AMD Giving Up on CrossFire with RX Vega

>>61678159
You don't need cf support when you don't drivers!

why would you even use 2 cards when you can have 1 powerful card. newfags

>>61678333
>vega
>powerful

Now instead of crossfire they will sell dual gpu cards.

>>61678159

tfw AMD cares for our safety and wont let some lighthearted users get burned alive while trying to cf two vegas

AYYMD IS FINISHED & BANKRUPT

AYYMDPOORFAGS CONFIRMED ON SUICIDE WATCH

>>61678333
But even Raja show than 2xpoolaris was faster than GTX 1080

>>61678159
>based AMD killing off the sli/crossfire meme
Thank fucking god

>>61678454
They won't need it if Navi works as intended.

>>61678159
God just finish off AMD already.

>>61678422
That was just their marketing spin on Polaris because it was a shitty architecture that didn't scale at all, hence the whole "85% of the market is under $299" buillshit and then they turn around with Vega and go "well the top end of the market represents 66% of the margin so we're shooting for that now!"

>>61678681
Yeah it was a stupid move, they bent over and basically asked Nvidia to fuck them in the ass by raping their profit margins and making up for it in the high end.

>>61678681
it was a shitty uarch..

you literally dont know shit about gcn eh

>>61678746
>GCN
>good
lol

>>61678757
>autistically lists a bunch of impressive stats
>no drivers

lol the tripfaggot is craving for some those (you)s

>>61678776
It's a bit ironic that the AMD fanboys started hyping up the Crimson series drivers yet they use incomplete drivers as an excuse for Vega's absolutely garbage performance.

To be fair Crimson drivers are very good now, my biggest gripe with my 1080 Ti is the garbage drivers, control panel looking like Windows XP software and the DPC latency issues which still haven't been fixed but lo and behold can't have the best of both worlds.

>>61678807
so amd lied about the drivers
the people that tested FE and said that it was almost as a fiji on gaming lied
but as usual /g/ was right!
why even bother going on beyond3d and discuss the tech there when your shills on /g/ have a better knowledge judging by pictures

>>61678842
So AMD lied about RX Vega performance in their own launch slides?

>unironically thinking AMD are sandbagging
Stop posting.

>>61678863
The BF 1 results they gave are the closest we have to real numbers. They were done on prototype RX Vega and prototype drivers, though.

>>61678873
>They were done on prototype RX Vega and prototype drivers, though.
lol

>>61678863
zen is calling they want your tears to cool them down

>>61678159
Friendly PSA: Dont forget to sage, hide and filter the tripfaggot.

>>61678842
AMD cares more about rebrands and marketing than drivers.

Literally right after Jensen Huang gave away a bunch of AI cards to a bunch of white CS students with his signature Kaduri gave away a bunch of Vega cards to minorities with his signature.

They are a fucking joke.

>>61678888
What tears? I'm not the one in denial about RX Vega's garbage performance and high power draw

>>61678159
Gaymers BTFO

File: 1501607764368.jpg (476KB, 2000x1125px) Image search: [Google]

476KB, 2000x1125px

>>61678882

>>61678896
instead of money they should pay you in history books

>>61678972
None of these endnotes are related to the BF1 results.

>>61678900
i wonder what your excuse about the power draw will be once wolfenstein comes and nvidia will have to use the cuda cores to get near amd

>>61679007
>nvidia will have to use the cuda cores
????

You mean use their shader cores like any GPU would?

>>61679016
if you dont know what i meant then dont fucking reply
f16 is op purely on cuda cores and cuda cores are underperfoming at a 1/32 for fp32 and 1/64 on fp16 they will have to use a lot of power to fucking use any kind of fp on 1070/1080/80ti
it was like when nvidia was running certain rage features on cuda...their fucking power draw skyrocketed

Get fucked shitcoin miners

Enjoy your Nvidia mining rigs

>>61679053
>two games will use FP16 for some calculations
kek

File: 1501598815616[1].png (186KB, 1682x974px) Image search: [Google]

186KB, 1682x974px

>meanwhile NVIDIA can just unlock performance because AMD are so far behind

>>61679101
fp16 is required by vulkan moron
but hey you arent really the brightest around here so i wont laugh at you

wew, the tripfag is on summer mode

>>61679170
He really is. Don't forget to hide and filter him, it's for the good of all.

>>61679169
>FP16 is the latest AMDrone meme

File: 1501606293127.png (37KB, 711x533px) Image search: [Google]

37KB, 711x533px

>>61679117
>using old-ass SPECviewperf instead of SPECapc

http://www.tomshardware.de/vega-benchmarks-workstation-leistungsaufnahme-gaming,testberichte-242375-6.html

>>61679191
>using cherrypicked benches to push an agenda nobody is buying
nice try

>>61679078
Mining doesn't need CF/SLI. The mining platforms recognize each individual GPU separately.

Ah, all that shitposting about Nvidida dropping 3/4 card SLi and now it's TOTALLY OKAY when AMD does it.

>>61679190
last time amd enabled fp16 for gaming usage this happened
https://arstechnica.com/civis/viewtopic.php?f=6&t=1122459

history repeats itself

>>61679217
Triple/Quad SLI was retarded, though.

>>61679228

I agree, but that doesn't negate all the AMD shill shitposting over it

>>61679209
>Anything NVidia isn't good at is now a meme
FP16 is for deep learning/AI and NVIDIA own that market.

>>61679211
>Being this much of a buttblasted NVidiot /v/ermin drone

I want you to go to Tomshardware right now and eMail the editor and writer of this article and tell them how much of agenda pushing shill they are posting cherrypicked benchmarks.

>>61679224
So why did they stop?

ATi played just as dirty as NVIDIA did.

>>61679254
holy shit you dont even know what the fuck you are talking about eh?
take a look
http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review/5
enjoy the cuckery that is called nvidia

>>61679278
cause the moron amd had as a ceo back then was the one that bought ati and was the one that pushed for bulldozer ....

>>61679311
Yeah buying out ATi was AMD's biggest mistake, look what happened now they had to split them off as RTG and are eventually going to sell them off because they're dragging the company into the shit

File: 1501504827746.jpg (537KB, 1920x1080px) Image search: [Google]

537KB, 1920x1080px

>>61679326
Counterpoint.

>>61679326
buying ati was ok the problem was the ceo wanted epicly stupid designs the bigger the better
i swear he was literally going the hitler's way
lisa literally took amd from a near collapse to what it is now
fun fact he is CIO on dell now and he was the one that burned off the evidence that intel was giving money to dell

>>61679353
As compelling a product as it is, nobody important is going to buy it, all the big studios are already invested in the CUDA ecosystem.

>>61678159
Makes sense when you consider the possibility that combining two of these in a cramped space could create a nuclear fission.

>>61679401
https://www facebook com/jarred.land/posts/10154878409465415?comment_id=10154878470535415&notif_t=like&notif_id=1501276190497990

cuda is just a set of api's and nothing more currently there isnt a single app that runs purely on cuda and since amd already told the press that sony pixar and adobe are already developing support well no cuda compared to what the ssg offers is ancient literally

>>61679436
You type like a 13 year old who has had one too many energy drinks

File: bearbro.jpg (72KB, 800x450px) Image search: [Google]

72KB, 800x450px

>>61679353
>It's another episode of a retarded AMD poster spamming same picture of a $7000 GPU as an argument in a discussion about consumer GPUs.
I hate reruns.

>>61679464
i guess trying to debunk red ceo the company that literally helds hostage the render segment on hollywood is too much

>>61679401
CUDA can be compiled into OpenCL pretty easily now, can't it?

>>61678888
Reminder that Zen runs cooler than kabylake which can literally burst into flames

>>61679481
Blame the guy claiming AMD is going to sell RTG off because reasons.

Oh no less gaymurs are buying our workstation focused GPU architecture, clearly we need to get rid of these fgts before they stink up the place with their anti-gaymur agenda.

>>61678159
Why have two housefires when you can have one giant housefire? Raja messed up big time, the only way im putting that shit in my system is if i get it free

>>61679326
Are you retarded? The graphics division and miners have been the only things keeping AMD afloat until just recently

Of course it's not supported, no residential home could provide enough power to drive 2 Vegas under load.

>>61679575
I get if gamers are mad they waited for a year to get an arch that's better suited to workstation tasks (pretty much what Zen was, really)

But let's face it, hardly any gamers were buying discrete add-in board AMD GPUs anyway.

>>61679591
If AMD were smart they'd just use all those GPUs to mine crypyocurrency for themselves

Don't really care. Crossfire and SLI have been more trouble than they were worth every time I've tried them, it would be good to see both die to be honest.

File: 1499203496073[1].jpg (265KB, 2048x1925px) Image search: [Google]

265KB, 2048x1925px

>>61679615
why would a shovel company dig when they can just make truckloads off of selling the shovels to fools

>>61678159
There are a couple of ways to look at this.

1) SLI is a pain in the dick to write drivers for and the whole DX12/Vulkan thing was to offload that work to developers. Since multi-gpu support is baked into those APIs it might still be a pain in the dick, but it's a much smaller pain in the dick for the developers.

2) Raja has a tweet about Infinity Fabric and Vega somewhere. It was about enabling IF in Vega drivers being hard or something. It's entirely possible that they are fiddling with IF in such a way that multiple GPUs don't look like multiple GPUs. Perhaps they are using IF to force GPUs to function like the CCX in Zeplin.

>>61679567
when nvidia puts a hardware sc their power draw will skyrocket too just like it did on kepler
but hey its amd bla bla

>>61679724
NVidia's "simple hardware complex software" approach can't last forever. I'm looking forward to the day hardware schedulers return to NVidia.

File: 1487789493624.jpg (156KB, 633x758px) Image search: [Google]

156KB, 633x758px

>>61678159

DELET DIS

>>61678681
AMD ALWAYS DOES THIS

IT'S FINE WHEN AMD CREATES HOUSEFIRES WITH EVERY MODEL

BUT IT'S NOT FINE WHEN INTEL OR NVIDIA HAS THIS ONE SHITTY MODEL THAT DOES IT

>>61678159
well duh
dx12 and vulkan support multi gpus mixed from any manufacturer
of course both AMD and Nvidia will want to kill it dead.

File: 1384116785181.jpg (174KB, 720x720px) Image search: [Google]

174KB, 720x720px

>>61678972

>muh fine prints

:^)

>>61678757
It was

And then they fired the original engineers and replaced them with rtg china

>>61679591
>But let's face it, hardly any gamers were buying discrete add-in board AMD GPUs anyway.
This, nobody buys AMD cards. It's time AMD realized who their real consumerbase is

>>61679845
>It's time AMD realized who their real consumerbase is
Judging by Vega's architecture design, they already have.

>>61679785
Except with DX12 you don't even need an SLI cable or CF mobo to make it work.

>>61679650
Cute girl.

>mfw the last good AMD card was the HD7990 and I had it
Was nice to watch it beat the GTX

I miss dual GPU cards desu

>>61679401
>jews not caring about saving hundred thousands of dollars

>>61678863
They aren't snadbagging, they are or were deathmarching the driver team to finish implementing DSBR before the RX Vega launch, which they essentially stealth delayed twice to August 14 to give them extra time.

Vega FE is like Maxwell with TBR turned off. Until we see how good RTG's implementation of DSBR is we don't know fuck all about how RX Vega will perform relative to Vega FE as already tested.

File: 1481393817965.jpg (120KB, 676x437px) Image search: [Google]

120KB, 676x437px

>>61679757
>I'm looking forward to the day hardware schedulers return to NVidia.

I'm not.

File: aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9SL0svNjkzMDU2L29yaWdpbmFsLzA4LTNkcy1NYXgtU2NvcmUucG5n.png (32KB, 711x533px) Image search: [Google]

32KB, 711x533px

>>61679211
>Not using the massively outdated version used in SPECviewperf is cherry picking.
?

>>61680776
Oh, it will be glorious.

>>61678159
Good
CfX and Sli are outdated garbage
Single GPU is the way to go.

>>61678159
>400 W power draw
>no SLI

No shit, you'd need a 1200 W PSU to run that shit comfortably lol

>>61678159

Not very surprising since the market is full of games which have effects and rendering techniques (basically all temporal effects, they require data from the previous frame so rendering of the new frame can't start before the old one is finished) completely incompatible with AFR multi-GPU.

NVIDIA already pulled the plug on low end and mid range SLI.

>>61679785
>of course both AMD and Nvidia will want to kill it dead.

If they wanted it dead then they would be shilling SLI/CF even harder.

If Vega is same price as 1080 and same performance its totally worth it to buy, cuz Freesync is easier and cheaper to get than Gay-Sync
Would totally get Vega myself, but currently trying to buy 570/580 for normal price, fuck you miners :)

>>61682948
>cuz Freesync is easier and cheaper

Any of that saved money is lost the moment you get your power bill

>>61683036
>power bill

I don't trust Vega FE simulated benchmarks sorry
lets just wait for legit RX tests

>>61682228
>Gamersnexus OCs FE Vega to 400 watts
>HURR DURR VEGA IS 400 WATTS
You retards have brains smaller than a fucking peanut.

>>61683088

>Simulated

AMD said the fucking power draw themselves. 345W for Vega 10 XTX. 190% the average draw of a 1080 for same performance a year late.

>>61683117
That is the watercooled version. Aircooled is 295 watts and that's 60% more power than a 1080.

Amd is right, nobody is retarded enough to buy two Vega

>>61683145

>B-but it's only 60% more power for the same performance a year later

Coming from the same shills screaming "HOUSEFIRE" if Intel uses 10W more than Ryzen.

So predictable.

>>61683157
It's shitty, yeah, but don't make it sound like it's worse than it already is.

>>61683117

dunno, Vega tests are soon, lets wait and compare that shit

File: 1501482882506.jpg (137KB, 1003x844px) Image search: [Google]

137KB, 1003x844px

>>61683097

>>61683774
They put that huge number to cover their asses when dumbasses put it in their PC with 300 watt OEM PSU and it explodes.

It's 345w for watercooled.

>>61683812
With a cf, you will need 1000 watts my friend

>>61678681
polaris wasn't shit. it actually scaled well. if you take the time to learn about it, its power efficiency is essentially on par with maxwell. if maxwell's power efficiency is shit then lord knows nothing will please you shills.
>hurr no its not!
yes, it is. take the time to learn about the architecture. a stock 480 running at its exact stock clocks with the stock power level will be as fast as a slightly overclocked 970 and use the same amount of power as said 970. just like the 970 though, polaris will go up in power usage if you unlock the power level and increase the clocks. hence why popular 970's like evga 970 ftw+ will hit up to 170 - 200 watts of power when boosting to 1.5ghz. just like a 480 at 1350mhz. polaris problem is that it was process limited. the first reversion of global's 14nm that the 400 series was slapped on was garbage. this is was made clearer than daylight with the release of the 500 series. where amd did not change a damn thing with the architecture, just updated it to a newer global 14nm stepping. for the same amount of power consumption they were able to up the clocks by 200mhz. my own sapphire rx 580 according to hwinfo, gpu-z, and monitoring with my p3 p4460 shows my rx 580 using about 170 - 190 watts for 1450mhz. sapphire also capped the power consumption to a max of 210 watts for it as well. jayztwocents also did a good video covering polaris power consumption with their xfx 480 review.

the dilemma amd ran into with polaris was two reasons:
>polaris was gcn 1.3 updated to 1.4 based off hawaii architecture.
amd took gcn 1.3 from fiji, updated it, and based the polaris architecture off of hawaii rather than fiji. fiji architecture successor is vega. its why all the tech talk and promo's from amd had it compared to hawaii rather than fiji.
>they increased the performance of hardware schedulers
>added in some new ones
>increased shader and cu performance by 15%
cont'd

>>61684255
>instruction prefetching and a increased the instruction buffer
>added a primitive discard accelerator & a new index cache which heavily increased its geometry performance (think tessellation)
>ROP/L2/Memory controller improvements and updates (polaris memory controller can actually use gddr5x)
>adopted amd's new delta color compression which copies nvidia's.
>reduce power consumption by 50% compared to hawaii

and that was a problem. amd was working on vega. amd adopted a new strategy with the launch of the 300 series and fury. one architecture to be low die and dedicated for low dies for the mid range and lower end, while another architecture for large dies for the high end market. amd didn't have anything in time for the launch of fury, so they just rehashed hawaii for the 300 series. but for the 400 series they had polaris ready to go. the problem? they didn't have vega. if you go back and look at amd's timeline vega was post to launch shortly after polaris. it really was post to launch in 2016 around the holiday season. but they ran into issues. first issue was the delay with hbm2 and yields. second was amd ran into some sort of issue with vega which has been clearly made apparent with the launch of vega.

polaris could scale higher. shit they even made its memory controller capable of using gddr5x. they said it themselves, that can literally just drop gddr5x chips onto it anytime they want if they decide to. which tbqh they should have as polaris is actually memory starved. even with 8ghz gddr5. which to be fair wasn't all that hard for them to tweak the memory controller to allow for gddr5x but they still took the two minutes to do it. but they had NO reason to as they were wasting time on the failure called vega. which was post to be their bigger die architecture and successor to fiji.
cont'd

>>61684266
>polaris wasn't designed to be their next gen professional, machine learning cards
i know, if that was the case why the hell did they release their wx series. well, to the same strategy mentioned above, they also decided to apply it to their pro cards. smaller, cheaper pro cards based off of higher yield small dies with cheaper ggdr5 and a bigger one based off of vega with hbm2.

also want to add polaris brought amd up to maxwell level of geometry and tessellation performance. it actually has faster geometry and tessellation performance than fiji itself. a 580 will actually be faster than a fury in heavy tessellation scenarios.

>>61683812
>he didn't watch the live VEGA demo
https://www.youtube.com/watch?v=UxQdS0pbpKo&t=13s

>>61684266
>low dies
what i mean is small dies for higher yields. there was no technical reason why they couldn't make larger polaris dies. they just had no reason as they were working on vega with hbm2. since vega was going to utilize hbm in general, hbm by default increases the size of the dies.

vega is a failure though and a mistake.

>>61678333
>why would you even use 2 cards when you can have 1 powerful card. newfags

why would you even use 2 cards when you can only have one hungry card melting psu. newfags

>Fixed for (you) macaco

>>61684315
>vega is a failure though and a mistake.
AMD is going to sell as many $6,999 Vega SSG cards as they can make. Vega would be a success even if they'd outright cancelled the RX version.

>>61678159
Existing CFX/SLI is a clusterfuck of driver hacks, and AMD has basically no cash to pay hordes of driver tweakers for something that maybe 1% of actual customers use.

OTOH, if Navi "scalability" works as intended with HW driven load distribution, all the per-game driver shit becomes unnecessary.

>>61678159
Lame I thought HBM and vulcan was going to pave a new era of stutter free decent scaling xfire.

>>61679169
Kek why doom is already running at 150 fps on 1440p on my 1080? It's some sort of magic?

>>61684315
Vega is similar to how Zen was a workstation CPU put on the desktop. Except Zen went up against a complacent Intel, while Vega went up against a very active NVidia who hasn't been sitting on their ass the past few years.

>>61683145
>Hey stop it's just using 60% extra power for sane performance on AMD approved titles®

>>61678159
>amd logo looks like a sim/sd card

>>61678159
But I thought there was going to be a Vega x2 card?

Will it use infinity fabric and appear to the graphics API/driver as a single GPU, then?

>>61680728
They already showed BF1 benchmarks you fucking clown

>>61686194
>Will it use infinity fabric and appear to the graphics API/driver as a single GPU, then?
no, this is (supposedly) coming on Navi

>>61683774
JUST

>>61687175
None of the benchmarks so far show the slightest indication that they were done with DSBR on.