMD products.
Nvidia is not employing single compute shader for all things geometry.
>if ever.
It's literally confirmed by Rys all is done by the driver.
Heck they are not planning on exposing it any time soon.

>>62128534
>is only for AMD products.
yeah, and what is next generation consoles are?
platform wouldn't mean anything

>>62128605
We're talking dektops.
Pajeets in charge of PC ports are not going to bother with pipeline implementation.
It's up to drivers to handle Vega-likes properly.

>>62125597
i think they should do the same thing that they did with ryzen
make new architecture and release new line of cards that are cheap and aimed at "everyone"

>>62128678
Well Vega is a new uarch.
Bretty fucking bonkers one at that.

>>62128641
>PC ports are not going to bother with pipeline implementation.
and again, you do not understand
there won't be ports in current sense of a word, it's all happens within GPU
what it does for consoles is easy way to get 100% performance from it, no overhead
what it does for pc, ports do not need special treatment to be good running on AMD
and nvidia would need special treatment
look, this is not a tomorrow thing, maybe in 7 years if we are lucky
but hardware is a long game, so who cares

Is it just GCN that's too power hungry?
Will they ever release anything GCN based that doesn't have these issues?

>>62128440
He's the same sand nigger shitposting for AMD in every thread about either AMD or Nvidia.

>>62128846
>sand nigger
?
>>62128832
>GCN
That's ISA. Vega is the uarch.

My god, you marketing faggots do not rest do you? Fiji, Polaris, Vega are either all Raja, or significant input from Raja. GPU design is much faster than CPUs. And he has been there since 2014. RTG with Raja at the helm was formed about a year later after his return.

Navi is just two to four shrunk down on 7nm node low powered Vegas connected by infinity fabric like Epyc and Threadripper.

Throw in that those Vega dies in Navi now have to be written as one single die at the driver level and AMD is fucked. They couldn't even get one Vega fully functioning, yiu exoect them to get multiple Vegas functioning?

Navi will be a massive failure. Just like Vega.

>>62128119
AKA the best Vega can do with 14 months extra development is equal or get beaten by the competition badly.

>>62129306
GPU R&D cycles are 3 years long at best, anon.
And you do understand that AMD has years of NUMA expirience?
Hell they brought NUMA to x86.

>>62125624

Its a good card when undervolted. But not at current price.

The 1080ti is 50$ more.

>>62129383
That still leaves Vega as all Raja. And since it's basically Fiji 2.0 how much input do you think he had on that chip?

All that theoretical NUMA experience meams jack shit when dealing with interposers and HBM and HBCC. They still have to make the Vega dies transparent at the driver level. And we see the teething problems Ryzen has with interdie latency.

Navi is fucked, bevause Vega is fucked.

>>62129508
>Fiji 2.0
U wot?
Vega brings the biggest fucking change since fucking unified shaders by merging everything but PS into one low overhead compute shader.

Doesn't matter how good Navi will be hardware wise because RTG will fuck the drivers.

>>62129594
The only way they can fuck up the drivers is merging pixel shader into generalized compute shader (think of pixel as a primitive).

>>62129526
Vega performs like an overclocked Fiji

>>62129693
It performs like big Polaris.
And of course it should.

>>62129703
But Vegs 64 draw more power than Fury X

>>62129693
and GCN 1.0 performed like underclocked VLIW4
where is VLIW now and where is GCN 1

>>62129756
no it doesn't, do you read only graphs that show the biggest number and not notice when it says "power limit disabled"?

>>62129756
And? What the fuck are you even trying to say?

>>62129779
That Vega is trash. And RTG should be fucking embarrassed that this is the best they can do.

Navi should be pretty good.

Vega shit the bed because Ryzen ate 90% of all of the R&D. Vega had just no budget towork with. I suspect now that Ryzen is out that the "low hanging fruit" optiiziations will be cheaper to develop thus the R&D budget will now shift to Navi.

>>62129826
Vega is bad only due to immature software
The hardware is more than sound, my child.

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

906KB, 793x877px

>>62125597
It's still too early to have any solid conclusions about how Vega performs yet, much less Navi.

Primitive shaders are one of the core uarch features of Vega and aren't even minimally implemented in drivers yet. Until they are, Vega will remain massively front end bottlenecked by being limited to 4 triangles per clock of culled geometry. This is why Vega 56 is 100% as fast as Vega 64 at the same clocks right now.

The minimal default implementation of primitive shaders should solve GCN's historical front end bottleneck issues by significantly increasing culled geometry throughput, but primitive shaders will eventually be capable of much more than that as the white paper describes them as generalized non-compute shaders, which is pretty fucking impressive.

>>62129860
It's much too late for Navi. By this time they're probably long done with the initial design and simulation and getting samples

>>62130439
I like you anon. Makes sense.

>>62130475
>getting samples
Tapeouts for 7nm products are late 2017.

>>62130439
>pretty fucking impressive
It's extremely fucking impressive.
Generalization of non-compute shaders is long overdue.

>>62130439
>It's still too early to have any solid conclusions about how Vega performs yet, much less Navi.

I have one solid conclusion about it: AMD failed to implement the architectural improvements which supposedly make it fast as fuck on launch day, after months and months of hyping

that marks them as incompetent for me, and quite frankly I'd rather buy a product that works as intended on day 1 instead of waiting & hoping for a miraculous performance uplifting driver update

>>62130682
Then buy fucking nVidia and stop crying you fag.
Wow, a faceless corporation threw extremely volatile market under the bus for the less volatile one.

>>62130682
not sure why you're surprised about this
remember how shit 7970 was on launch and how it completely dominated 680 after a year of driver improvements
AMD was on the verge of bankruptcy until a few months ago, hopefully now they can spend some money on a driver development team.

>>62130682
That's budget constraints for you. They didn't have enough driver monkeys to have the pro and gaming drivers both ready by launch, and they clearly prioritized the pro drivers over the gaming drivers.

Hopefully they get prim shader geometry culling working by the time AIB cards hit the market.

>>62130776
They already spent money on driver development team, it's twice the size of the one from three years ago.
It's just implementing mini-pipelines is fucking difficult because no one besides Intel did that before and Larrabee was canned anyway.

>>62130829
>They already spent money on driver development team, it's twice the size of the one from three years ago.
Sauce?

>>62130889
everything they earned went into R&D that also means drivers
they could've been profitable q1 2017 if they were stupid with money

Anyway why the fuck it's AMD of all companies that is moving towards software defined GPU?
Ahoy, NVIDIA, wake up, the train is moving again.

>>62130947
Nvidia didn't get console market for a reason.

>>62130921
So basically no source.

>>62131029
not the guy, I don't think they doubled the team, it would be very expensive but they added focus to software that's for sure

>>62130986
Well technically Switch is a console (and not a rebranded Shield with custom *bsd-based OS).

>>62131068
well, it is also technical disaster

>>62131099
No, but it's nothing special.

>>62131196
no, he's right, switch is a broken piece of shit

>>62125597
Navi won't release until late summer 2018 at the earliest. You won't be able to buy one at msrp until December 2018. If you are fine waiting for over a year, then wait.

File: 1317726561959.jpg (11KB, 250x250px) Image search: [Google]

11KB, 250x250px

>>62125597
>yfw wait for navi threads

AMD fanboys are walking memes

So Vega 56 is better than GTX 1070 and cheaper, innit? "AMD failed boohoo". The shills are truly amazing.

>>62131247
it's actually confirmed 2019.
2018 best we get vega refresh or if very lucky vega 48 for $280-300

>>62125769
Why would they be? At these prices the 1070 and 1080 are not only the better absolute performance in most cases, bit also the better price/performance which used to be amd's bread and butter.

>>62131286
yeah, but you can't buy it, it's sort of a problem

>>62131286
It's a bit faster, but also burns a heck of alot more power, and thus outputs more heat.

1070 prices are also inflated as fuck right now because of mining

And yes, releasing an architecture than can only match the third fastest card of the competition over a year later with much more power consumption is pretty unprecedented failing in the gpu market.

The gap hasn't been this big between both parties in a very long time

>>62131344
>but also burns a heck of alot more power
20w more.

>>62125597
AMD never did a good GPU

>>62131344
The software stack surrounding Vega is unfinished (well for gayms). And the card needs one.

File: 5.jpg (173KB, 846x1154px) Image search: [Google]

173KB, 846x1154px

>>62131368
No

Also considering HBM2 draws about 35-75w less than regular GDDR solutions, it makes these results even more pathetic

File: 7.jpg (121KB, 691x1073px) Image search: [Google]

121KB, 691x1073px

>>62131432
Keep in mind, this is on balanced mode, kek

>>62131507
>>62131432
Okay, it's not 20w, it's 60w but temperatures are magically the same, why?
Also it should be compared only to AIB 1070 because you can't, and won't buy reference.
I'm not even going to compare electricity bills, $3 per year not worth mentioning.

File: 1481660299502.png (32KB, 711x533px) Image search: [Google]

32KB, 711x533px

>>62131575

Oh goddammit why did they even launch RX Vega.
I mean kids were already screaming murder but now they are screaming murder even more.

>>62131619
kids aren't buying these

>>62131641
Well, man-kids. Whatever. Preemptive launch was totally unnecessary, they should've launched it mid-October.

>>62131663
stock holders will get angry

>>62131683
Radeon Pro would tame their anger.
For such a ballsy move towards software-defined GPU the software is blatantly unfinished.
Launching the cards in this state is dumb given the gaymen market is 98% mindshare and 2% decent hardware.

>>62125610
vega was all his hands on work

I don't understand why Vega exists at all. The plans for a massive, expensive die with HBM2 should have been laughed out the door when they were drawn up 3-4 years ago. AMD should have taken any R&D money that was allocated to Vega and thrown it towards making the Polaris architecture the best it could be in addition to better software and driver budgets as well as planning for future MCM. MCM is a hail mary but it's also the only way AMD could ever crawl back into the lead at this point since it's the one area they still may have an R&D lead on Nvidia, though that grows increasingly unlikely with Nvidia's revenue.

Every time AMD/ATi has tried to make a big fat GPU they have gotten their fucking doors blown off. Fiji should have been their last attempt and you could argue that that never should have existed either.

>>62131826
Who cares the software is not there. Hardware dudes can do only so much.

>>62131826
It wasnt. He stated before that Navi will be the first project with him from start to finish

>>62131847
Polaris was a dead end.

>>62131847
ATi has a history of making killer big die GPUs, starting with R300.
Vega10 also targets more than once fucking market.

>>62131847
If anons are correct, than the plan wasn't to just try and create hueg die. It was to try to change the game

>>62131930
Yes, it's the first (in 6 years) move towards a fucking software defined GPU.
Fucking hell the GPU market is hilariously static.
Just throw moar ALUs!

>>62131344
>>62131432
>>62131507
>>62131608
Gee, you mean the perf/watt of Vega is shit when core uarch features crucial to its performance are still disabled? Wow. I'm shocked.

Once Vega is no longer front end geometry bottlenecked, it's perf/watt will be fine.

>>62125597
>Scalability
This kills nVidia

>>62132060
No, nVidia also consider MCM.
Anyway who cares I want a Larrabee-like that actually works.

>>62132083
Larrabee was and is a terrible idea, why do you want it?

>>62132094
Larrabee was terrible because of retarded idea of in-order cores running HUEG vectors.
The concept of software-defined GPU is very good in itself.

>>62132121
No it's not. It introduces software overhead as the drivers have to detect what sort of code is being executes which adds an unnecessary level of driver responsibilities. And for a company that has historically poor drivers it's a nightmare.

Maybe a stupid question, but still.
If GPU is "software defined" What does that means for CPU? Load increases/decreases/changes?

>>62131372
how old are you?

>>62132178
A little bit of driver overhead is certainly better than wasting die area on retarded amount of FF units.

>>62132183
decreases, there is no need for driver or API if engine pipeline is in the GPU already

>>62132178
are you familiar with concept of "dumb" driver? that's what software defined gpu means
if engine supports it, it always will run perfectly on hardware, it basically becomes ASIC for any engine coded for it

>>62132183
It introduces a little bit of driver overhead. Driver identifies what pipeline/mini-pipeline to execute based on input data.

>>62132247
Dumb driver is a meme since it's basically an edge case generator.

>>62132248
the giant cache system is for that
you are confusing what prim shaders will do in auto regime, and what is real software defined gpu

>>62131977
Too bad AMD can't into software.

>>62132298
not if your engine runs the gpu, not other way around

Don't even need to read this thread to know it's full of fanboy cope and bullshit. They fucked up just deal with it, stop being a little bitch.

>>62130986
Because they A) Burned their brisges by being.liney grubbing fucks. And B) Refused to chase the razor thin margins chasing after AMD to the bottom.

That's it.

>>62132355
And Navi is going to be even more software dependant than Vega is.

>>62132492
>implying nvidia didn't offer their solution to console makers

File: 1502086187799.png (806KB, 638x958px) Image search: [Google]

806KB, 638x958px

When the FineWine™ Primitive Shaders driver comes out shit is seriously gonna hit the fan.

>>62131864
>>62131874
wow he sure shouldent have talked out of his ass that vega would be good then :^)

did he reveal this information before or after it looked like vega was another massive disappointment?

>>62132386
Not every engine dev will have enough balls to touch the GPU directly, albeit only in some delicate places.
>>62132355
They very much can. They just need more manpower.
>>62132299
I mean 45megs of SRAM sounds sane if you consider that.

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

255KB, 1280x1440px

>vega
>disappointment
It's literally shitting all over the gtx 1080

>tfw waiting for primitive shaders to be enabled in gaymen drivers and see vega rape nvidia in 8x msaa
>tfw 24" 1440p monitors don't even need AA

>>62132579
Jensen please.
Your meme machines are being hunted by ASICs and you still need to beg for EBYN, pal.

>>62125597
For servers/farms I can see how this can work, but for the end-user and gamer? They can't compete with Nvidia, no way.

>>62132583
>Not every engine dev will have enough balls to touch the GPU directly, albeit only in some delicate places
the ue4,unity, ea and ubisoft will
they want to run their games 60fps 4k on next gen consoles, and if they do that for consoles it will work on pc due to lack of API

>>62132623
wrong quote?

>>62132616
That's an edge case.
And no, it's GP102 competitor, the performance right now is abysmal given the die size.

>>62132616
no one plays dirt 4 anyway peasant

>>62128391
It's called Vulkan. All real engines support it. Developers just don't want to do

>>62130719
awww, so angry, did I hurt your feefees?

>>62130787
I hope so, I want to see AMD do a comeback, but I doubt it's going to happen with Vega

File: 1491757511426.png (72KB, 253x307px) Image search: [Google]

72KB, 253x307px

>>62132824
Fishing for (You)s, pal?

>>62132616
>this one outlier shows that vega is faster than 1080!

no, in terms of averages - worst case scenario vega is on par with 1080

best case scenario it's a bit better

vega launch was a d i s a s t e r no matter in what light you try to paint it

>>62132840
They've already made a comeback.
Just not in gaymen space.
And I don't think they'll ever offer anything but overvolted trashbins to gaymen, ever.

Honestly? No

Mcm Navi will probably be garbage and have a predictably horrible launch.

>>62132878
Just like EBYN, right?
Oh wait, no.

>>62132849
no, you're just a brainless fanboy
"once these MAGIC drivers get released, it'll get a HUGE performance uplift and beat the shit out of nvidia!!!"

where have I heard this before? oh right, when rx 480 and fury x was being released
>1 year later
>5% faster than nvidia competitor
give me a break

>>62132894
Dude ryzen launch issues are barely fixed I am still having remember trouble months later

>>62132910
Oh, totally hungry for (You)s.

>>62132961
Ryzen doesn't even have launch issues. The segfault issue only happens because normie birdbrains don't adjust the soc voltage like you should and even then it only happens in scenarios where cache thrashing occurs.

File: 1447174425222.jpg (119KB, 796x805px) Image search: [Google]

119KB, 796x805px

>>62132728
>>62132869
(You)

>>62133012
>Have to undervolt Vega to get perofrmance it should have had
>Have to overvolt ryzen so it's even stable at stock speeds
Wow ayyyymd

I can't wait to fucking ditch x86 both sides are so shit

>>62133012
ryzen didn't have issue only with one thing at launch - price and power consumption
same fucking complaints as with vega, it was nightmare first couple months

>>62131344
Difference is 1070s and 1080s are in stock for less than vega 58 and 64 which are not in stock.

>>62133055
>I can't wait to fucking ditch x86 both sides are so shit
I guess get comfy, it will be a long wait.

>>62133012
>Ryzen doesn't even have launch issues.

holy fuck, you're either a shill or don't own a ryzen cpu

here are the problems that I've when I first switched from my i5 4670 to a r7 1700:

>post takes ~40 secs
>can't set memory higher than 2133mhz
>random pc restarts (regardless of CPU load)
>random pc restarts when switching monitors on and off
>fails to wake from sleep

here are problems that I still have with the most up-to-date bios:
>random pc restarts (regardless of CPU load)
>random pc restarts when switching monitors on and off
>fails to wake from sleep

I bet you're this fucking guy too >>62133005

>>62132816
Vulkan just goes to prove that the game develpment community is saturated with kiddies who just got out of uni cutting their teeth on DX9/11 and don't want to learn new shit (especially when on tight schedules set by their employers).

File: good job idiot.png (411KB, 2282x744px) Image search: [Google]

411KB, 2282x744px

>>62133044
brainless idiot

>>62133081
Are you me? Try overvolt ram to 1.365-1.4 then try the 1.0 soc volt

I got lucky I got 3200 ram to run at 2933 stable on my 1600x

>>62133044
Its true, thats the face i had a few weeks after buying my current card, the GTX 970.

>>62133081
>random pc restarts (regardless of CPU load)
>random pc restarts when switching monitors on and off
I think something is wrong on your end.

>fails to wake from sleep
removing battery might help, microcode update probably didn't take in
they fixed that

>>62125597
I don't know why you people find these questions so difficult to answer on your own. As it is right now, do you find that you're having to make too many compromises in your stupid video games with dual 780Tis? If the answer is yes, go out and buy two 1080Tis and call it a day. If no, then fucking wait.

The idea that you NEED to upgrade just because it's been awhile is retarded. If you don't NEED the performance boost, then wait, not just because "V-VEGA" or "N-NAVI" or "AWW VOLTA IS GONNA", wait because why the fuck are you so inpatient about spending money on things you don't need in the first place? Wait until you actually need it, then buy what you think is the best, Nvidia or AMD. Waiting for waitings sake when you NEED an upgrade is almost as retarded as upgrading for no reason.

>>62133081
You sound like a retard who can't even trouble shoot your own issues. I have a 1700 @3.8ghz on all cores, c6h, and 16gb 3200 trident z. No problems with my build other than one of my RAM sticks being faulty and needing RMA, but that's got nothing to do with Ryzen

>>62132616
Fucking kek, I knew we could count on one AMD faggot to take this outlier and post it. He already explained what had caused it in a follow up video, it had to do with him one type of AA

>>62133103
Vulkan was only good in one game though and that's because of their FUCKING wizard software culling skills

Everyone moved to dx11 multithreaded and dx12

File: 14821.png (21KB, 200x194px) Image search: [Google]

21KB, 200x194px

>>62128172
In summary :
> RTG is abandoning gaymen performance
> Is instead going after compute
> Have no mindshare, drivers suck dick, as does software stack

You don't capture the compute market with a shit software stack, drivers, dev tools, and a lack of mindshare.

So, this is equivalent to a 4/10 chick leaving her man (8/10) for greener pastures not realizing how low her worth is and then getting dicked down by chads who have no interest in her beyond a casual fuck (miners). Yes, it appear RTG is this dumb.

When I think general purpose compute, the last thing that comes to mind is RTG

>>62132027
>Once Vega is no longer front end geometry bottlenecked, it's perf/watt will be fine
So you mean never

>just wait

>>62133157
I'm pretty sure it's one guy.
But I find it stupid to use AA on 1440p and higher.

>>62133169
Dx12 is no easier. The difference is Microsoft finished the preliminary spec sooner than vulkan then seeded studios with money to use it. And for the record, some studio had 4 guys implement a dx12 renderer for cryengine in like 6 weeks

>>62133189
>But I find it stupid to use AA on 1440p and higher

Can this meme end, it might be less noticable on higher resolutions but you definately need AA on 1440p, I have been gaming on it for over 2 years

>>62133118
>I think something is wrong on your end.
no shit?
>removing battery might help, microcode update probably didn't take in
this has been since version 1.0 AKA launch, my current bios version is 1.7 and I've been having these issues in every single BIOS version

>>62133116
I actually got a bit lucky in terms of memory, I have 2x8gb RAM sticks rated at 3000mhz, but run them at 3066mhz with the DRAM voltage at 1.4v and SOC at 1.1

>>62133155
shut the fuck up, idiot

>>62133183
390x still has abysmal performance in newer games as well past 2016 they just stopped trying

>>62133187
>Tahiti gained 30% performance with one single driver

>>62133183
Productivity apps are not compute. Its 90% the same rendering but the workloads are different from games

>>62133213
Oh sweet might try that tomorrow

>>62133209
me too, I can't see aliasing

>>62133220
Even if that happened, (it didn't) that's no guarantee it will happen again in the future

>>62133246
it did, but year later

>>62133238
If you have Rise of the Tomb Raider and use the benchmark (second test) that ledge on the left hand side looks atrocious no matter what level of AA used. I bet it looks shit at 4K too.

>>62133257
>it did
Show proof

>>62133213
>no shit?
I mean I never had random restarts once for 4 months. It's either your PSU or GPU drivers used to do that.

>>62133118
Restarts are a psu or motherboard issue. A CPU won't trigger a no warning restart unless it's exceptional like thermal protection kicking in.

Turn off quick start in bios and/or hybrid start in windows. Or reinstall Windows. Being unable to wake is almost always a Windows issue and my 2500k build with win 8.1 occasionally had problems too with sleep until I clean installed to win10

>>62133289
oh, well it's been like that since launch and I'm fairly sure it's the bios at this point, because I've also switched out a 9 year old 620W PSU to a brand new 750W PSU yet the issue still persists

>>62133230
I'm speaking about OpenCL.
So I'm speaking about compute.

CUDA is the standard because it has good dev tools, drivers, and software stack. Rendering suites are the same... They're dominated by the company with the best drivers, mindshare, and software stack. A place RTG lacks.

>>62133279
oh but if you move camera 13.7 degrees to the left and 7.2 degrees up it will look even worse!

point is I can't see it during gameplay, that's all I need

>>62133323
Vega is no good for compute outside of a handful of cases where AMD worked with the developers to optimize it, like blender

>>62133303
Try a different outlet or get a ups

>>62133373
done that too, didn't help

>>62133392
Reseat the 4/8 pins and the 24 pins on your motherboard. If that doesn't change then rma. CPU shouldn't cause what you're describing

>>62133282
give him a few minutes to boot up photoshop m8

>>62133373
>>62133392
Get an UPS, you probably have shit power delivery in your entire building.

>>62125652
Its not bad, just undervolt it

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

37KB, 711x533px

>>62132910
Core uarch features are not magic drivers or finewine. The default implementation of primitive shaders is well described in the Vega white paper, and if even 50% of the benefits claimed in the white paper are realized in practice, Vega's culled geometry throughput will literally double and Vega will no longer be front end bottlenecked like Fiji was. Your inability or unwillingness to read and understand the white paper are your problem.

>>62133187
RTG already got prim shader geometry culling working on their pro drivers, it'll be in the gaming drivers soon enough.

>>62131826
vega is literally fiji with better microcode moron

>>62133886
?

>>62133865
no primitive shaders are NOT working the discard is working
if the primitives were working creo would have been a field day for amd

>>62134061
The gaming drivers have fixed function culling too. If you want to claim the pro drivers don't have the default version of primitive shaders enabled you are going to have to provide a coherent explanation for Vega 10 providing GP102 performance in pro 3D rendering applications but only GP104 performance in games.

It is my understanding that the massive difference in performance in the two geometry bound workloads is mostly down to primitive shaders culling working in the pro drivers already but not in the gaming drivers.

>>62134103
Shrout said primshader culling is disabled in any public drivers.
Looks like it's PDA that does the heavy lifting.

>>62134103
the shaders wont work on a program like creo that needs everything to be rendered
the discard function is active and that only so far and not because of the drivers obviously its working in safe mode not even catching the 4 number (its 3.75 actually)

>>62134103
>coherent explanation for Vega 10 providing GP102 performance in pro 3D rendering applications but only GP104 performance in games.
Cache thrashing in games

>>62134234
Vega flushes less often than Polaris tho.

>>62128896
Wait, EVEN FUCKING GPUS USE µOPS CISC RISK CONVERSION? WHY? You don't need machine compatibility like with x86. Just implement an API. WHY REUSE THE SAME ISA FOR A DIFFERENT ARCHITECTURE DESIGN?

>>62134450
>WHY REUSE THE SAME ISA FOR A DIFFERENT ARCHITECTURE DESIGN?
Because you need to write drivers for it retard... GPUs remain dumb devices

>>62134461
Doesn't Nvidia start from scratch every generation? Like, aren't Fermi, Kepler, Maxwell and Pascal completely 100% different from one another?

>>62134450
No, both GCN and G80 derivatives are pure RISC SIMT.
>Just implement an API.
You mean just write a driver?
>>62134484
It's still G80, somewhere underneath.

>>62134484
low level bait
or
just ignorant

>>62134495
>You mean just write a driver?
Why is writing a driver for hardware you designed yourself considered a challenge?

>>62134495
>>62134500
>It's still G80, somewhere underneath.
>just ignorant
How do you know that? Do you work with the reverse engineering team for the linux drivers or what?

>>62134538
>Why is writing a driver for hardware you designed yourself considered a challenge?
Oh boy.
GPU is fucking retarded.
Guess what driver does.
Yes, doing that is HARD.
>How do you know that?
I dunno maybe both companies publish the ISA specs along with whitepapers?

>>62134562
Since when does nvidia release specs to the public? All I know is that they have some IR for cuda shit.

>Guess what driver does. Yes, doing that is HARD.
Even with Vulkan and other low level api shit? Vulkan is basically the C of GPU programming. Portable low level control.

>>62134634
>Even with Vulkan and other low level api shit?
Oh yes.
Vulkan/DX12 makes driver layer thinner, but that's it.
There's still a fat slab of drivers.

>>62134651
How do you know that? Do you work for nvidia or amd? If so, please leak me the VESA specs for DisplayPort adaptive sync and nvidia gsync.

>>62125597
They can't even spell next-gen right so probably not

File: t3_fa1t8.jpg (236KB, 1177x805px) Image search: [Google]

236KB, 1177x805px

>>62134538

>>62134694
>ati
What year is it?

>>62134662
>please leak me the VESA specs for DisplayPort adaptive sync
what?

>>62134774
I want to find a way to convert Gsync signals to freesync signals. Both use VBLANK timing, can't be that hard if I have the spec. Also I would love to generate DisplayPort signals in software, since I bought a USB-C to DisplayPort adapter cable for the lulz recently.

>>62133358
Correct. RTG's radeon cards aren't even supported in Tensorflow as well as other packages. This was eye opening to learn.

>>62133865
>>62134103
>>62134234
Does anyone have a link to any technical documentation as to what in the world HBCC does beyond on-demand paging that video cards already do and what pinned memory does in CUDA?

Also, does anyone know if FP16 is going to be available for compute?
I am hoping to close out my consideration of Vega for good today.

>>62134890
http://radeon.com/_downloads/vega-whitepaper-11.6.17.pdf

ever tried to read whitepaper?

>>62134694
Does this really work?

>>62135158
Only if the right demon is summoned.

>>62134931
>http://radeon.com/_downloads/vega-whitepaper-11.6.17.pdf
I have. This reads like marketing slides not a white paper.
This is what a white paper looks like :
> https://www.nvidia.com/docs/IO/40049/Dual_copy_engines.pdf
^ I hope this isn't want HBCC is... RTG finally catching up to Nvidia.

HBCC :
HBCC still hasn't been detailed in terms of how it functions or what makes it better than how memory is paged back and forth today in gaming mode. HBCC was clearly designed for the Pro cards and is an afterthought for consumer cards that lack NVME. No understanding what is required of developers to get it to work. Something is required which is why there are videos from bethesda saying they're working with RTG. No understanding of if this will be open to compute data flows or general code developers.

FP16 :
Wording suggests its available to compute.

1/2 inquiries addressed.

>>62135411
HBCC is completely transparent (on Windows, at least). One tick in Radeon Settings to page-based memory management.
That's it.

>>62135555
> Technical details ..
> Technical inquiry

>>62135411
yeah, "real" whitepaper

https://international.download.nvidia.com/geforce-com/international/pdfs/GeForce_GTX_980_Whitepaper_FINAL.PDF

>>62133865
>just wait

Every time

>>62135758
>https://international.download.nvidia.com/geforce-com/international/pdfs/GeForce_GTX_980_Whitepaper_FINAL.PDF

Congrats, you now understand what isn't a "real" white paper that doesn't detail technology to a degree worthy of being called a whitepaper.

Glad you were capable of linking to a similar 'non white-paper' example. You now understand what level of detail i'm looking for.

>>62133886
stay in denial

>>62136397
Just Wait™ for magic drivers to unlock everything and beat the Titan!

>>62128172

>chasing performance

That's AMD's problem. They designed a very efficient chip with rock solid perf/watt ration, and then threw it straight into the trashcan by overclocking and overvolting to the absolute limits of the fabrication process coupled with everything else going on, on the chip.

Making it a clusterfuck of multitude of failures.

They should have instead pushed the absolute boundary of perf/watt efficiency, stopped there; and then with the remaining money, paid developers to create CrossFire profiles for their GPUs and optimized for them. Having a single GPU that's 75-80% of a 1070, but with unparalleled perf/watt efficiency w/ proper CrossFire profile would be more valuable to the consumer than having a GPU that performs between a 1070 and 1080 while drawing 2-300W more than the 1080Ti on average.

>Design a 150W GPU
>that is able to perform around 75-80% of a 1070
>focus driver optimization and developer outreach on CrossFire tech (as a setup to Navi with developers getting used to optimizing for multi-GPU configs)
>2x150W GPUs cost equivalent to 15-20% higher than a 1080, while costing as much as a 1080
>and drawing only 50W more than a 1080

Vega all in all, is just a complete waste of everything and a collective disappointment for the market as a result. Nvidia can now proceed to continue their price gouging for Pascal AND Volta, because Vega can't even compete with 14 months old Pascal.

Fuck you RTG.

>>62125597
No, no no!

Vega was NOT a disappointment

the waiting meme was preemptive

you need to wait AGAIN for drivers

vega will be AMAZING in 3 months just WAIT

>>62132535

Nvidia had the better GPU to offer with the efficiency. But they lack an x86 license, which they tried to get from Intel in a cross licensing agreement; which Intel looked at and went "that's cute." And instead gave Nvidia $1Bn in cash for their GPU/ARM IP licenses.

If Nvidia had managed to snag an x86 license, its possible that consoles would have had x86 AMD + Nvidia in them. But without an x86 license, Nvidia's pretty worthless in console space.

AMD on the other hand, has had a solid history of integrating a full functioning GPU with a CPU into an APU, and they have both an x86 and GPU division under one roof. Which makes them perfect for Microsoft & Sony for console spec design requests.

Thus, AMD was effectively able to monopolize Xbox & Sony consoles under their belt. Xbox Scorpio & PS4 Pro are AMD all around. Xbox Scorpio II and PS5 will once again be AMD all around; because arguably upwards of a billion dollars (combined) has gone into developing an optimized software stack for AMD CPU/GPUs.

Nvidia won't ever make market penetration into console market again.

>>62133204

And 100% of all DX12 implementations are garbage; and 100% of all Vulkan implementations are well executed.

Hmm.