AMD does it again, releases a new arch with disabled features

AMD does it again, releases a new arch with disabled features.

Clearly the software was not ready, why even release it unfinished?

Mommy got pissed.

Honestly though, high clocked Vega 56 AIB cards will save this arch for gaymen

>Clearly the software was not ready, why even release it unfinished?
They didn't release it.
Vega launches on August 14th (yes, another delay).

>They didn't release it.
What is FE?

That's not RX Vega though.

Early Access
really shows that gaming drivers are not a priority for AMD at this point.

I guess that's reasonable.
SSG is the product nVidia can't compete against.

Gamers got shafted by AMD again... another arch for compute... why AMD why do you do the same shit every time, fuck you AMD FUCK YOU, Nvidia makes gamer hardware not this psuedo enterprise crap you make, you have tricked me for the last time, no gamer will ever want your GPU again..!

Guess what Volta is centered around? Spoilers: It's memes, memes everywhere. Gaymurs are no longer priority #1, deal with it.

AMD did their best releasing the best gaymer cards being HD4000 - HD7000 series.
No one bought them.
Now they are carving themselves a niche.

>rx vega special instructions for mining
REEEEEEEEEEEEEEEEE

Those pajeet at amd are getting torn apart I bet.

Managers probably brought in cots for those bitches to sleep on (for a couple hours at most of course). They're never going home until VEGA is released.

Fuck you AMD!!!

Gaymers got pissed.

They managed to make Pro drivers work, AMD was quite happy to present WX9100 and SSG.
Now only gaymen shit left.
Also Sweeney on AMD keynote?
The fug.

No amd make shit Gpu, dont buy it

I had a 7950. It was a great card.

They have two weeks left to polish up the graphics on level 3.

Yes, Tahitis were fucking stellar.
They sold worse than anything 600 series though.
More like a little bit more than 1 week.
Reviewers need to do the benchmarking.

>two weeks left
reviewers definitely already got vega drivers
it looks like ryzen all over again

Fuck them, they can benchmark it in a day. Just express ship the GPUs with air mail.

Not likely.
Though RTG may share their internal driver branch with them.
They're all under NDA anyway.

AMD know where the bread and butter is. They don't care if you buy it for gaming or mining anymore just as long as they make a sale.

it was weird how they released new feature set drivers last week without slight mention of vega in them
something brewing, hope it's good

AMD doesn't include mining into financial prognosis at all

They got Pro drivers working, that's good.
RX Vega being paperlaunced on SIGGRAPH is more of a desperation move.

ryzen launch was paper as well then
the date is solid 14th, even asus announced it

Good.
Vapor-X Vega56 when?

Also, 7k shekels for SSG?
Faggots know nVidia can't touch it.

early september for strix, meaning sapphire ones will be mid september, late september at best
sapphire is slow

they created niche nobody knew existed

>wait for Vega
>now wait for Sapphire
Oh well.
At least Sapphire delivers consistently good cards.
Heck, they don't even need custom PCB, just slap a cooler.

well fuck looks like i'm going to have pre-order day 1 if i even want a chance at getting one. at least i have until august 14 to decide if i want a vega 64 or a potentially cheap 1080ti.

>first card with 2tb memory and 8k video decode/playback
They can charge 10k usd easily.

WX9100 is priced very conservatively though.
Not just decode/playback. SSG allows one to do live 8k video editing.

>he doesnt play it the way its meant to be played

Jensen can your dudes make SSG competitor?

>WX9100 is priced very conservatively though.
Because it competes with P6000, which is disgustingly overpriced.

2x of FE sounds reasonable, and 9100 is 250W TDP

>disgustingly overpriced
Let's call it "nVidia tax".

They learned that move from Intel

>why even release it unfinished?
to meet the Q2 launch time frame

Actually Vega is great at geometry, it's just bandwidth starved because DSBR/TBR isn't working yet. Wait till you see how well they implement DSBR before you rage. As retarded as this sounds we still have barely any idea of what Vega's ceiling is.

Couldn't they just say it was 2.2x performance of the original? That chart is completely unwarranted.

They have the hard numbers in the endnotes anyway.

No, because you need a fancy chart and at least the baseline is either 0 or 1 and the jump to two is linear

Except that's only big Volta. Nvidia always makes reduced dies for the gamer market. Vega' s double rate half precision and half rate dp probably adds like 60-80mm to the die which is an added cost

We already know what Vegas ceiling is. 70% better than Fiji. One of the bench nazi tom'shardware did was a shader compute heavy test with minimal geometry, textures, etc. It performed 70% better at 1600mhz

Gaming Vega won't even be bad as people think it is, considering the tested cards are reference V64 that are clocked under 1400MHz while Vega can reach over 1700MHz.

With a good overclock it can handily beat a 1080, the other disabled features should help it as well, can you power supply handle that though?

No 1080ti though, but as long as it's cheap it's less important.

I don't think it's reasonable to suggest that you can use that information to infer how well AMD's implementation of DSBR is going to work and therefore how much of that theoretical performance gain will actually be realized, no?

It's still not going to be the huge generational leap we've seen from NVidya in the past.

There's no double node jumps left anymore.

It's guaranteed to be lower in real games. The performance in compute benches suggests there are other problems than bandwidth. It could be any number ofnthinga, maybe latency, poor scheduling, etc. The shader compute bench is a best case scenario

DSBR isn't a performance feature, it indirectly impacts performance by making the GPU do less.


Currently Vega is memory bandwidth bottlenecked in games, AMD expects DSBR to reduce the bandwidth requirements by 30%
You can also see that FE memory overclocking increases performance quite a lot compared to core clocks.

Secondly, doing less work the GPU can stably maintain higher clockspeeds, and for Vega this is critical because reference cards run nowhere near their boost clock, some 1400MHz for FE but even under that for reference Vega RX with Polaris cooler, DSBR should enable higher stock clocks.

Yes, but perf at Creo/CB15/AutoCAD/whatever was stellar.

gcn was always a compute uarch nothing has changed

the volta nvidia released has nothing to do with the volta that we know its coming..its just a rebrand 4th in a row

Iirc Maya and max had worse performance than p6000 while the others had better performance. They just did a good job with drivers. And at the same time those workloads typically don't have a steady stream of assets being streamed into and discarded from memory like games.

Yeah, but that's FE, AMD released new drivers for pro just yesterday and it's FE compatibilte.

FE also has no ECC or certification so it's gonna be slower than a equal die that has WX firmware.

Certified drives increase performance in specific applications from either 5% to 300%, the latter obviously won't happen but WX should neatly leap over a P6000 by some low digit percentage, and that's really good.

Low double digit.

That sounds nice.
But what about support and ecosystem?

7 year warranty and support.

Ecosystem? Who knows, as said these really aren't ML cards, the ecosystem is infantile there, and I don't think their support for Maya plugins and whatever are bad.
It should be fine, WX Polaris cards were pretty good sellers, these should do a lot better.

Wait WX9100 will beat P6000 while having the same TDP?
Why can't gaymen cards do the same.

>gaymen

You're asking why AMD would rather sell and keep good dies for $2000+ cards instead of a $500 ones?

AMD even when it had good cards didn't sell as expected, professional cards are guaranteed sales.

No I mean why can FE compete with P6000 without certified drivers and yet RX Vega barely grazes full GP104.

Because AMD

HOW CAN YOU EVEN DO THAT?

Ask AMD

>DSBR isn't a performance feature, it indirectly impacts performance by making the GPU do less.
No shit Sherlock.

My point was that we can't accurately assess how successful AMD will be at implementing DSBR and thus what proportion of the theoretical benefits will actually be realized.

>No I mean why can FE compete with P6000 without certified drivers and yet RX Vega barely grazes full GP104.

Do pro rendering workloads typically require less overdrawing and thus tend to avoid the bandwidth bottleneck Vega FE currently has, whereas games have tons of overdrawing and thus are bottlenecked hard by memory bandwidth in the absence of DSBR?

IF that's the case, it could easily explain the disparity of results.