Vega56 VS GTX 1070 benchmark

>>61889303
Oi WUT?

>>61889303
>20% difference vs. partner GPU
Doesn't seem likely.

>>61889420
Reminder to report shitposters.

>>61889303
Based on the microscopic amount of information we have so far, it could be. Seems like the Vega 56 is a much more of a midway point between the 1070 and 1080 than a direct competitor to the 1070 in performance. The missing 8 compute cores probably won't hurt it that much in games and the lower thermals and TDP mean it should run cooler as well. I don't see why it couldn't be a great overclocker too if the Vega 64 is clocked the way it is. Vega 56 seems like a GTX 1070 killer, Vega 64 not so much a GTX 1080 killer due to the power consumption.

>>61889688
You literally need to OC HBM to get it to 64 perf.
Also he locked the video, he broke the NDA lmao.

He removed the video it seems and probably broke the NDA.

>>61889719
>>61889797
Well if it was real .. Vega 56 might be worth it. Vega 64 is probably still not, though.

>>61889797
He obviously broke the NDA.
Notice the WX9100 on his table.

>>61889303
So 56 is great and 64 is trash?

>>61889800
It means they fixed the bandwith bottleneck. 56 performs like that with silly low clocked memory.
Even 64 is worth it now.
>>61889822
Both are great assuming Vega10 is no longer BW-bottlenecked.

>>61889822
it s not that it s trash , it s cause NVIDIA with the 1080ti basicly went light years ahead , and they still use gddr5x ... AMD is 1 year behind

File: o0700100012696567516.png (127KB, 700x1000px) Image search: [Google]

127KB, 700x1000px

>>61889303
I want to believe.

>>61890352
>believe
It was on the video.
OC the memory and it's going to be even faster.

>>61889303
post mirror

File: Radeon-RX-Vega-56-1440p.jpg (92KB, 955x602px) Image search: [Google]

92KB, 955x602px

repeat after me:
1070 performance
https://videocardz.com/71953/are-these-radeon-rx-vega-56-benchmarks

>>61890769
>repeat after me:
Vega 56 is slightly faster than 1070 performance over 25 games

>>61890794
That's hilariosly bad for 3.5k ALU card with a fuckton of uarch improvements.
Bigger Polaris would've performed better a year ago.

>>61890769
Looks to be a great 1440P GPU and since that is what I have I think that is what I am going to get (Miner fags permitting).

>>61890805
I actually wonder why they simply didn't go with a die-shrunk Fiji-XT with GDDR5 memory controller... The clock difference alone would probably put it against GP104.

>>61890891
Reasons?
Who knows.
It would definetly be better (and earlier) than Vega.

>>61890891
RTG don't have the resources so they had to go with one GPU and that was aimed at the pro market. RX is an afterthought.

>>61890944
Making bigger Polaris would've been silly easy. It would also launch a whole fucking year before the abortion that is Vega.

>>61890805
The theoretical product from my brand is better than the actual product from your brand.

>>61890957
The money is in pro machine learning cards. Just about every auto manufacturer is looking at nvidia right now for self driving car research. AMD needed to get back in that market.

>>61891050
>The money is in pro machine learning cards.
There's none considering meme ASICs are coming.
>Just about every auto manufacturer is looking at nvidia right now for self driving car research.
Even less money and even more memes.
No, AMD needs to make good iGPU.
That's it.

>>61889303
Where do you buy these cards at release?

Do they just show up at ebay/amazon/newegg or can I order them somewhere before they run out of stock?

>>61891124
>Do they just show up at ebay/amazon/newegg or can I order them somewhere before they run out of stock?
Yes.
But they won't run out of stock, Vega is bad at both gayming and mining.

>>61890891
The Vega was on development ages ago. Raja said that it started before her joined the team so it's not his architecture, he only finished it.

>>61891149
>Vega is bad at both gayming and mining
>and mining


>RX VEGA 64 MINING EDITION?
>It was confirmed by OCUK’s Gibbo that
>“Seems the hash rate on VEGA is 70-100 Mh/s per card, which is insanely good.

>>61891151
Bigger Polaris would still be better and cheaper for gaymen.
Consumer Vega is a total disaster.
>>61891236
Shrout says it's the same as FE hash rate.

>>61891243
I'm saying that they might started polaris and Vega at the same time. They didn't think that Vega would be such a failure.

>>61891327
You can always can the design if it's bad.
They could've also launched 64CU Polaris.
It would performs the same as Vega.
Heck, probably better.

video is delet

>>61889303
No video?

>>61891357
i still believe they are sandbagging. They'll bring down the power usage while keeping the computing power and put two on a card with IF. But nvidia needs to get comfy and fuck up their next generation first.

Anyone managed to mirror the vid?

>>61891432
>sandbagging on launch day
?
And AMD never sandbags anyway.
RX Vega is DOA.

>>61891460
It was only up for a few minutes. I saw it, though.

>>61891467
Keep wasting your time nvidshill, no one will ever care.

>>61891475
>20% ahead on the cut-down version is "struggling"
You realize this is Vega 56, not Vega 64?

>>61891504
You do realise it's 3.5k ALU card?

>>61891484
Was it just the benchmark result?

>>61891504
https://videocardz.com/71953/are-these-radeon-rx-vega-56-benchmarks

STAY MAD, AYYMDPOORFAGS

>>61891467
i mean they sandbagged until now.

>>61891475
>>61891506
Did you really just autism so hard over your phones auto correct that you deleted and reposted?

File: 1439137596524.jpg (23KB, 481x368px) Image search: [Google]

23KB, 481x368px

>>61891523
I'm alright with that

>>61891523
Why would I be mad? Vega 56 is 400 dollars while the 1070 is 450-500.

>>61891523
>25 games avarage
SOPAq

>>61891523
Someday we might find out how to cure this kind of autism.

https://rbt.asia/g/search/text/%22AYYMDPOORFAGS%22/
https://rbt.asia/g/search/text/%22AYYMD%20HOUSEFIRES%22/

>>61891560
>launch day
>3.5k ALU card performs the same as heavily cutdown GP104
>sandbagging
?
>>61891570
>acceptance
Good job.
Lower your standards.

File: stfu.png (6KB, 560x86px) Image search: [Google]

6KB, 560x86px

>>61891475
>>61891506
I saw your embarrassing spelling monkey, just stop

>>61891510
>3.5
DOA

>>61891593
There is no bad card, there is only a bad price.
Faster you learn that, better it be for you.

>>61891593
>Same performance = lower standards
Ok. The power draw isn't as much of a dealbreaker as it is with Vega 64.

>>61891622
It's a bad card with a bad price.
Fucking hell 1070 has $349 MSRP, it can be OCed and it's more efficient.
>>61891637
It's still 210w vs 150w.

>>61891593
It's today your first day in the internet? There were dozens of "leaks" in the past. We had tonnes of info about these cards a half year ago that created false expectations.

>>61891653
>1070 has $349 MSRP
Not anymore, nvidshill.

>>61891653
>1070 has $349 MSRP
Good luck finding it at that price. Good luck finding Vega 56 at all, after it launches.

>>61891668
Why would anyone even buy 56?
It's shit at mining.
And shit at gaymen.
What's the purpose of it?

>>61889303
Is it just fantasy?

>>61891677
You keep saying that, but do you have [citations]

>>61891677
>shit at gaming
it matches the 1070 right there
that's not "shit at gaymen"

>>61891677
NVIDSHILL ALERT
NVIDSHILL ALERT

>Struggling against 1 year old GTX 1070 while consuming 210W

TOP KEK

>>61891700
>3.5k ALU card barely matches 1070 in gaymen
>not shit
?
Just buy a fucking 1070.

>>61891718
it'll be crippled in a few months

>300W AYYMD HOUSEFIRES losing to 180W GTX 1080
>210W AYYMD HOUSEFIRES losing to 150W GTX 1070

AYY

>>61891718
>3,5k ALUS!! THE ALUUUS
Yep, it has 3.5K alus and guess what it's a big chip.
You know who gives a shit about that? Nobody.

>>61891726
Maxwell was not crippled.
Please stop your delusions.
Vega is late, hot, big and underperforming.
There's no magic drivers.
Nothing.
They conceded one generation to nVidia and about to lose the next one!
Formiddable efforts, RTG.

>>61891677
>shit at gaymen
>>61891709
>Struggling against
>>61891729
>losing to
The damage control shilling has already started and the card's not even fucking out yet.

You know AMD might be onto something when the shitposting intensifies

>TFW to intelligent to use GDDR5/5X

>>61891748
People already benchmarked that.
It's 1070 perf.
>>61891736
3.5k ALUs are supposed to perform much better, given the uarch improvements.

>>61891757
>3.5k ALUs are supposed to perform much better
It does, it performance like magic in where ALUs matter. That aint games.

>>61891710
Founders Edition

>>61891746
>losing to over year old cards
>damage control
madtards are special kind of stupid

>>61891570
How is that acceptable?
Preforming on par with a 1 year a old 1070, while consuming 2x power
The minute miners get ahold of it, prices will go up 30%+ over MSRP
Putting it at gtx1080 money

We all know AMD fans don't have money.
If they had $500+ to spend, they would have gotten a gtx1080 or ti a long time ago.

https://www.reddit.com/r/Amd/

AYYMD shilling central already imploding from the disappointment, top kek

>>61891769
>given the uarch improvements
It's like you can't fucking read.
Vega was supposed to remove inherent GCN bottlenecks so it would be able to fucking saturate the ALUs.
And well fuck me it brought MORE bottlenecks.
Bravo RTG!

>>61891778
It's acceptable because I don't have to give money to nvidia. And that's priceless.

>>61891775
Even though it was just show to be pretty much dead even? Yeah, you're a fucking twat. https://videocardz.com/71953/are-these-radeon-rx-vega-56-benchmarks

>>61891779
I feel bad for them

File: AMD-Radeon-RX-Vega-56-Prototype-4-1000x493[1].jpg (133KB, 1000x493px) Image search: [Google]

133KB, 1000x493px

https://videocardz.com/71958/prototype-radeon-rx-vega-56-pictured-with-three-fans

You're a big heatsink

>>61891778
>while consuming 2x power
40% more power is not 200% more power, you are bad at math. Also, NVidia calculates their TDP differently anyway.

>>61891804
FOR YOU

>>61891510
this, i really wonder what the 10.5Tflops are doing

>>61891804
40% more power, 40% more heatsink. Seems legit.

>>61891812
>NVidia calculates their TDP differently anyway.
another amdtard lie that has been debunked countless times

>>61891785
What are you even talking about? Do you just like throwing random shit in the air and hope it sticks?

https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1070/22.html

Keep on being forever assblasted, AYYMDPOORFAGS

>>61891787
>Rewarding a bad company because you hate another.
I will understand you guys.

>>61891804
Holy cringe. If the partner cards don't come out early sept im finna just buy a 1070

>>61891822
Holy fucking fuck; Vega was supposed to bring higher perf/alu compared to Polaris, but it actually fucking regressed.
>>61891818
Choking some cocks.

>>61891812
Do you know even what TDP and power draw mean, retarded drone?

File: AMD-Radeon-RX-Vega-56-Prototype-9[1].jpg (417KB, 1600x1200px) Image search: [Google]

417KB, 1600x1200px

>>61891804
More

>>61891844
>Vega was supposed to bring higher perf/alu
Where the fuck did you read that from? It was supposed to bring high end GPU to the market.

>>61891841
Just buy a 1080 for 500 bucks
You can't do wrong

>>61891833
But I'm not rewarding nvidia, you know, the bad company that shits on you and commits actual crimes to fake benchmark results.

>>61891861
>WX9100
Neato.
I wish someone would benchmark it.
>>61891862
And it failed.
486mm^die that does nothing but sucking dicks.
Bravo RTG!
It's fucking worse than Fiji and that was a fucking bottleneckfest.

File: AMD-Radeon-RX-Vega-56-Prototype-6[1].jpg (258KB, 1600x774px) Image search: [Google]

258KB, 1600x774px

>>61891861
8 pin + 6 pin

Reviews tomorrow
Wait for Navi TM (wait for Volta)

>>61891864
Don't need a 1080. Man this crypto bullshit is truly the worst timeline to build a PC

>>61891882
>486mm^die that does nothing but sucking dicks.
Guess what, nobody should gives a shit about die size other than AMD, it's their money they are wasting.

>>61891897
Fucking hell, 486mm^2 die barely beats 314mm^2 one.
You do understand it's nv30-tier disaster, don't you?

>>61891882
>It's fucking worse than Fiji
Vega took route of pascal, less performance more frequency, worked in favour of nVidia.

File: 4f62a4f0-abcb-4045-b244-2de392f4054f_1399569625883.png (576KB, 1364x1808px) Image search: [Google]

576KB, 1364x1808px

>>61891861
>>61891894

File: Screenshot_2017-08-13-17-13-16-1.png (223KB, 1080x802px) Image search: [Google]

223KB, 1080x802px

>>61891812
Based on Vega FE, Rx Vega will use twice the power of a 1070
Any overclock and 2.5x the power is needed.

>>61891897
>hurr durr nvidia hueg dies lol lmao huang
>w-who gives a shit that amd big die is shit
You're desperate.

>>61891910
Because Maxlel was not a fucking bottleneckfest.
It needed moar ALUs and clocks, Pascal delivered exactly that.
Fucking hell Vega is so fucking bad it's a total disaster.

>>61891909
What does size of the die give you? What kind of information is there that you really hate the size of it? What can you gather from that so infuriating that you just have to scream?

File: AMD-Radeon-RX-Vega-56-Prototype-5[1].jpg (435KB, 1600x1200px) Image search: [Google]

435KB, 1600x1200px

>>61891894
8GB, XT, Vega 56

>MADE IN CANADA
A
FUCKING
LEAF

>>61891929
wtf i want my taiwan cards back

>>61891924
YOU'RE SUPPOSED TO BRING THE SAME OR BETTER PERF/MM^2 THAN COMPETITION WITH NEW UARCH.
NOT WORSE.
VEGA IS SO FUCKING BAD ITS HILARIOUS.
NUKE RTG CHINA PLEASE BEFORE ITS TOO LATE.

>>61891918
That is the cut-down Vega 56, not Vega 64.

>>61891923
So, Vega failed because it has more frequency?
it already has enough ALUs to compete in pro market why would they add more?

File: AMD-Radeon-RX-Vega-56-Prototype-1[1].jpg (364KB, 1600x1002px) Image search: [Google]

364KB, 1600x1002px

>>61891937
Friendship ended with Tsai Ing-wen
Justin Trudeau is now my new best friend

>>61891929
Did Linus personally fuck this up?

>>61891944
How delusional are you?
new uarch is supposed to improve old uarch and nothing else.
nvidia has failed to do any of your requests back in the golden age of ATi/AMD cards, don't see why you angry about it now.

>>61891966
Linus personally produces each and every Vega card

>>61891909
Holy fuck, are you really going to keep knowingly and deliberately shitposting about this constantly in every thread about Vega from now on on /g/?

AMD obviously decided to make one size of GPU for WX/SSG/FE/MI25/RX56/RX64 and to segment RX 56 and 64 against the 1070 and 1080 by memory bandwidth rather than ALU count. Calculating perf/ALU or perf/mm^2 on Vega10 is retarded because AMD is knowing wasting ALUs and mm^2 on the RX cards for economies of scale. That's also why the Vega 56 reference cards have the exact same design as the 64s (even down to the dual 8pin connectors).

>>61891929
I wonder what kind of cool shit you could do w/ JTAG access like that.

>>61891951
BECAUSE VEGA IS NOT SIMPLE FUCKING DIESHRINK, IT'S A FUCKING NEW UARCH.
THAT'S SUPPOSED TO BRING ACTUAL PERF/WATT AND PERF/ALU IMPROVEMENTS.
But it regressed perf/ALU wise compared to Poolaris.
>>61891975
>b-but nvidia
WHO THE FUCK CARES WE'RE TALKING VEGA.
Vega was supposed to fix GCN, not make it worse.
>>61891989
64 has the same peak bandwith as 1080ti, hell, it even has more ALUs.
YET IT PERFORMS WORSE.
WITH ALL THE UARCH IMPROVEMENTS AND MOAR ALUS IT MANAGES TO PERFORM FUCKING WORSE.
HOLY FUCK YOU NEED TO BE COMPLETELY RETARDED TO DESIGN A GPU THIS BAD.
Literally anudda nv30.

File: patchy4.png (275KB, 600x600px) Image search: [Google]

275KB, 600x600px

>mfw amdrones overtook appletards in the reality field distortion competiton

>>61892005
>IT'S A FUCKING NEW UARCH.
Nah, it's a Fiji it's not even new uarch it's old uarch with new shit added to it like new geometry engine.

>>61892018
Your days are counted. When Intel is finished we go after you. Cunt.

>>61892005
>Vega was supposed to fix GCN, not make it worse.
You keep saying this, but you can't provide one example of what is wrong with GCN, Frankly i don't think you even know what CGN is.

>>61892022
It's a new uarch, down to ISA and CU-level changes, anon.
But it's worse than Polaris perf/ALU wise hahaa.

>>61892005
dude go jerk off, your autism reaches new levels. you can thank me later for preventing yourself embarrass yourself here

>>61892034
>what is wrong with GCN
Shit geometry perf.
Shit per ALU perf.
Shit DCC.

Hold up, why does Vega have memory bottlenecks with HBM2??

the GTX 1080ti has just ddr5x and a smaller bus, but has no such issues.

>>61890769
>$50 more msrp
>performs the same as a 1070

>>61892048
only ALU has anything to do with GCN and if anything, ALU is only thing that works properly in Vega.

File: 900x900px-LL-689b3a99_6.png (124KB, 900x775px) Image search: [Google]

124KB, 900x775px

>>61892005
>64 has the same peak bandwith as 1080ti
64 is limited to the same bandwidth as a 1080 and performs like a 1080 as a result. Give 64 the same memory bandwidth as Fiji and it will crush a 1080. Coincidentally, I still don't believe that ~300 GB/s is all you can get out of two stacks of HBM2.

>>61892054
You do know that HBM2 and G5X work VERY differently?

>>61892035
>I know how Vega CUs are build
Do share

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

37KB, 711x533px

>>61892048
>Shit geometry perf.
LMAO

>>61892066
>Give 64 the same memory bandwidth as Fiji and it will crush a 1080.
But (surprise-surprise) it's not there.
And the cards are launched tomorrow.
>>61892083
So where's the gaymen perf impact?
Oh, it's not there.

File: Vega_FE_Memory_Bandwidth.png (129KB, 800x574px) Image search: [Google]

129KB, 800x574px

>>61892054
>Hold up, why does Vega have memory bottlenecks with HBM2??
THIS is the correct question, yes.

It should be self evident that RTG could not have intended this, and if it was a hardware issue there was PLENTY of time to do a respin, so WTF is up with this memory bandwidth regression? Everything about RX Vega 56 and 64 comes down to whether this can be fixed with BIOS updates.

what if it is just driver issues then

>>61892005
Indian codes makes drivers ....

>>61890891
https://youtu.be/ejVNhOvwtuI?t=962
Watch this part and you know exactly why.

>>61892091
>So where's the gaymen perf impact?
You keep inflating things you do not understand together thinking you have a point.

>>61892102
It's launch day.
They were supposed to fix everything for launch.
Yet it's not fixed.

File: k0gC6m5.png (53KB, 1029x744px) Image search: [Google]

53KB, 1029x744px

>>61892091
>So where's the gaymen perf impact?
>Oh, it's not there.
You said geometry performance, not gaming performance. Vega 10 has at least the geometry performance of full GP102, but as you already know it is memory bandwidth bottlenecked in games.

>>61891896
It's only a 100 bucks more and it enable 144hz and/or 4k gaming depending on the game
I run f1 2016 at 144hz 1440p and it's pretty nice
You don't really "need" a high end gpu anyway

>>61892068
>>61892102
Nevermind
Did some research, and the 1080ti actually has a higher memory bandwidth than Vega.

I assume that's just Nvidia using their shit more efficiently?

>>61892124
Whether it IS fixed by launch is less important than whether it CAN be fixed by BIOS updates. As long as it CAN be fixed by BIOS updates then it WILL get fixed sooner or later.

>>61892137
>I assume that's just Nvidia using their shit more efficiently?
Oh yes, nVidia's DCC is unrivaled.
>>61892143
>sooner or later
Fucking hell, LAUNCH is everything for GPUs.
No one cares how it would performs in a year.

>>61892137
nvidia cards run at their 100% like in last gen, while amd cards don't. nothing new.

>>61892102
>that 1 trick anon that hangs around hoping someone mentions memory bandwidth so he can post it again

>>61892160
Meh that big 200% perf upgrade on the titan recently make me really doubt this statement

>>61891897
Die size tends to go along with power util. So, I tend to give a shit about how much power my vidya wastes.

486mm^2 die getting whooped by a 314mm^2 die is nothing to sneeze at.

>>61892102
Vega has less memory bandwidth than Fiji, that was confirmed in the official slides. 484Gb/s vs 512 Gb/s

Typical retarded and poor redditor
https://www.reddit.com/r/Amd/comments/6ti2rq/im_going_to_run_vega_64_on_a_480w_psu/

File: 900x900px.png (379KB, 900x506px) Image search: [Google]

379KB, 900x506px

>>61892102
Why is vega so slow in this test?

>>61892137
>I assume that's just Nvidia using their shit more efficiently?
Well, there are two different things going on with this and its important to understand both:

There are two separate issues going on here with Vega 10 and you need to understand both:

1) There are clearly some issues with Vega 56 and Vega 64 having less memory bandwidth than they should given the type of memory they are using and the clock speed its running at. Vega 64 should at LEAST equal Fiji's memory bandwidth rather than having 20% less. This may be fixable by RTG with BIOS updates and if it is, there is at least 20% performance improvements potentially on the table.

2) RTG wanted to make Vega 10 for a lot of different markets, but they only needed two stacks of HBM2 for pro market uses, so Vega 10 in gaming is limited to competing with Pascal rather than Volta, because two HBM2 stacks just isn't enough to exploit Vega to its full potential in gaming. RTG will need a four HBM2 stack version of Vega to realize its full potential in games.

>>61892210
Well power consumption is relative, I now have 250W nvidia GPU if you have new Pascal then sure, you would think that Vega is huge power consumer, but then again you aint looking to buy Vega.

>>61892179
That was because the titan and the quadro are the same but they can't sell it for a different price if it's the same.

>>61892246
And nVidia can extract full gaymen perf from 3.5k ALU card with only 480GB/s peak memory bandiwith.
Basically RTG is shit.

>>61892172
Listen up, asshole: I've been running around the internet for WEEKS now trying to get people who actually know way more than me about GPUs to try and figure out WTF is going on with Vega in the hopes that someone will actually figure it out, while you've been here shitposting.

I'm basically the only reason people who actually know shit about GPUs are talking about the memory bandwidth issue and might actually fucking ask RTG what the deal is.

I go on about it because it is the KEY question for understanding WTF is up with Vega 10 in gaming.

>>61892268
Ok i guess that makes sense

>>61891779
Not seeing much disappointment there desu. It's performing as expected.

>>61892228
Yes, but the difference in realized memory bandwidth is much worse than that. Vega 10 has ~300 GB/s vs Fiji's ~360 GB/s

>>61892270
>And nVidia can extract full gaymen perf from 3.5k ALU card with only 480GB/s peak memory bandiwith.
Are you literally retarded? Vega 64 is only getting 300 GB/s, give Vega 64 480 GB/s in raw memory bandwidth and it will utterly dumpster a 1080.

>>61892286
They are talking about "peak" performances, averages can be well different because of how bw compression responds to load etc

>>61892286
That is a 20% difference which is entirely possible.
>>61892313
Too bad it will never get that realized bandwidth.

File: AMD-Fiji-GPU-High-Resolution-Shot.jpg (77KB, 900x507px) Image search: [Google]

77KB, 900x507px

>>61892246
My issue is the memory bandwidth of two stacks is yuge and yet.. the performance isn't there. As for 4 stacks, fiji had 4 stacks
Why the fuck does this shit only have two?

>>61892313
1080ti also has less than 484GB/s effective BW.
Yet there's night and day difference in perf between Vega10 and GP102.

So the mining hash rate was a total lie.

Hype lie disappointment cycle continues

>>61892339
>Too bad it will never get that realized bandwidth.
Can we maybe actually get an on the record statement from RTG about whether and to what extent the memory bandwidth issues are fixable with BIOS updates before coming to a conclusion?

>>61892343
Because WX/SSG/FE don't need more than two stacks and you don't need more than two stacks to make RX cards targeting the 1070 and 1080. There WILL also be a four stack version of RX Vega either instead of Vega 10 x2 or piggybacking off Vega 20 development.

>>61892403
>You need silly expensive memory setup to make Vega10 perform decently in gayms
Maybe it's fucking time for RTG to invest in proper fucking DCC?
Navi is coming you know.

>>61892343
>Why the fuck does this shit only have two?
Yields, cost

>>61892156
>DCC is unrivaled.
Explain
Google didn't find me anything

>>61892376
I assume it was to help sell Vega

>>61892439
Nvidia has pretty much unrivaled delta color compression. It saves bandwith.

Hell like what the fuck, fetch once is supposed to save memory bandwith, but Vega is STILL bandwith starved.

>>61892403
Even if it got 10% of that bandwidth it'd only be ~30 GB/s more. Don't think it's going to happen anyway. If there was something to fix, it should've been fixed by now.

>>61892376
How much do you wanna bet retarded miners will still buy it without factoring energy costs?

File: B3Dbandwidth.png (5KB, 518x269px) Image search: [Google]

5KB, 518x269px

>>61892371
>1080ti also has less than 484GB/s effective BW.
1080ti has ~350 GB/s effective texture bandwidth, notably this is right around Fiji's 350 GB/s effective texture bandwidth as compared to Vega 10's 303 GB/s effective texture bandwidth.

Now deduce for yourself how Vega 10 would perform with the same effective texture bandwidth as a 1080 ti.

File: b3d_fillrates.png (62KB, 1016x1203px) Image search: [Google]

62KB, 1016x1203px

>>61892463
>Vega is STILL bandwith starved.
Because it has a 20-30% bandwidth regression vs. Fiji. Give it back that 20-30% memory bandwidth and you'll gain exactly the same % of gaming perf.

Now i just wonder if you could get one.

>>61892478
>Now deduce for yourself how Vega 10 would perform with the same effective texture bandwidth as a 1080 ti.
Oh, very well.
Too bad it has no such bandwith right now.
A-a-a-and that's the fucking problem.
It's a 4K ALU card with the bandwith of 1080.

>>61891062
>asics for machine learning research
No
>igpu
Different market entirely

>>61892520
>Oh, very well.
>Too bad it has no such bandwith right now.
>A-a-a-and that's the fucking problem.
>It's a 4K ALU card with the bandwith of 1080.
Yes, I understand, and exactly what I want to find out is whether and to what extent RTG can fix the memory bandwidth with BIOS updates.

And since you've now admitted you understand exactly what's going on, can you please stop shitposting about perf/mm^2 and perf/ALU when you know exactly that Vega 10's perf in those two categories is entirely bound by it's memory bandwidth?

>>61892506
Ding dong diddly nope. Miners will buy all the gpus no matter how bad. Only gpu seem to not be affected are the 1080+ and 1050 ti's

>>61892057
plus the mining tax of 200usd and your all set

>>61890769
1070 performance

>>61892574
>can you please stop shitposting about perf/mm^2 and perf/ALU when you know exactly that Vega 10's perf in those two categories is entirely bound by it's memory bandwidth?
But that's the funny part.
Anyway they failed to sort out BW issues for launch. That's very bad, first imression is EVERYTHING for consumer GPU.

>>61891841
>finna
Tyrone can you even afford a new GPU

>>61892607
>But that's the funny part.
>Anyway they failed to sort out BW issues for launch. That's very bad, first imression is EVERYTHING for consumer GPU.

I actually think RTG is in a damned if they do, damned if they don't situation. Even if they COULD have had a BIOS update to fix the memory bandwidth regression done prior to launch, all it would result in is miners buying every single card because fixing the memory bandwidth also necessary increases the ETH hashing rate.

>>61890769
>>61892604
Yes, Vega 56 is 1070 performance and Vega 64 is 1080 performance.

>>61892643
The reviews would still make a good impression of new uarch.
That's much more improtant than GPU shortages.

>>61892574
>what I want to find out is whether and to what extent RTG can fix the memory bandwidth with BIOS updates
You're lucky if they get another 10 GB/s out of it.

>>61892648
1070 performance

>>61892665
>You're lucky if they get another 10 GB/s out of it.
Why?
It's not a hardware issue, since they had all the time in the world to do even a base layer spin.

>>61889822
No, 56 is the sweet spot just like 1070. A 1080 have no porpouse becausr it cant 4K and 1070 is good enough on 1440p. If vega 64 prove itself good in 4k it will be a nice card if it shows to be another 1080 it will be trash because powerdraw.

What botters is if amd will limit the power level or voltages so you wont be able to reach the 1600mhz with a decent cooling on a 56.

>>61891896
>timeline

kys redditfag

File: 1477263429243.jpg (298KB, 2162x1236px) Image search: [Google]

298KB, 2162x1236px

forgotten reminder to ignore shitposts and nvidia can' async cant dx12 cant vulkan and gimpworks certified garbage

Well APPARENTLY the dude who benched that has contacts with AMD and APPARENTLY buildzoid knows him (and APPARENTLY buildzoid is also under the NDA).
What the fuck is going on.

>>61892735
fuck you 60hz pleb

>>61893013
The most likely : It was a Vega 64 he was testing.

https://videocardz.com/71771/final-radeon-rx-vega-64-3dmark-performance

Look under "AMD Radeon RX Vega 64 3DMark Fire Strike Ultra"

>>61892881
How do you know im from reddit?

>>61893031
Back to /v/ now, fgt.

Anyway AMD needs better DCC for Navi. Interconnect bandwith is not fucking free.
>>61893034
Video was explicitly about 56 vs 1070 though.

Ow fuck buildzoid is spilling the beans.

>>61893054
"Leaked" HW unboxed results put Vega 56 right around 1070 levels. https://videocardz.com/71953/are-these-radeon-rx-vega-56-benchmarks

>>61893091
And "leaked" that guy results (who has direct access to card from AMD) suggest it's better than 1070.

>>61893103
I would be shocked if the change to a fan style cooler helped that much over the reference blower.

>>61893117
I don't think ref blower would have troubles cooling 210w board.

>>61892115
Buildzoid is smart, but can you guys place this cuck's accent? Is this like a Newfie accent or something? It's like vaguely Canadian/British. Or is he just a faggot?

Can't wait for this to finally launch so a bunch of retards can cherrypick the absolute worst game that Vega runs on vs. Pascal and screech about how AMD is shit and 1050 ti performance at 999 niggahurtz.

>>61893400
>the absolute worst games
It's usually geometry heavy shit.
But Vega is hilariously good at geometry.
I dunno what you have to cherrypick.

How long till the reviews drop?

File: 580 1060 game breakdown.png (505KB, 1920x1080px) Image search: [Google]

505KB, 1920x1080px

>>61893411
See those games on the very bottom there? Yeah, those ones.

>>61893604
>geometry heavy goyworks shit
Not gonna work with B-bega (probably).

>>61893631
it's pronounced wega

>>61893633
its geyga

>>61889800
well, same way 1700x is kind of worth it, but 1800x not worth it at all

crypto miners fuck off

REEEE

File: 1502687980150.jpg (219KB, 792x504px) Image search: [Google]

219KB, 792x504px

>516W

THE HOUSEFIRE RISES

File: 5f0.gif (3MB, 540x300px) Image search: [Google]

3MB, 540x300px

>>61895581

>tfw ryzen vega apu is a flop

>>61895661
Yeah, there was some benches a few days ago. Its GPU scores way lower than HD550.

File: b6b4f6095944affde44975b3a8cd23c8_1502679280_9008.jpg (240KB, 719x1006px) Image search: [Google]

240KB, 719x1006px

>>61895581
>>61895631
1070 performance!

>>61895671
link?

>>61895687
Nope, no link. Probably even seen it there, so I thought there's no point in saving something which can be fake. The difference was like 1800 to 5K points, I think. Can't find it in the history, sorry.

>>61895685
>cherrypicking
>making the same post in every thread
>doing it for free

>>61895661
The Ryzen/Vega APUs aren't out for a long time. They just launched updated last generation arch APUs.

>>61895751
enjoy your badly implemented dx12!

>>61895765
they will be out in 3-4 months, confirmed by CEO.

>>61895581
>Power limit +50%

>>61895836
Su said 2018Q1, not anywhere soon

>>61892939
Not sure if ironic shitpost or genuine retarded amdrone

>>61895685
>>61893400

>>61895671
There were a few leaks but I don't remember anything near that bad. Vega's desktop performance isn't going to tell us much about how it performs in low power embedded space.
>>61897246
Those are the desktop APUs not Ryzen Mobile.

File: aida64_2017-08-14_14-16-55.png (167KB, 400x574px) Image search: [Google]

167KB, 400x574px

>>61892102
>>61892242
My stock GTX1070 as a comparison.