Leaked Vega Benchmarks

Could be real, could be fake. You decide.

Other urls found in this thread:

videocardz.com/71771/final-radeon-rx-vega-64-3dmark-performance
videocardz.com/71280/amd-vega-10-vega-11-vega-12-and-vega-20-confirmed-by-eec
twitter.com/SFWRedditVideos

how is the 1080 Ti such a big jump from the 1080 anyway?

Moar ALUs.
Like 1k more ALUs.

hahaha
no.

>Wanting the competition to fail
>Not wanting a competitive market

ISHYGDDT.jpg

The FE versions supposedly throttle on air cooling. They also cost more than a 1080 and take more power.

kys nigger poorshit

Reminder to report shitposters.

The 1080 Ti is a fucking monster. I immediately switched to it the instant I got my GTX 1080. I don't regret it at all. I'm going to stick with my Ti for a while.

>Wanting hyped up products that can't compete be shilled as if they were good.

Lol, still slower than a midrange 1080? Fucking AMD is a joke

>1080
>midrange
?

Titan - Ultra high end
1080ti - high end
1080 - mid range

?
You've been drinking JHH's sperm for far too long.

1060 is mid range.

Oh good it's this retard again. Tell us again how 1060 is for poorfags despite also costing hundreds of dollars.

elitist "pay say gay mers" are fucking cancer

>AMD literally can't beat Nvidia's midrange GPU after 2 years

*WHEEEEEEEEZE*
*WHEEEEEEEEZE*
*WHEEEEEEEEZE*
AHHHHHHHHHAHAHAHAHAHAHAHAHAHHAAHHHHHHHHHAHAHAHAHAHAHAHAHAHHAAHHHHHHHHHAHAHAHAHAHAHAHAHAHHA

>1927mhz core
>1100mhz HBM2
wew lad

>Clearly is slightly faster than a 1080
>slower than
>midrange

What's the point when it's bandwidth bottlenecked?

You realize the Titan is basically just a slightly, keyword being slightly, better binned 1080 Ti right?

>that bar


DIAL IT UP TO CHINK WANG FANG, WE'RE GOING NUCLEAR

Biggur numbers.

>bandwidth bottlenecked
Which bandwidth?

Memory.

Memory.

Titan has GDDR5X , and more cuda cores

actual benchmarks

how can AMD even compete?

You realize that there are literally 6 GPUs faster than a 1080? P100, V100, Titan Xp, Titan XP, P6000, 1080ti , if anything the 1080 is at the very bottom midrange card.

Literally DOA

1080ti also uses G5X.
Now that's what i call mental gymnastics.

Gonna get the 56 and overclock that nigga if this is true.

From (945Mhz) 483.8 GB/s to (1100Mhz) 563.2 GB/s I think that allows a little more bandwidth.

Not enough to justify the cost in any capacity. In addition, the recent Titan X(Pascal) and Titan Xp fiasco has killed any smart consumer's trust in the Titan brand. Even Nvidia has no fucking clue why people buy Titans anymore, it used to just be what they were considering "absolute best we can physically make, but at a cost" but the release of the 1080 Ti and how it performed better in most games meant the people who paid almost double for the original Titan Pascal got screwed over.

The extra CUDA and GDDR5X don't really help the people who'd want to buy a Titan when performance can't even be guaranteed to be better than the less expensive card anymore.

The other user should post effective BW numbers.
That OC will result in practiaclly nothing.

Need re-tests on memory bandwidth when it launches.

>53 C at 1927mhz
W E W

G*rmans will do that.
Hopefully.
The rest of "tech press" forgot B3D suite ever exists.

Vega 10 has 20-30% regressions vs. Fiji in raw memory bandwidth, effective texture bandwidth, and texel fill rates.

The real question that remains to be answered is whether the absurdly low memory bandwidth can be fixed with BIOS updates, because as it stands right now, Vega 10 is so bandwidth starved that Vega 48 with 945mhz HBM2 would be just as fast as Vega 64.

>literally a 300W 1080

WEW

LAD

I'd speculate that this is where they actually do place, it makes sense from a corporate and engineering standpoint. While a lot of people do in fact have the 1080ti, and it beats out the whole of the market, there are plenty of people (read majority) of people playing at 1080, and looking forward to 1440, as such it makes sense that they would offer a GTX1070 competitor and a GTX1080 competitor. Consider the market share that it offers being slightly better but similarly priced. Especially the 56 since it will be a key consumer card and directly parallels the 1070's price.

Looking forward if AMD adopts nVidia's release pattern it may indicate they too will release something kin to the ti in the near future, which will be an interesting battle.

I'd say people that bought titan xp hardly got screwed by release of 1080TI, sure the card has basically same performance at half the price, but also came out like 8 months later ? If you are the target market for titans, the extra 600 bucks you pay for "early access" is completely worth it, and not like 1080 ti is faster, it's only cheaper.

D O A
O
A

It's all about target markets, not performance.

A Vega 64 Liquid hit 25,000 this morning.

That's shit.

There should be fappening style leaks for hardware
Or some reviewer just making it stupidly easy to get their benchmarks early and then claiming to be hacked,

I was considering getting one but since I have a 1080 SLI since launch going for one ti is kinda of downgrade. Majority of titles iolay support sli

t.Jensen

Breaking the NDA is a bad idea m8.
No, it's shit, given the power consumption.
Perf/watt is atrocious (in gayms, that is).

>Titan has GDDR5X
so has the 1080Ti and the 1080

>aka your avg astroturfing stock shiller.

AMD NUMBA ONE!!!!!

It's not willfully breaking NDA if you are "hacked"

You're not beating their legal department, dumbo.
You're not smarter than a 15 page clause

I know, 3DMark is my favourite game and if Vega does not show well there, then I don't care how it performs in none synthetic situations.

3DMark is a good indication of gaymen performance.

Maybe for a certain balanced game engine.

>Still posting Vega FE benchmarks a day before launch
Trying to get your last few shitposts in? videocardz.com/71771/final-radeon-rx-vega-64-3dmark-performance

Link pls

Not since about 2001

That's barely better.
It's still abysmal performance for 486mm^2 die.

I'm pretty sure they consider NDA to be a strict liability thing, it won't matter if you personally released the benchmarks are not. Ultimately they just won't trust you because you will have partially fucked up their marketing strategy and will have indirectly caused them damages.

The FE is better than the RX

If true, Vega 56 best card
> 1080 performance
> 1070 price

In reality

>1080 price
>1060 performance

1070 MSRP is $350. What you're seeing is inflation due to miners. Once initial stock is gone, expect Vega 56 to go up to even higher than a 1070 in price.

t. 815mm^2 die man

>It's still abysmal performance for 486mm^2 die.
Ah, I see you are still pretending to not understand that AMD is segmenting by memory bandwidth rather than die size? When a four HBM2 stack version comes out and trades blows with full GP102 in games (as it does now in geometry performance) what will the implications be for perf/mm^2 hmm?

>Titan Xp price
>GT 710 performance
>hurr durr look how funny I am hysterical laughter all around

Kepler Titans were literately "Quadro/Telsa" rejects. They were a cheap way for general compute and prosumers to get near-Quadro/Telsa-tier hardware.

Nvidia gimped their Maxwell and Pascal Titans because it hurt their Quadro/Telsa margins.

>4stack Vega10
Hooo boy $1000 MSRP?
The BoM is going to skyrocket.

At that point you're not gonna be able to buy cards so that's pointless.

no, 1060 is entry level.

>four HBM2 stack version comes out and trades blows with full GP102
And when exactly are we expected to see this? And why would they design their entire GPU around memory bandwidth that isn't there, lowering yields in the process? Sounds stupid.

I honestly don't get what you gain from this shilling. A competetive market would benefit even Nvidia fanboys.

If this is correct then Vega is a giant step down from the 480, and the 480 actually is close to the 1080 in crossfire.

However, where would this chart put the 1080 Ti and the Titan?

Late year or Q1

videocardz.com/71280/amd-vega-10-vega-11-vega-12-and-vega-20-confirmed-by-eec

Also you need to redesign Vega10 with 4 HBM PHYs in mind for 4-stack configuration.
Vega10 is DOA.
Maybe Vega11 will be less shit.

Vega20 is FP64 card.
It's not a bigger Vega10, user.
It'll never be a consumer SKU.
Kill yourself my dude.

>And why would they design their entire GPU around memory bandwidth that isn't there
Because they wanted to use the same GPU for RX/WX/SSG/MI and for all of those segments except gaming vs a 1080ti, two stacks was enough.

Pic related. In pro rendering applications where Vega 10 isn't memory bound it trades blows with full GP102.

>1050 is entry level
>1060 is low-mid end
>1070 is mid-high end
>1080 is high end
>1080ti is ultra high end
>titan xp is rebranded workstation card with virtually no benefit over the 1080ti in the gaming department but 200 dollars more expensive

>And when exactly are we expected to see this?

It's Vega 20, coming next year, meaning it'll be on the market in 2 years.

They could've made Vega10 with 4 stacks in mind from the get-go.
Vega10 is a bottleneckfest.
It's almost as bad as nv30.

>According to the list from EEC, Vega 20 is currently only expected as GL variant, which means Radeon Pro/Instinct series.

It's not a gaymur card. Although 32 GB RAM sounds crazy.

All previous "FP64 cards" from AMD had a consumer version, so no, Vega 20 is likely to end up in this space eventually.

All previous FP64 cards from AMD had reasonable BoM.
Vega20 is gonna be bigger.
Much bigger.
FP64 is not free.

>Vega10 is DOA.
This is idiocy. SSG locks down a whole segment unchallenged, WX is a solid competitor for the P6000, and FE (with updated drivers) will be excellent an excellent OG Titan type card. Vega 56 > 1070 and Vega 64 > 1080.

The only segment Vega 10 isn't great in is vs. 1080ti in gaming.

>titan xp is rebranded workstation card with virtually no benefit over the 1080ti in the gaming department but 200 dollars more expensive

Titan X (pascal) =/= Titan Xp

>This is idiocy. SSG locks down a whole segment unchallenged, WX is a solid competitor for the P6000, and FE (with updated drivers) will be excellent an excellent OG Titan type card.
These are hilariosly niche.
Like no revenue-tier niche.
>Vega 56 > 1070 and Vega 64 > 1080.
This is what matters.
And it's BAD.
VERY FUCKING BAD.
Gaymen perf is everything.

>They could've made Vega10 with 4 stacks in mind from the get-go.
Two stacks was enough in every market they wanted to compete in with Vega10 except one. They already have Vega10 x2 on the schedule for 2H 2017, should be easier to go to 4 hbm stacks than doing a full on dual GPU card (which there would be no point in doing with Vega10 anyway).

4 stacks require GPU redesign, user.
Vega10 is DOA for gaymen.
Good look to RTG designing a gaymen chiplet for Navi.

GCN ALUs can all do FP64 natively at 1/2, they do cost more registers and power that's why you don't see them.

Unlike Nvidia, which needs dedicated FP64 units independent of FP32 units, said FP64 units cannot do FP32 or FP16 like GCN can.


In short you're making assumption based on limited knowledge, fuck off.

Reported for shitposting, your posting style is too transparent, reminds me of a certain forum nigger.

>These are hilariosly niche.
>Like no revenue-tier niche.
Yeah, it's not like every studio in Hollywood could use these. Oh wait. They are also $6,999 a card.

>4 stacks require GPU redesign, user.
They are already doing a four stack design anyway, they can just piggyback a four stack RX version off that work.

>expecting amdrone to understand chip design and on die memory controller configuration

kill yourself

Report a report is a report newfag

>Yeah, it's not like every studio in Hollywood could use these. Oh wait. They are also $6,999 a card.
That's still no volume.
Volume > margins.