[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vr / w / wg] [i / ic] [r9k] [s4s] [vip] [cm / hm / lgbt / y] [3 / aco / adv / an / asp / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / qst / sci / soc / sp / tg / toy / trv / tv / vp / wsg / wsr / x] [Settings] [Home]
Board
Settings Home
/g/ - Technology



Thread archived.
You cannot reply anymore.



File: v64 vs v56 shaders.jpg (198 KB, 2549x1444)
198 KB
198 KB JPG
Vega 64 and Vega 56 are LITERALLY the same! The only difference is Vega 64 has a higher power limit and higher default clock state.

http://www.youtube.com/watch?v=jHRP1uVBqfY
>>
File: VEGA.png (1.09 MB, 1294x860)
1.09 MB
1.09 MB PNG
>>62416044
>p-please buy our cards goy

Good job with Ryzen, Vega is a fucking failure
>>
>>62416044

They are only the same at gaming right now because the primitive shaders aren't working for gaming workloads.

Vega for gaming is pretty much an overclocked Fury or an sup-up Polaris with more VRAM until AMD fixes it.

Vega 64 is faster at general compute and professional graphics though (a.k.a primitive shaders are working).
>>
File: 1454039405929.png (49 KB, 808x805)
49 KB
49 KB PNG
>>62416088
DONT UNDERVOLD NOOOOOOOOOOOO MTY 1070 REEEEEEEEEEEEEE
>>
>>62416088

Vega isn't a failure idiot. You are just mad that it didn't kill "Pascal" and force Nvidia to do price cuts.

AMD selling the crap out of the Vega cards with decent profit margins. It is a different matter whatever the sales will hold up when Volta comes around. Besides, high-end GPUs aren't big money anyway for GPU vendors. It is the mid-range shit that makes the bulk of the money and Nvidia has that market by the balls via compentive products and sheer mindshare.
>>
>>62416044
Well of course it would be the same.
You can trim it to 2.5k ALUs and it would be the same.
The ALU saturation is basically nonexistant due to choking frontend.
It's up to RTG to fix it.
>>
>>62416288
It is a failure of a consumer card (well for now).
Radeon Pro looks bretty kickass though.
>>
If I was Lisa Su right now I would be pushing the driver team harder. The hardware is there. It's the software side that is letting team red down.
>>
>>62416319
The only thing they'll be pushing is ROCm releases.
>>
>>62416288
>high-end GPUs aren't big money anyway for GPU vendors
Tell that to NVIDIA after selling more GTX 1080s than AMD sold RX 480s. I'm sure they made quite a pretty penny.
>>
>>62416375
nVidia sold more Thermis than AMD ever sold Evergreens.
Mindshare rules the consumer market and not good hardware.
>>
>>62416044
>youtube video
Got a proper link to a tech website?
>>
>>62416389
>VHS > Betamax
>MP3 > CD
>iPhone > ?
>>
>>62416389
What does have to do with anything? It's literally beside the point, that being that saying high-end aren't big money isn't necessarily relevant. There are more people willing to spend $600-700 for a 1080 than $200-250 for a RX480.

>Mindshare rules the consumer market and not good hardware.
Perhaps, but AMD has neither currently and they will definitely not get any mindshare if they don't get the good hardware first. You might argue that Vega is good strictly from a hardware perspective and the software is holding it back - fair point, but also irrelevant since it doesn't matter how good the hardware is if it cannot be taken advantage of, especially in the consumer market.
>>
>>62416417
Literally on the channel

https://www.gamersnexus.net/guides/3053-vega-64-vs-vega-56-clock-for-clock-shader-differences
>>
>>62416483
>Perhaps, but AMD has neither currently and they will definitely not get any mindshare if they don't get the good hardware first.
They will never, ever, ever have the mindshare back.
Even RV770/870 failed to do that.
>There are more people willing to spend $600-700 for a 1080 than $200-250 for a RX480.
More people buy nVidia in general.
>>
>>62416375

GTX 1080 sold units than RX 480s which is expected because high-end GPUs don't sell in large volumes. It is only outselling the Vega family right now because it has been in the market for a year now. It might be a different story next year.

It is the GTX 1070 and 1060 that are selling like hotcakes. AMD is mainly selling RX 470s and 480s due to the crypto-currency craziness though not so much for their gaming prowess.
>>
>>62416510

That's completely wrong. There aren't many people getting $600+ GPU in general.

High-end GPUs have never sold in large volumes ever. By volume and revenue they are dwarfed by mid-range and low-end discrete solutions.
>>
>>62416044
wtf they are bottlenecked to the same performance? that's crazy
>>
>>62416632
Right now it's only 4 frontends and nothing else.
>>
>>62416510
>More people buy nVidia in general.
Yeah, but those people who bought 1080s didn't even have an AMD alternative until very recently. They cared more about the performance than AMD's value proposition.

>They will never, ever, ever have the mindshare back.
Realistically? Probably not. They would need multiple generations in a row of top-notch cards which beat NVIDIA by solid margins and do not suffer from various software issues. They need better performance or at the very least nearly identical performance while having significantly better value. That's just implausible considering how NVIDIA hasn't (yet) made the Intel mistake of sitting on their ass and doing nearly nothing for a decade.

>>62416559
According to the current Steam survey, if it's in any way accurate at all, NVIDIA has sold more 1080s than AMD sold Polaris cards, all of them put together. I don't expect the Steam survey to be incredibly accurate, but I don't see why it would be extremely biased either, so the trend should in general hold up.
>>
>>62416299
>literally the same issues fury had
>repeated again

Raja and his entire management clique that did this need to be fired.
>>
>>62416658
>That's just implausible considering how NVIDIA hasn't (yet) made the Intel mistake of sitting on their ass and doing nearly nothing for a decade.
Making faster GPU is hilariosly more east than making faster CPU.
>Realistically? Probably not.
And that's precisely why AMD is wandering into other markets right now with wacky products like SSG.
They KNOW they have no chance in consumer space.
>>
>>62416510

ATI was at its strongest in mindshare back when R3xx-R4xx were dominating Nvidia's FX family.

They went toe to toe with NV4x-R5xx families but Nvidia started to get ahead again with G8x-G9x mostly due to 8800GT. R6xx was a misfire that managed to get back somewhat with RV7xx but Nvidia kept on going a roll until GF100 misstep but quickly turn that around with GF104-GF110 and have been cruising along ever since without a hitch.

ATI/AMD never quite recovered from the damage done from its R6xx misstep.
>>
>>62416689
They did not repeat them again.
In fact they came up with a very elaborate solution to that.
They are just not ready to launch it for consumers.
>>62416704
You conviniently forgot GT200 vs RV770.
>>
>>62416650
4 frontends? what does that even mean
>>
>>62416389
>nVidia sold more Thermis than AMD ever sold Evergreens.
>Mindshare rules the consumer market and not good hardware.

No, what that means is Nvidia's marketing is so good they can still outsell their competitor while having an inferior product. Likewise, AMD's marketing is so bad that they cannot outsell their competition in spite of having a superior product.

Mindshare has duck-all to do with it. That's an excuse AMD fans use to excuse AMD's poor performance in the market. Zen hasn't needed mindshare to sell - because it got decent marketing behind it AND it's a superior product.
>>
>>62416717
>They didn't repeat their mistake
>They just repeated their mistake but it'll be fixed soon^tm
>>
>>62416717

GT2xx was nothing special though. It was just a G8x-G9X on roids. The first generation GT2xx SKUs were buttfucked by AMD's aggressive pricing but Nvidia was able to turn that around with second batch of GT2xx and kept their mindshare intact.
>>
>>62416746
>No, what that means is Nvidia's marketing is so good they can still outsell their competitor while having an inferior product.
No, it's because of mindshare.
Fermi's marketing was HORRENDOUS with denial at every stage and they still sold more of them.
Heck only Maxwell onwards had decent muh PCMR marketing.
>>62416781
>turn that around with second batch of GT2xx
These were overpriced turds.
>>
>>62416692
>Making faster GPU is hilariosly more east than making faster CPU.
A faster CPU isn't all IPC and clocks, as Ryzen has eloquently demonstrated. I had a C2Q in 2007, 10 years later Intel's mainstream is still quad cores.

>They KNOW they have no chance in consumer space.
Yeah, sadly they really do not seem to have a chance nowadays. They weren't beating NVIDIA but they had some pretty solid, desirable products up to the 7970, which was an absolutely excellent card. Now they don't even have that anymore. Vega is a year and a half late and can't get anywhere near NVIDIA's top card.
>>
>>62416825
>A faster CPU isn't all IPC and clocks, as Ryzen has eloquently demonstrated.
Ryzen just accidentaly happened to be good for consumer tasks.
It's server first uarch.
>Vega is a year and a half late and can't get anywhere near NVIDIA's top card.
It is, in compute and CAD.
And they don't expect them to sell it anywhere else so it counts.
>>
What's the max overclock you can get on a 64?

My 56 seems to be stable at ~1680Mhz core / 1100MHz memory . Haven't tried to push further.
>>
>>62416288
>selling the crap out of the Vega cards with decent profit margins.
Quite the opposite, AMDs Martine on Vega are extremely small.
Some reported they sold at a loss at the original $500 MSRP.

And this undervolting meme is also funny.
Utilities are reporting the wrong voltage and power usage.
Users think their Vega is suddenly as power efficient as a 1070.
Using a volt meter connected to the PCB shows a different story.
>>
>>62416803
No, it's because of marketing
>>
>>62416871
No, it's because of mindshare.
Fermi had fucking horrific marketing, with woodscrews and onwards.
>>
>>62416848
>It is, in compute and CAD.
Weren't we talking about the consumer market? That's where the "mindshare" problem is most severe.
>>
>>62416803

Nope, GTX 260 (216) and GTX 275 were competitive to their AMD counterparts.

Nothing could touch the GTX 285. HD 4890 barely caught up to the GTX 280 for slight less $$$.
>>
>>62416894
AMD is not betting at consumer market.
They know they lost all mindshare looooooooooooooong ago.
That's why you got RX Vega as it is now.
>>
>>62416902
>Nope, GTX 260 (216) and GTX 275 were competitive to their AMD counterparts.
And were still pricier even AFTER the price cuts.
>HD 4890 barely caught up to the GTX 280 for slight less $$$.
280 was a HUEG die compared to 4890.
>>
>>62416869

You are an idiot. High-end GPUS are never sold at less or thin-margins ever. Those reports that claim $500 production cost are baseless rumors.

The problem is that are small volume so that decent to handsome profit margin per unit sold doesn't really amount to much.

Mid-range shit is what makes the dough via sheer volume for both vendors and Nvidia rules that roost.
>>
>>62416886

Actually, Fermi did quite well. The GF100 was really the only sore spot while the rest of the family outsold and yield comparable performance/price to their AMD counterparts. GF110 took the crown.
>>
>>62417017
It still had horrific marketing.
It was late, hot and pricey.
Even meme GPGPU wins were not wins at all.
And yet nVidia still sold more of them.
I love consumer GPU market, it's pants-on-head retarded.
>>
>>62417073
GTX 460 really wasn't that horrible.
>>
>>62417222
It was merely okay because AMD had no product to fill in that niche (well besides 5830 that came later and was about as meh).
>>
>>62416886
It's because of marketing, AMDrone

get better at it
>>
>>62417249
Fermi had horrific marketing.
>>
>>62417242
Yes, but the x60 NVIDIA segment is the one that usually sells shitloads. Even HD5k vs. GTX 400 isn't as clear cut as "Fermi was absolutely, completely horrible"
>>
>>62417288
No, it's x70 cards that sell the most.
Really (just look at 970 sales).
And 5850/5870 were uncontested.
>>
File: Capture.png (26 KB, 922x108)
26 KB
26 KB PNG
>>62417305
Doesn't seem to match up with statistics
>>
>>62417346
>steam HW survey
It's opt-in.
Aka it's useless.
>>
>>62417305
>And 5850/5870 were uncontested.
Way better bang for buck but Fermi was still a bit faster.
>>
>>62417377
10% faster for $150 more.
And 5970 still existed.
>>
>>62417352
I assume you have some evidence regarding it's uselessness, or why it would be biased against higher-end products at least? Are x60 owners more likely to opt-in than x70 owners? If so, why and based on what evidence are you making the claim?
>>
>>62416958
Holy shit, you are retarded. Vega is not profitable. RTG has barely ever been profitable and that's with the bulk of their offerings being high margin low/mid range components. Vega is the definition of a money pit.

It's a 500 sq. mm die with 8 GB of HBM2. The only reason these aren't $1500+ GPUs is because they're not being sold by Nvidia. AMD is getting BTFO on Vega margins because they know they're fucking AMD and they can't overprice them or LITERALLY nobody will buy them.

Honestly I've been against Vega from the start. The thing should have been axed in favor of ramping up MCM earlier. Unless it has huge success in the business sector it will go down as one of their biggest failures.
>>
>>62418250
>ramping up the MCM earlier
Moar ALUs are as good as useless without the changes that Vega brings.
>>
>>62418365
>changes that Vega brings.
What changes?
From what we've seen, Vega preforms on par with the furyx clock per clock.
>>
>>62418365
What changes are those? When I said ramping up MCM I meant scrapping GCN altogether and starting the fuck over. What the fuck does Vega bring to the table? AMD hasn't been able to figure out how to saturate their shaders for like six fucking years straight.
>>
>>62418612
>Hasn't read the Vega whitepaper
>Shitposts anyway
(You)
>>
>>62418637
>Also hasn't read the whitepaper
>Also shitposts anyway
(You)
>>
File: technocracy_scary.jpg (229 KB, 1600x1600)
229 KB
229 KB JPG
>muh whitepaper
>just wait guys you will see
>>
>>62416375
the 1080 is a mid range card sold at a high end price.
>>
File: 1481393540560.jpg (44 KB, 615x409)
44 KB
44 KB JPG
>>62418640
>Vega whitepaper
>meaningful
WEW LAD I'm sure it totally isn't filled with corporate PR bullshit.

Tell me how the Vega white paper is important when discussing how plainly evident it is that the only thing separating Vega 56 from Vega 64 performance is a BIOS that kneecaps 56's power delivery. In other words, the bottleneck is somewhere else. This would be excusable if yields were so poor on 64 that they needed 56 to exist to get some revenue out of the rejected dies but this isn't likely the case because the 56 is the harder GPU to find at retail and all reports are that the process is mature and the yields on 14nm are fine. Either way, that wouldn't make the 64 a good buy if all it takes is a fucking BIOS flash on a 56 to make it irrelevant. I'm beginning to think you're a dumbass m8. Go ahead and tell me now how Vega is a necessary step for MCM. I'm sure Lisa Su would agree with you after firing Raja earlier today lmaooooooooo

By the way, I haven't owned an Nvidia graphics card since the Geforce 4 but even I am getting so sick and tired of the AMD defense force. Newsflash: Vega fucking sucks. It's a year and a half late, slower, less power efficient AND more expensive than the competition.
>>
File: oiK8Bb9.jpg (906 KB, 793x877)
906 KB
906 KB JPG
>>62418731
Gee, I wonder if there was some new software feature described in the whitepaper that hasn't been implemented yet that would do something to alleviate GCN's historical front end geometry bottleneck? The same bottleneck responsible for Vega 56 being 100% as fast as Vega 64 at the same clocks presently?
>>
>>62418731
Final conclusions cannot be made yet. Even if Vega came fucking late. Wait for drivers.
>>
>>62416958
Amd has always sold its gpus at very low margin to compete.

the current vega rumors are based on people with fab experience estimating prices of hbm, with the one Im most familiar with quoting it in the 150-175$ range, then you see wafer size, vs yeilds and you can estimate cost there too.

amd is on very low margins with their vegas, but distributors are making shit up well above msrp before it even comes close to consumer hands.
>>
>>62418250
Vega is a real world test for many of the things that would allow an mcm gpu possible, in this regard vega is a necessary stepping stone.
>>
>>62418815
The issue is the hear and now, and vega is disappointing in that regard as neither can even come close to msrp.

in a few months, when im back in the market for a gpu, the driver may have come out and fixed some of the shit. but right now all we have is a confirmation that primitive shit isnt in driver.
>>
>>62416745
the current driver only handles 4 Xes at once, the card is capable of 17 Xes at once, but the drivers need nearly a 100% rewrite to implement. This is a massive undertaking, even if amd had a stellar driver team, they got competent people, but god damn they have the work cut out for them.

It should be noted, I can't remember the exact name of the bottleneck situation I just remember the numbers.

Part of this can be seen in dirt 4, where the gpu isn't being hammered on X as hard, and it flys passed a 1080ti, what the came does hit a bottleneck area is compute, which nvidia gimped to hell and back.
>>
>>62419022
see
>>62418808
>>
>>62418250

It is profitable but it is small money (several million USD of revenue at most). The real money in discrete GPUs has always been mid-range and low-end cards (hundreds of millions to a few billions of revenue). Nvidia dominates that by a mile via sheer mindshare and volume.

Nvidia isn't making big money on 1080Ti and 1080s sells. It is all from 1070s and 1060s with a considerably smaller portion coming from Quadros/Telsas. 1080 and 1080TI are just halo products meant to strengthen mindshare of lesser SKUs.
>>
>>62418365

Vega 10 is AMD's version of GF100.

GF100 wasn't so hot at launch because Fermi was trying to reinvent the wheel, once Nvidia software and engineer teams got a hold on it.

Fermi started to rock as seen with GF104 and GF110. Kepler push it further along and Maxwell took to the next step. Pascal was just Maxwell on roids.

I suspect later version of Vega are going to be more potent, but real goal of Vega => Navi is taking over the iGPUs and murdering mid-tier discrete GPUs (bulk of the revenue in the discrete market).
>>
>>62416658
What is miner demand for $400?
>>
>>62419520
GF100 was hot because nVidia went full retard and tried to build HUEG die on a new node after failiing to yield 113mm^2 one.
And not because of whatever magical reason you have thought.
>>
>>62417305
What are you, fucking 12? The 970 was literally the first card that Nvidia marketed heavily in a good way and managed to convince people that it was a mainstream card for the prices it was going(it fucking wasn't). Before that it was nvidia x60 cards and amd x700 cards selling at roughly $200.
>>
>>62418612
But the fury x can't clock that high.
>>
>>62418250
Are you retarded? The GPU division was the only part of AMD that was profitable for ages.
>>
>>62420471
Nice reading comprehension, you massive faggot. Where did I say the rest of AMD was better? I just said that they've been in the black on occasion, but barely, and that has had nothing to do with high end offerings like Vega. Trying to argue that Vega isn't a massive waste of money (or "investment in the future" if you're delusional) is idiotic. It's clearly not going to be profitable at MSRPs based on the components on board and all of the R&D.
>>
>>62416869
I measure at the wall which is close enough to give a ball park figure. My RX Vega 56 on 64 BIOS with an undervolt and overclock uses about 100w more than a GTX 1080 in order to match a 1080. So a bit more expensive per quater in electricity I agree. Other than that though it's fine and I got it at stand alone MSRP.
>>
>>62417420
No evidence but based on the mindshare of AMD fans they tend to be more savvy regarding things like sharing information/privacy and in their buying choices. Many understand how team green operate and see AMD as more of an underdog (even if they really are not). Nvidia fans are more about the brandding/gaming. AMD fans are more about professional/cost savvy/privacy advocates. So are less likely to take steam surveys or even use Steam at all. Plus the miners of course who would never need Steam anyhow.
>>
>>62420761
$7 more expensive per year
PER YEAR.
>>
>>62420944
No one really cares about ebin efficiency in the real world (okay Maxwell marketing slides do care).
What matters is performance and Vega is seriously lacking it due to being more or less bigger Polaris right now.
>>
>>62416288
>being this delusional
>>
>>62416319
Pajeets can't do miracles babe, barely beating a 1070 us all the damage AMD was able to do this generation...


HAHAHAHAHA.
>>
>>62420981
>R300
>>
Give it a month or two and we'll be shitposting how Vega is destroying 1080Ti's once the drivers actually make use of the GPU features. It's the reason Dirt 4 is is so high on Vega because the Dirt 4 engine is not geometry bound using CMAA.
>>
>>62421025
DiRT4 is an outliner for both vendors.
Anyway, TensorFlow port to ROCm when?
>>
>>62421025
>Give it a month or two and we'll be shitposting how Vega is destroying 1080Ti's once the drivers actually make use of the GPU features.

Just Wait TM

>muh drivers

Vega is housefire power hungry sack of shit, muh drivers won't change that
>>
>>62421071
>housefire
So is anything gp102. And no one complains.
>muh drivers won't change that
Muh drivers one killed 8500.
Anything is possible if you try hard enough.
>>
wait for drivers tm
>>
why dont reviewers test nvidiot cards with the earliest drivers? could it be that theyre SHILLS
>>
>>62421675
>why dont reviewers test nvidiot cards with the
Nvidia day 1 drivers are used for testing.
The difference is Nvidia does a damn good job optimising their drivers from day 1.

AMDs "fine wine" bullshit forms from their launch drivers being so horribly bad.
It's easy to gain 10% when ur day 1 driver is a half assed attempt to optimise.
Look at the 290x, took a while fucking year to get the driver right.
Then the gtx980 rolled around and the 200 series didn't matter anymore.
>>
>>62420962
it is as fast as 1080, is 1080 crap at gaming?
1080 - zero potential
vega - black box
>>
>>62422333
Checked

Vega is horribly overpriced and ends up being $100 more than a base gtx1080.
So it's slower on average
More expensive
Close to double the power draw.
At its current price point, the GTX 1080ti makes more sense.
>>
>>62416472
>>iPhone > ?
iPhone > iPod
>>
File: prease.jpg (44 KB, 341x544)
44 KB
44 KB JPG
Oh look, it's another thread of "is /g/ actually retarded or just full of spambots?"
I can't wait to find out how this ends, maybe in 50 more threads like this one!
>>
>>62420987
>12 years ago
amdpajeets living literally over a decade ago
Roman Empire also ruled the world.
>>
>>62416768
>I haven't read the Vega whitepaper
>>
>>62421748
> optimising
Nvidia press drivers contain 'bugs' that don't render games correctly in order to increase fps. See: AotS snow




Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.