AMD Vega spotted - performance between the 1080 and 1080Ti

videocardz.com/67242/amd-vega-with-64-compute-units-spotted

The T-Rex is the most relevant because it's a 3D bench:

>1080: 12.876 fps
>1080ti: 19.773 fps
>vega: 14.784 fps

Other urls found in this thread:

compubench.com/compare.jsp?benchmark=compu15d&did1=26470338&os1=Windows&api1=cl&hwtype1=dGPU&hwname1=AMD Radeon R9 Fury X&D2=NVIDIA GeForce GTX 980 Ti
gpuopen.com/professional-compute/
twitter.com/AnonBabble

WOOOOAHHHHHHHHHHHHHH

If AMD aims the price to be similar as a the GTX 1070 (maybe between the 1070 and 1080) consider me sold... Considering both 1080 and 1070 just got a price cut.

I don't want to leave my Freesync monitor

Finished and bankrupt?

>The T-Rex is the most relevant because it's a 3D bench
How is it relevant when Vega is a completely new architecture that has 50% higher fp8 performance which that bench isn't optimized for?

>15% higher results than the 1080
>80% larger die size
>HBM2

There is no fucking way AMD is going to be able to compete with Nvidia with this.

With a die size that big it's going to be very expensive to make, definately with HBM2.

>80% larger die size

Is that a good thing or a bad thing?

Bad for production costs.

The larger the die, the less you can fit on a wafer, which means more expensive to produce

WHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA

At what Price would you say its a good deal. Assuming if OP's benchmark is valid?

15-25% higher than the 1080.

But it might be even more expensive to produce than a 1080ti

Performance obviously doesn't scale linear with price.

And even then, I'm not even taking into account stuff like power draw and heat output.

Which will most likely be significantly higher than the 1080 with a die size that big

1000mhz and 1200mhz boost?
This is clearly the low end model. Perhaps a "Nano" model that's under 175w TDP yet still outperforms the 1080.

The high end model is at least 1436mhz.

Probably something for mITX builds that will sell like hotcakes thanks to that.

Those clockspeeds make no sense, that's not a 12.5tflop chip with 1200mhz.

we'll have some room at reduced margins but you're most likelt correct. 50+/- aimed at the 1080ti price point is what I'd bet. Pleasant surprise if slightly bellow 600 and i'll the praise the lard if it's hovering around the 500's.

that boost clock will make many a fanboy unpleased. I myself was putting some faith into Adored's comments on speculative clocks.

wait raja told us that the base clock will be 1500mhz and it shows 1200mhz..

oh this is going to be good lol

>OpenCL benchmark

As I already said: because what is seen there is likely a R9 Nano successor.

They are likely launching at least 2 different RX Vega cards.

i'll clean Raja's loo if it's 499,99 - delivering on his wildly speculative, informal off-camera, don't take it home, en passant, whisper of a high end desktop at 1000 bucks. Ryzen 1800x + Vega flagship could technically fulfill that in the hearts of fanboys.

Vega Mi25 instinct is clocked around 1520mhz, this is 1200mhz.
So either someone was testing IPC or this is an ES without final clocks

kek

just to let a hint regarding my promise of heavy duty cleaning.

Cuck

this is a 4096 monster it cant be a small one

I have a nano, and it has the full 4096 core count of thr fury x. It's just down clocked for consumption

compubench.com/compare.jsp?benchmark=compu15d&did1=26470338&os1=Windows&api1=cl&hwtype1=dGPU&hwname1=AMD Radeon R9 Fury X&D2=NVIDIA GeForce GTX 980 Ti

your point?

kek at your sad life of SS'ing user comments. I could understand the satisfaction of making someone eat their words if there were handles, but to save a copy from a random user, sometime, somewhere, somehow said something silly is pathetic.

if vega comes out around or under $400 its a victory if its in the 600 range its DOA

and for all you morons out there spending $700 on a 1080ti enjoy playing AAA games at 4k for about a year then you will have to lower settings to low just to stay above 60 fps

4k wait for volta
1080/1440p 144hz vega will be good for years to come

> 1700X - $349
> Vega Nano - $399
Total: $748 for top notch, 8-core/16-thread hardware

This is the best timeline ever, meanwhile at Intel/Nvidia:
> i7 7700K - $349
> 1080ti - $599
Total: $948 for a 4-core/8-thread rig;

Thanks AMD.

Yeah, this Amdrone faggot was so funny.

you are overrating how fast gfx cards losses its performance, i can still play most games at medium at 1080/60fps with my 670gtx after 5years. Yes this card could max any game at 1080/fps at release.
Yes 1080ti wont be playing 4k/60 for 5years. But it will support highest settings for atleast 2-3years at 4k. Why? Because consoles.

*it's 1048 user. add another 720 bucks if you're a real good goy buying the greatest from ROG.

I'll lick Raja's loo clean if it's 399,99 for the top end model.
For realsies this time.

500-600 is a more realistic price...

The low end model was rumored to compete around 1070 level. If you look around for the device ID mentioned in the article you'll find that it's the big one too.

Or a 1400X + Vega Nano mITX PC.
Under 250watt total power consumption at load, under 30watts at idle.

I'm going to have to build 2 PCs with all this $AMD stock and all the good options AMD is giving.

Maybe $450 for the Nano and $550-$650 for the top end card if lucky.

The top end card is likely going to roughly match the 1080Ti in un-optimized games and shit-stomp it in optimized ones like Prey.

Shit rumors from currytech. Disregard.

>if vega comes out around or under $400 its a victory if its in the 600 range its DOA

simply because you are moron the world doesn't need to accomodate your retard obtuse ways. If it launches at 800 dollars while destroying the 1080Ti it'll be a success by any metric, including perf/dollar.

I hope it comes out at better pricing scheme, but i don't think it would be very smart from AMD to do charity for you, personally (you).

People assume it competes at a certain level due to the TFlops or memory bandwidth. You have to factor in the wildly different architecture.

> 1400X
no, get the 1600X

Nothing relevant uses OpenCL and that benchmark is obviously worthless, if it gives a Fury X and a 1080 at the same score

AMD pajeets on suicide watch

The Fury X is a monster at compute dumb shill. most OpenCL benchmarks aren't relevant for muh gaems, but AMD's intention with Vega is to rape Nvidia and Intel on coprocessors/GPGPU as demonstrated by the Radeon Instinct event

it's probably safe to assume AMD filling the price points at 100 increments
rx 580 - 250
rx 590 (?) - 350
Rx Vega smallest - 450
Rx Vega mid - 550
Rx Vega biggest - 650

Are you stupid? If Vega is behind the 1080 ti, the card is DOA

With a whole CCX disabled, the 1400X will be lower power for miniITX.

RX580 will probably be $200, though.
And RX590 won't exist.
You failed.

Are you stupid? The Benchmarks gives the Fury X and 1080 the same score.

too bad amd doesn't provide software support like cuda
they shit their turd and leave you to deal with it yourself

>rx 590 (?) - 350

RX 5xx are confirmed rebrands

or ditch the 590 replacing it with the smallest vega and cut the line at 550.

Last AMD gpu i had sounded like a vacuum cleaner, heated better then my oven.

Have they improved?

Jesus...look the T-Rex

>Shit rumors from currytech. Disregard.
Of course, I'll disregard the AMD demo with a 8+6 pin card too, because a fanboy on Sup Forums told me so.

Yeah, you do of course. Not to mention the performance we saw from AMD's own demos put it closer to a 1080 than a 1080 Ti/Titan XP.

isn't the larger die size due to hbm2???

sounds like you had a gtx 480 than a amd

No, nobody is talking about the size of the interposer or of the HBM2, it's about the actual GPU.

>its a Sup Forums speculates and gets it totally wrong again episode

Yawn

avoid reference designs, nvidia isn't much better with stock coolers either

But they do

7870

Vega is 45% more raw FLOPS than Fury X and that's before you get into all the architectural improvements.
A benchmark that shows it to only be 18% faster than Fury X is obviously wrong.

your kidding right 7xxx series is known for the low temps
in fact some HIS cards had a max temp of 62c at full load..

45-50 in idle with idle fan being around 40%
70+ easily

It died after about a year

Lent a 970 from a buddy of mine after that, felt fucking amazing having a card not using its fans until it hit 60c

In idle it went down to 30c

Now i got a 1080 which cools pretty similarly.

Nshitia btfo

DEELET THIS

>Vega Nano - $399

good luck

Small one as in weakest cooling which would necessitate a lower clock on the core and possibly memory, it also might have a less clean power delivery which also hurts its clocks

>Vega Nano - $399
See, we know the 1080 is priced stupidly above of what it's worth, but there is no way a fat chip like Vega will be $400. Not only that, it also has expensive HBM2 so expect prices starting at $600. I'd love to get one, but please be realistic of prices.

>realistic prices
margins are inflated as fuck. there's nothing realistic about current prices for hardware. No one can find exact figures for the total production costs for cpus and gpus, since they're all guarded under trade secrets and the best we can do is extrapolate from the old production costs for the virgin wafers and wildly add this and that to come up at a guesstimation.

>b-but that tech guy wrote that it costs 398,972572 burgers and 3 dead chinks to produce a titan xp.

Guess what, he took that figure straight out of his smelly ass, apart from the chinks; he totally went to visit the mass graves.

The price finding mechanism for hardware manufacturers has no concern for supply, competition, logistics, they go straight to end user market perception of value focusing on how much they can rape you before you start to bitch - hint, your a big boi, you can take it real good and a lot of it.

They do now.
gpuopen.com/professional-compute/

Usually costs are high in R&D but Pascal cost almost nothing in R&D. Especially not the 1080Ti.

The cost of the cards themselves isn't very much. In the $30-$150 range, I'm pretty sure. There's no way even the most high end consumer card costs over $150.

And Titans are one of the cheapest cards to make since they're made from garbage chips that would have been trashed and high margins are already made on the server cards they are garbaged from. The only reason they cost $1000 are to keep them from being bought instead of Teslas and shit.

Looks like the same disaster as the Fury X back then.

B-b-b-b-b-but wait the new drivers

Yeah ok but how stronk is the 580

% larger die size
why would this be a problem?

...

I know it's a rebrand, that doesn't tell me its actual specs

It's literally a 480.

It was between 1080 and 1080 Ti on a 1.2 GHz clock, but enterprise release clocks have already been promised at ~1.55 GHz due to the 25 TFLOPS fp16.

This thing will completely rape the 1080 Ti if the leak is accurate and consumer clocks are roughly in line with the enterprise parts.

Needs 1436mhz to hit 25 TFLOPS half precision, no...?

>How is it relevant when Vega is a completely new architecture that has 50% higher fp8 performance which that bench isn't optimized for?

fp8 isn't a thing. you've thinking of fp16. there's also no real world application that uses half precision floats outside of a few niche applications for ai/machine learning stuff.

exactly.
Most people neglect logistics costs, for instance. The final impact on pricing is a big factor when you're shipping worldwide out of a single factory in Taiwan. But that becomes moot with these guys. We could very well have 500 dollars top tier cards (consumers, server grade is another story entirely) while "they" could still maintain very decent profits per unit sold. But why the fuck would they even bother doing the math, the focus groups or hiring consultants on price finding, when their target audience actually have a passtime of calling other folks too poor to afford their flagships?

the titan price finding probably went like this:

>...so there's that. How much can we charge for it?
>800?
>nah, I feel we can go higher.
>1200?
>hm... maybe, let's try it a notch down.
>a notch? 1000.
>k, let's do it.

straight out of Wharton guise.

Colors in every fucking game are fp8 or fp10 you stupid fucking piece of shit go kill yourself

No, it needs 1525mhz

that doesn't make them 8 bit floating point numbers.

That's up for the driver and shit to decide. The texture format is 8bpc. Normally the driver may convert back and forth 32bit floats but it sure as shit doesn't have to.

Nvidia fancies itself as luxury brand, which takes into account that its bottom line must be at least at such a height the perception of value amongst its wealthy consumers doesn't start to decline. They would lose customers if they started to put Titans on the bargain bin.

ps That's messed up, because they aren't selling 50k watches nor 20k purses or shoes. This goes against mass production of goods, it should, at least in principle.

of course, no need to even touch the inherent value of a 50k watch, or designer haute couture. The very nature of such items antagonize mass production mechanics.

Funny enough the Fury X was sort of an attempt at "luxury". It shipped with an AIO.
But it wasn't good enough quality or quite enough performance. But it's the sort of thing you'd expect with the absurd price and lack of performance in Titan cards.

Hope Sup Forums doesn't fall for the, year of the AMD meme again

That's my take on it too.
Monkey see, monkey do.
Although my criticism was glaringly aimed at Nvidia I never said that AMD, or Intel, as a matter of fact, don't follow the same rules. Well, AMD attempted to sevral times in the recent history and never managed to catch that halo (the last possible item to attain was maybe the 6990, but that's debatable for sure).

If that's the sate of the markets, the competitors tend to protect and reinforce the staus quo that is so sweetly beneficial for their final margins. Not to mention that Nvidia was protected (because of the nature of the market) from AMD even trying to perform a aggressive dump strategy, my opinion is that they would've shot themselves int he foot if they tried it.

The thing is going to run hot as fuck, isn't it?

No, the top end server vega is supposed to be 25TFLOPS fp16 and passively cooler. So 220w TDP or less, I'm pretty certain.

I've noticed AMD meme'ing crossfire a bit more lately. I wonder if this will be a good thing or a rerun from the Rx480 launch.

Absolutely fucking clueless retard here. Looking at the reviews of it people are saying that it tends to get quite hot while working. How big of a deal is? Would it cause problems and should I wait for the better cooling versions to come out instead?

>amd
>gpu
... Since when has amd branded their gpu's amd instead of ATI?

The RX 480 is 45%~ behind the GTX 1080, thats with 36CU clocked at a peak 1266mhz.
Vega 10 has higher IPC, 77% more CU, and is clocked over 20% higher. It'll slap the shit out of the GTX 1080ti.

The card in the OP has its max frequency listed as only 1200mhz, its a test sample not running at its full clock. To make 12.5TFLOPS Vega 10 is clocked at 1526mhz.

2006 i believe

Modern graphics cards have a target temperature you can set, it will basically run as fast as it can while obeying your temperature, power consumption, voltage and clock settings. By default it seems to run at 84C. You can easily increase fan speed to bring the temperature down if that's what you want to do.

But yes, waiting for custom cooled cards is often the best thing to do in general. They come with the downside of dumping heat inside your case instead of straight out the back, but that generally only matters once you get to using 2 or more graphics cards, doesn't matter much with just one unless you want to have an especially small and poorly ventilated case, in which case it may very well help.

Is this Vega 11 or 10??!?