So the Vega results are in and its shit, its performance is

File: poo.png (312KB, 595x654px) Image search: [Google]

312KB, 595x654px

>>61181554
wait for drivers

>>61181554
All you fags are forgetting RX Vega will be watercooled unlike $1000 FE so look for $1500 FE watercooled reviews for a meaningful comparison to what the gaming card will perform like.

PCPer and this review makes it quite clear Vega is not running at it's promised 1600 MHz clock and is throttling and thus underperforming

Vega at it's proper 1600 MHz clock will be GTX 1080 competitive at the very least.

>>61181554
It's running with TBDR disabled (why? Who the fuck knows). So the results are in line with it being a bigger Fiji with wider pipeline.

Dumbfuck gaymur retardniggers need to be gassed.

File: trashman3.jpg (31KB, 352x450px) Image search: [Google]

31KB, 352x450px

>When overclocking, underclocking, or even restoring a change to defaults, HBM drops from 945MHz (reported) to 500MHz (reported).

>>61181658
Fiji drivers confirmed

File: 1491349230554s.jpg (10KB, 224x250px) Image search: [Google]

10KB, 224x250px

>>61181554
HBM really is a meme.

Imagine what AMD could have done if they put their limited resources into better drivers and arch refinement.

Instead we got Mantle and HBM, fuck raja

>>61181633
wow I guess I'll wait :^)

Vega is not released yet and probably suffering from drivers not optimized for gaming.

>>61181699
>mantle
Literally the best API since forever.
>HBM
Literally the best DRAM since forever.
>>61181658
Lmao what the fuck.

>>61181750
It's suffering from nodrivers period.

>>61181769
That's a first for ati/amd

>>61181783
No, I don't mean bad/unoptimised drivers. I mean uarch features simply now working. At all.

>>61181760
>best api
They paid dice fucking millions to redesign their game around it and it gave +10% over dx11

C'mon anon, you aren't even trying

HBM was also rushed out with shitty drivers. According to AMD's justification "the one thing devs want more of, is bandwidth" was just flat out wrong, they want more SP/Cores at higher clocks.
HBM has a huge sunk cost that they will never recuperate.

>>61181554
so?
in a years time it'll be way above those cards
screenshot this post

>>61181554
why is it lower than a demo? this whole FE thing is weird

>>61181819
>a huge sink cost that they will never recuperate
I can say this about everything since R700 days.

>>61181934
Their gpu division saved them from the abortion that was faildozer, retard. Without it AMD FINISHED AND BANKRUPT wouldn't be a meme.

>>61181819
Wrong. Mantle was incorporated into Vulkan and became the heart of DX12. Obviously if you do not leverage the advantages of DX12/Mantle/Vulkan you don't get the "performance gains". You can write the same shitty code in DX12 as DX11, just more work. If you put that work towards optimizing the game, and improving parallelism, using compute shaders, and saturating the wavefront/warp to exploit ILP, newer APIs obviously have a lot to offer. Just because it isn't practiced in games that are incredibly large and unwieldy to port doesn't mean there don't exist inherent advantages to the API. Have you even read the Vulkan spec?

HBM had a memory controller capable of handling only half the bandwidth of the theoretical maximum thoroughput of HBM. HBM isn't targeted towards gamers you absolute dolt. It's literally designed to shrink the size of the card to near mxm sizes. It would do well in texture mapping, and it does. Fiji was limited by the render back end in most games, nothing to do with HBM except when textures got too big to fit its tiny 4GB space.

Also wrong about more cores, unless you're only talking about GPGPU. Fiji was at its limits and so is Vega at 4096 ALUs over 4 shader engines and both dies are very big. You need high res textures loaded live, need to go 4K or 8K rendering realtime? Memory bandwidth is absolutely essential. HBM makes it scalable without needing to put the controller at inane clock speeds.

But then again, why reason with a gaymur :^)

>>61182102
Dice literally got PAID to "leverage the advantages of Mantle" and it fucking sucked. Go look at the benches.

Low level API's are good, I'm not debating that, I'm purely saying for AMD to invest this much into it and get so little back (+10% at best when specifically targeted) is where it all went wrong. they should have used that time to deveop their shitty DX11 drivers that 99% of their userbase wanted. Mantle was a bad business decision.

>HBM isn't targeted towards gamers you absolute dolt.
Stopped reading right there, I'm pretty sure this is bait. you're a retard if serious.

>>61181819
Look no further than Doom Vulkan performance to see why the API is so good.

>>61182162
>HBM isn't targeted towards gamers you absolute dolt.
It really isn't. Believe it or not GPUs can be used to do things other than gaming, it's just another technology that could be leveraged for gaming. The JEDEC chairman even states
>“GPUs and CPUs continue to drive demand for more memory bandwidth and capacity, amid increasing display resolutions and the growth in computing datasets. HBM provides a compelling solution to reduce the IO power and memory footprint for our most demanding applications,” said Barry Wagner, JEDEC HBM Task Group Chairman.
Nothing to do with gaming bud, just saying that it is more power efficient and smaller than previous memory paradigms.

>Dice literally got PAID to "leverage the advantages of Mantle" and it fucking sucked. Go look at the benches.

OK? How can we be sure that it uses all the features Mantle in every single class and piece of HLSL? Reduced driver overhead, parallelized DMA means nothing to you? What about better interoperability with console low level APIs like GNM which run off of AMD SoCs? For the record, DICE also contributed to Mantle a massive deal. It's not like AMD just tasked all of their software engineers on this project.

>Between 1070 and 1080 in meme AMD shilled games
>1070 performance in Nvidia optimized games
>Pretending to be workstation card but only being able to compete with meme prosumer cards like Titan Xp - not actual workstation cards like Quadro.

This card is fucking useless.

AYYMD IS FINISHED & BANKRUPT

AYYMDPOORFAGS CONFIRMED ON SUICIDE WATCH

>>61182540
>>Pretending to be workstation card but only being able to compete with meme prosumer cards like Titan Xp - not actual workstation cards like Quadro.
I'd like to see your source for that.

>>61181554
who would have knew a workstation card performs just as bad as every other workstation cards!

meanwhile lets check what vega FE does on the market that amd create it
https://www.pcper.com/reviews/Graphics-Cards/Radeon-Vega-Frontier-Edition-16GB-Air-Cooled-Review/Professional-Testing-SPEC
oh shit 90% of the benches show it AHEAD of titan xp who would have knew!

>>61182624
>no ECC and certified drivers
>workstation
pick one

>>61182638
hbm a priori has ecc moron

>>61181633
>Vega at it's proper 1600 MHz clock will be GTX 1080 competitive at the very least
>1500 dollarinos
Wow what a steal

>>61182652
Vega FE has no ECC and certified drivers. Kill yourself asap.

>>61182638
Actual WX Vega comes later for much moneys.

>>61182678
and is actual workstation card unlike Vega FE

>>61182677
hbm 1 had ecc and data masking capabilities a priori but you couldnt use both of them
hbm2 has ecc and data masking and its enabled through the MRS registers
so next time you try to answer back like a smartass try at least to search a bit before you embarrass yourself further
also yeah which card has certified drivers at their launch? NOT A SINGLE ONE cause it takes time blender just released their CF yesterday

>>61182698
It doesn't matter if HBM supports ECC if its disabled on hardware level you retarded sack of nigger shit with down syndrome.

>>61181633
>water cooling

Enjoy your fried mobo's fags.


TEAM GREEN

>>61182705
oh god dont fucking answer back if you dont know shit please

https://www.design-reuse.com/articles/41186/design-considerations-for-high-bandwidth-memory-controller.html

spend some time and learn something

>>61181554
Wait for the dedicated gaming cards next month.

>>61182711
You do realize GPUs draw the majority of power directly from PSU by way of 6/8-pin connectors and only use PCIe for data?
This invasion of /v/edditors needs to stop.

>>61182597
the source is GN review, it literally gets smacked in 60-70% of the "professional work" by Titan, not to mention it got murdered in blender as well.

>>61182756
really cause pcper review here>>61182624
says otherwise

>>61182624
>>61182763
actually the score is 7 : 4 for vefefe vs titan in pcper, and they used differnt benches than GN and didnt even include blender bench.

>>61181554

I hope yoúr whore mother becomes paralysed in a car crash you subhuman /v/ermin.

>>61182770
they used spec which by any means is the gold standar on industry

id say its far more reliable than using a version of blender that kept crashing cause it didnt had the yesterdays CF

>>61182713
gddr has ECC as well, so what? just because hbm2 can have them enabled doesnt mean it has it, did you AMD market vefefe as having ECC? no, so it doesnt have it, if it would they would have plastered all over it.

>>61182753

That's irrelevant. Water-cooled components always have a high chance of failure, is AMD offering any warranty on the card? And what about my other components? What am I supposed to do if there's liquid from the GPU leaking on my rig?

>>61182785
anyway its competes with titan, literally its anyones guess in which bench which card can win. If you want consistent performance with certified drivers buy quaddro if you want cheap(by prostandards) fake procard buy titan or vefefe, performance is case by case.

>>61182803
exactly water cooling pc components is biggest meme ever, there is literally 100% chance that at some point your rig will start leaking and you just hope it wont destroy to much of the components.

>>61182803
Are you retarded or just pretending?
In any case, gaming cards will be produced by 3rd parties and they usually make different variants so you can select what suites you.

>>61182844
>>>/v/

>>61181554
In reality it's going to do worse than the 1070 in most games, as that benchmark is clearly very AMD friendly. Just look at how well the rx 580 is supposedly doing.

>>61182844
Not him but only custom loops are a meme, closed loops are exactly as fucking described, CLOSED and SEALED.

>>61182756
>GN
I'm not sure I believe them after the shit they pulled with Ryzen.

>>61182803
the fluid is non conductive and even if it was the voltages involved are so low that it's not going to cause damage. The chance of it damage your system is minuscule.

>>61182844
My CPU block leaked distilled water onto the back of the GPU for like two days. Literally the only effect was occasional system freezes I couldn't explain. After I stopped the leak everything worked fine.

>>61182868
yet they still manage to leak, a quick Google will show you the thousands of horror stories involving closed loop.

>>61182863
Well, then it's a good thing VegaFE isn't a gaming card you retarded /v/edditor.
Why don't you shell out three thousand bucks on a Quadro so you can masturbate to playing CS:GO at 320p on a workstation card?

>>61182785
>>61182756
SpecPerfView 12.1 uses only the software from 2013; Maya 2013, NX9, Creo 2, Solidworks 2013. The WX7100 literally competes with a M2000 in SPV12 but will readily compete with a Quadro P4000 on SPECapc which uses the 2016 versions of the software. Blender should be using the 2.79a version (which is actually OpenCL compliant) not the 2.78 version GN uses in his benchmark.

>>61181554
JUST WAIT PLEASE I SWEAR WE'VE GOT A GAMUR VERSION PLEASE

File: download.jpg (11KB, 298x169px) Image search: [Google]

11KB, 298x169px

>>61182787
you lad listen to master bratac and he tells you to know when to shut up

you have no clue what you are talking about

>>61182994
spec updates its suite only when they see that intel is in trouble..

shows why they never thought to change it up untill 3 weeks ago while intel was hacking through the numbers...

>>61182983
Titan isn't a gaming card now?

>>61183083
the target group of it isnt really the gamers but it can play perfectly fine

>>61183083
Titan is targeted to be something in the middle. Not a 100% gaming card but not a true workstation card either.
I'd probably compare it to Radeon Pro Duo as far as target market goes.

File: SHOLVA.png (770KB, 1024x576px) Image search: [Google]

770KB, 1024x576px

>>61183041
KREE SHOL'VA

>invest into expensive meme ram tech
>it doesn't do anything

lmao

>>61183592
Yes let's all be like nVidia and wait until g-ds will gift us new tech.

>>61183565
WE

Oh my god, so the latest AMD turd that got shilled and hyped to hell and beyond has failed to deliver yet again?

I am so shocked! I have definitely not seen this a dozen times before and nobody could see it coming a mile away!

#wow
#whoa

>>61183622
t. Jen Hsun Huang

>>61183611

So being way below year old cards on every metric while not even dreaming of touching the current flagships is what this exciting new tech has provided? Oh boy I am thrilled for this revolutionary tech! It totally paid off!

>>61183633
Yes Huang, you should be eternally grateful to ATi/AMD engineers for pioneering memory tech you're using THRICE.

File: 1451932503908.jpg (59KB, 675x450px) Image search: [Google]

59KB, 675x450px

>>61183643

Seems like those engineers also love pioneering the last place in performance.

>>61183651
At least they are pioneering something.

P O O
O
O

I N
N

G P U
P
U

>>61183643
Nvidia not being retarded like AYYYMD saw that the only benefit of HBM is smaller PCB package for stacking teslas up the ass. AYYYMD being the retarded poojets they are, decided to push it into gaymen market where it delivers zero (0) benefits for now.

>>61183674
Yes, nVidia is more retarded than AMD since they can't engineer hardware for shit.

>>61183680
Nvidia was behind HMC which was better than HBM but more expensive for the time. Enjoy your discount "technology" poojet.

>>61183700
HMC failed. Horribly.

>>61181554
>be AMD
>release card without drivers
How can you fuck up this badly? It would be better for their image if they delayed it.

>>61183730
It's a card for people who want to tinker with Vega as uarch and ROCm platform. Also for gamedevs.

>>61183680

That probably explains why current nvidia offerings are years ahead in performance when compared to amd's while consuming half the energy, right?

>>61183730
More like:
>Be AMD
>Release workstation and datacenter versions of card with the proper drivers for those uses
>Use placeholders for the game related drivers as you continue working on those for the release of the gaming cards
>Expect tech sites not to be retarded enough to benchmark the workstation cards for games using those placeholder drivers without some massive warnings

Then again this probably to be expected. Google and Facebook have together pretty much killed the market for ad funded news and review sites by gobbling up a lion's share of all the ad revenue so the only people left running these sites are the people too retarded to get a job actually making something (rather than writing about something other people have made).

>>61183779
Using PowerVR technique from 1994 is not that hard, and AMD uses it in Vega too.

>>61183784
They should have never released Vega FE for general public.

>>61183700

>salty that JEDEC told Nvidia to fuck off.

>>61183784
AMD themselves compared their card to Titan so everyone expected it to able game also.

>>61183812
>AMD themselves compared their card to Titan
In compute.

>>61183833
Why would you compare your workstation card with glorified gaming card?

>>61181658
lol literally Fury drivers confirmed

>>61182983
Do you even know what a workstation card is? They usually perform even better in games than the equivalent gaming card.
>>61183103
>>61183169
The first Titan one had some non gayming features, every single one after it was a 100% pure gaming card. You are retards who easily fall for the marketing.

>>61183843
Titan Xp is still a fully enabled gp102. Also FE is "prosumer" card, not a WX workstation one.

>>61183855
>Also FE is "prosumer" card
So which is it? Prosumer card like Titan or workstation card like quadros?

>>61183873
It's prosumer like OG Titan. Actual WX Vegas will be announced on SIGGRAPH.

>all this arguing over what Vega FE is or isn't

Just proves what a fucking turd AMD's marketing is for there to be this much confusion.

>>61183971
Let's not talk about Titan X and Titan X and Titan Xp.

>>61181554
well, there goes the dream of having a full AMD build. I guess they can't pull off another Ryzen on the GPU department. It is a miracle they even made Intel shake in their boots. Here is hoping that Navi will be the one.

>>61184004
They can, but it takes time, and money, and software engineers. And since GPU market is much much less profitable than selling EPYCs with absolutely stellar margins, they don't care about it that much to throw money at RTG.

>>61184020
In a way it makes me happy that AMD is finally getting their shit together and focusing on the things that matter. EBYN will surely bring them lots of profits, that is why I hope Navi will get enough budget to slaughter Nvidia.

>>61184035
If Navi is coming 2019, it's already well into development, and extra R&D budget from Ebyn sales probably can't save it now.

>>61184035
Vega will already fight nVidia the moment they write fucking drivers for it. For fucks sake why TBDR is not fucking working who the fuck though releasing not-Maxwell with TBDR disabled was a good idea?
Also if AMD takes at least 20% of the server market and 35-40% of desktop CPU market, nvidia's revenue will be pocket change to them.
>>61184044
It doesn't need to be saved. It needs even better drivers if it's actually MCMed GPUs.

This thread is excellent proof that /g/ knows absolutely nothing about gpus and AMD and can only regurgitate what equally stupid people on reddit say.

File: capsaicin roadmap 2016 .jpg (447KB, 3000x1688px) Image search: [Google]

447KB, 3000x1688px

>>61184044
>Navi is coming 2019
didn't the latest roadmap from FAD show it at 2018?
previous roadmaps showed it at early 2018 (although they also showed vega as late 2016/early 2017)
pic related roadmap from 2016
I think it should be nearing the end of it's development at this point and should be taped out H1 next year (since 7nm volume production starts at H2)

>>61184106
If Navi is anything sizable on 7nm it NEEDS EUV which means Q1 2019 at very least.

>>61181554
>gaymens anus . the ultimate shill site. no1 should listen to them on anything
>Vega
>relevant
Wow anon, a thread died for this?

>>61184020
>And since GPU market is much much less profitable
lol nope
I think vega was delayed just so they could target the deep learning market after seeing how well it worked out for nvidia

>>61184122
You should totally compare Intel's and nVidia's revenues.

>>61184116
glofo will offer 7nm DUV at 2H 2018 and EUV in 2019/2020
source: http://www.anandtech.com/show/10704/globalfoundries-updates-roadmap-7-nm-in-2h-2018

>>61184142
Yes, but as i said, if Navi dies are anything sizeable they need EUV to lower complexity to achieve acceptable yields.

>>61184132
not saying it's as profitable, but it's not much much less profitable and easier to get into than the server market which will be slower to transition to AMD
In other words GPU market is a better short term investment.

>>61184154
if Navi's scalability is MCM then it shouldn't be a problem

>>61184173
>In other words GPU market is a better short term investment.
nVidia can compete with AMD in HPC/meme learning. Intel can't compete against EPYC in anything but bribes and 4S+ systems.

Also margins on EPYC will be BEYOND stellar, it's fucking 4 Zeppelin dies (which have 80%+ yields) in one MCM.
>>61184186
That depends on how big base die is going to be.

>>61184198
>Also margins on EPYC will be BEYOND stellar, it's fucking 4 Zeppelin dies (which have 80%+ yields) in one MCM.
true, didn't consider that.

>That depends on how big base die is going to be.
with 7nm they could probably fit 2k cores in under 150mm^2
I think that's good enough for 4 die MCM.

EPYC exists to kick intel right in the shekels - they physically cannot build a 1S chip big enough to compete directly.

>>61184224
That's only 8k shaders. And if the rumors are true, AMD wants HPC niggers which means bigger base die to achieve some 10-12k ALUs in one MCM.
>>61184236
Well theoretically they can, but no one will buy it.

>>61184238

>Well theoretically they can

Well, in so much that single digit yields are possible but realistically for all of Intel's fab advantage the nature of monolithic dies means they aren't going to be able to build such a huge chip that doesn't clock really, really low. Plus you get the issue of cooling given how dense Intel's design is (whereas zen has a fucking huge package in comparison).

>>61184270
Yes, hence why EPYC has no direct competition, only bribes.
Now they have to write fucking drivers for Vega, apply MCM to Vega, and finally write drivers for Navi (i pity RTG software team, it's gonna be hell).

>>61184238
>That's only 8k shaders. And if the rumors are true, AMD wants HPC niggers which means bigger base die to achieve some 10-12k ALUs in one MCM.
well they could probably fit 48CUs in 200-250mm die. That still seems reasonable.

>>61184297
Yes, that's it. Now comes the hard part: drivers.

So, this means I can get vega for 249$?

>>61184454
No, that means you can't get one at all.

File: Captain kirk.png (860KB, 847x481px) Image search: [Google]

860KB, 847x481px

>>61183565
Indeed.

>ryzen flopped
>vega flopped
glad i didnt wait for either

>>61184563
your brain flopped

>>61184563
Hi Brian! Remember to bribe OEMs to prevent them from adopting EPYC.

Hmm, hopefully AMD put out the frontier editions so they can iron out bugs and optimise drivers, Ideally getting it sorted to by the time the RX cards come out.

>>61184675
FE literally exists for people who want to tinker with Vega uarch or ROCm platform and for gamedevs. That's it. Also it's only good for compute, at least now, since key uarch features are simply disabled.

>>61184616
Don't have to mate, AMD delivering such a turd has put them off anyway. It's like their entire range is an indian shit street.

>>61184840
Brain you should stop be delusional and start bribing OEMs ASAP.

File: _id1489488558_343178.jpg (71KB, 634x581px) Image search: [Google]

71KB, 634x581px

>>61181554
The Vega Frontier version was not optimized for gaming you fucking braindead nvidiot. That one would be the RX Vega which will be released in a few weeks and will have around 20~30% better gaming performance than Frontier.

Just wait for newer drivers. I bet AMD has a card under their sleeves and the cart will perform higher than GTX 1080ti.

>>61184935
They have the hardware. The drivers are lagging faaaaaaar behind.

>>61184935

I would hope so as I want a full on AMD build (already have ryzen) for my move to 4k.

>>61184926
>That one would be the RX Vega which will be released in a few weeks and will have around 20~30% better gaming performance than Frontier.
I really hope this is true... I mean 300W for a card that barely performs better than the 1070 and costs twice as much as a 1080? WTF are they thinking?
If RX Vega won't be <300W, can't compete or beat the 1080 and cost more or equal the 1080 costs AMD should stop producing GPUs at this point.
After what we've seen i have ZERO hope for RX Vega. I'll wait regardless but my expectations are low. I'll likey have to get a 1080 or used GPU or 1050Ti and wait for Volta or Navi.. if AMD doesn't give up after a possible disaster.

>>61185019
300W is TDP and we don't know how much power it'll chew with TBDR enabled in graphical/mixed workloads.

>>61181554
AMD should have done a rx 480x2/580x2 ages ago.
It would perform better and draw less power.

>>61185049
It would not perform better than actually werking Vega and you can CF 480/580 anyway.

>>61185048
Benchmarks of the FE showed approx avg of 280W during gaming with a ridiculously stupid spike up to 450W or so.
If RX Vega will draw the same amount of power it has to beat the 1080 at the same price point, otherwise they won't sell shit.
Imho this one spike to over 400W is extremely dangerous. If you have something like a 1700 OC @4.0GHz, Vega and a 550 W PSU you can and will overload it.

>>61185096
FE works like a fucking overclocked Fiji. Any non-compute tests are TOTALLY fucking irrelevant and Kyle or that gaymen nexus dude can shove them up their anuses.

>>61185106

No you don't understand - gaymen is the only possible use for a graphics card.

>>61185106
Gaming performance aside, that spike in power draw still is an issue if it's reproduceable. If that's the case there will be reports of burned PSUs due to this issue.

>>61185076
It literally performs way better.
Gamernexus has a line in its graphs for crossfire Polaris, but you must have missed it.

>>61185147
Spikes in power draw are totally up to drivers since Vega can hold ~220w power target, indicating it's ability to disable parts of SIMT.

>>61185173
>actually working
Can you read?

>>61185201
Implying we will get anything more from RX Vega?
Implying it will beat crossfire 480s.

>>61185183
Doesn't matter if it's a driver or hardware issue. If RX Vega goes to the shelves with that issue people WILL put those cards in PCs with PSUs that aren't top notch because that's how the majority is and they will eventually have a housefire.
Maybe that won't be an issue for RX Vega, maybe they will have fixed it, idk but unless they do something housefires will happen.

>>61185223
It's like you are TOTALLY obvious to how GPUs work. You are also unable to read the thread.

>>61185228
It's unlikely for this to be a hardware issues since RTG had all the time in the world to do testing and respins if needed.
I wish we had some clear uarch details about Vega, specially their kind of doing TBR.

>>61185238
No, I have those cards.
I know how they perform very well.
I was scared it would just micro-stuter like hell, like back when I crossfired Radeon HD 4850s.
But it didn't.
Ok it doesn't work on some games, but when it doesn't, said game generally is shit early access piece of crap that doesn't need that horsepower anyway.

>>61185289
I'm not talking about crossfire. I'm talking about you being totally obvious to features implemented in Vega. Said features are still not working.
Why the fuck did they launch a glorified compute board for consumers?
Hell if only I knew.

>>61185262
>unlikely for this to be a hardware issues
Doesn't matter. If the issue is present at launch there will be consequences.
>I wish we had some clear uarch details about Vega
>ayymd
>releasing details
This is why their marketing department needs to die in a fire. Nobody knows nothing about anything. We're all in the dark. Leaving consumers in the dark is so stupid.

>>61185321
No, leaving everyone in the dark worked wonders for EPYC. Who the fuck knew Zeppelin has AES-128 crypto engine built into each and every memory controller?
The less you know, the better.

>>61185305
I was waiting for Vega to replace both cards, and maybe buy 2 for the lulz, as my crossfire experience last year was pretty good.
Now, I'm very much reluctant on having 2 of these housfires running on my 1000W PSU.

>>61185337
>leaving everyone in the dark worked wonders for EPYC
they didn't release a fucking "pro" version of EPYC that failed monstruosly and kept secrecy for another month though

>>61185349
just don't oc them

>>61185349
Every flagshit card is a housefire, anon. You can't have fucking have fuckton of ALUs clocked really high without having gigantic power draw.
>>61185355
>they didn't release a fucking "pro" version of EPYC that failed monstruosly and kept secrecy for another month though
Ryzen was exactly that. One Zeppelin versus four Zeppelin.

>>61185366
>Ryzen was exactly that. One Zeppelin versus four Zeppelin.
Ryzen wasn't a complete pos housefire
What tf are u on about

>>61185366
>>61185364
Maybe I'm dumb, but I was thinking HBM2 and process improvements would at least limit the power draw to 225-250W per card.
But right now, I'm contemplating a 1080Ti or SLi 1080 for something that wont kill my PSU.

>>61185337
EPYC isn't a consumer product. Take Ryzen for example. Ryzen was a huge success but imho it didn't reach it's full potential. AMD took too long to release it and kept it too quiet.
How many people got sick and tired of waiting and have gone with a 6600K or 6700K, i wonder? How many of those builds have we seen and recommended here before Ryzen?
You don't need and should not reveal everything to the public, obviously but keeping it a quiet as AMD does is stupid if you ask me.
If RX Vega won't be a huge success after such a long wait and the innitial disappointment with Vega FE AMD will not only not sell their product but even worse, lose a significant amount of consumer goodwill which is disastrous for any company.

File: 1364573208543.png (453KB, 600x375px) Image search: [Google]

453KB, 600x375px

>>61184675
>>61184692
literally the only reason FE exists is because they wanted to make the Q2 target
even for pro-application I'd say it's half optimized since their claim of 175% faster than titan xp in maya was proven wrong
Vega FE is literally a half finished product

>>61185399
Zeppelin has a fuckton of redundant hardware we knew nothing about. Vega FE works fine as a compute board, no more, no less.
>>61185404
It's fucking 4096ALU chip with high clocks. Like the fuck, you can't make big and powerful GPUs sipping nothing unless you magically invent a way to throw fuckton of low clocked shaders since more shaders is inherently more efficient than clocking less shaders higher.
>>61185411
>lose a significant amount of consumer goodwill which is disastrous for any company.
Radeon is totally dead for almost every consumer since forever. No one buys Radeon. There's no consumer goodwill.

>>61185404
Nigga if u sitting on a 1000w psu it doesnt fucking matter that vega pulls a lot

>>61185439
>Radeon is totally dead for almost every consumer since forever. No one buys Radeon. There's no consumer goodwill.
This is either something AMDshills use a reverse psyop tactic or only thing I hear from ex-AMD neckbeards.

The 7XXX and 2XX did fine as far as perf and sales.

>>61185436
Vega FE works fine as a compute board and as a teaser for MI25 and ROCm platform.

>>61185442
I'm fucking weary of PSU and advertised power.
I wouldn't personally push them pas 70% of that.
It ensures durability and is more efficient.

File: 1482574053190.jpg (7KB, 267x323px) Image search: [Google]

7KB, 267x323px

>>61181554
Just wait

>>61185467
>fine
Simply doing fine is not enought when you're selling better hardware for cheaper. Like, it's totally not worth it to throw fuckton of money into GPU R&D if you're not going to outsell your competitor anyway. Fucking Evergreen failed to outsell Thermi.

File: power-witcher3.png (249KB, 1577x803px) Image search: [Google]

249KB, 1577x803px

>>61185096
stop being so tech illiterate. A millisecond power spike is not doing any damage to a PSU. A PSU rated for 500W continuous power can take a lot more peak power for short periods of time.

>>61185404
it's only drawing ~15% more power than a 1080ti and way less than 1080 SLI would. It's not going to kill your PSU.

>>61185505
It's doing suprisingly fine considering it's 4k shaders monstrocity.

>>61185484
>Like, it's totally not worth it to throw fuckton of money into GPU R&D if you're not going to outsell your competitor anyway.

Being retarded on the internet: the post

>>61185518
>let's throw fuckton of money to barely recoup R&D costs afterwards
GENIUS!

>>61185505
As I said, I want 2 Vega stuff.
One's just gonna be a regression from my crossfire.
Now I'm contemplating 700-800W total power draw.

File: 7523978463285.png (519KB, 762x647px) Image search: [Google]

519KB, 762x647px

>>61181554
Why is the Titan Xp so shit? You'd figure 256 extra cores would count for something over the 1080 Ti.

>>61185505
Enjoy your housefire.

>>61185546
It's only ~300 something shaders. That's not a lot of difference.

>>61182054
Now it looks like the Cpu division is gonna have to mop up Raja's mess

>>61185568
>mess
One compute board without drivers is a mess? Well whoa Instinct line is literally DOA.

>>61182624
But it's slower than the quadro p5000 which is a workstation gtx 1080. So it's slower than gp104 in all scenarios and consumes twice the power.

>>61185629
Quadro uses certified drivers with software profiles. FE drives everything on pure compute power. Actually WX Vega will be announced at SIGGRAPH.

Oh well they should've given FE boards to gamedevs and key ML partners and told the rest to Wait™ for RX Vega.

>>61181554
The FE has more memory, but its slower than the upcoming gaming card.

I'd imagine the top of the line gaming vega will sit close to 1080 in performance, but cost significantly less.

It'll also be a housefire so we'll get to remake memes from fermi era once again.

Wow, I actually thought the "AMD has no drivers" meme was finally dying, and then AMD goes ahead and releases a GPU with literally no drivers.

>>61185893

You can already get a 1080 for around 500, how much cheaper do you think vega will be? rx 480 owner here, but honestly do you think it's even possible?

>>61186362
>how much cheaper do you think vega will be
300$ for 1080ti level of performance

i wonder how fucking amazing these are for minig buttcoins.

>>61181554
t. gaymer

>>61186410
Shit since HBM2 trades latency for bandwith. And everyone who mines BTC does that using ASICs.
>>61186430
Vega is all about gaymen though. Too bad RTG software team is pretty incompetent.

You nvidia shills need to fuck off. In a few years AMD is releasing something big.

File: file.png (2MB, 921x2201px) Image search: [Google]

2MB, 921x2201px

>>61181554
I don't know

I remember when the 290x came out it was between the 770 and 780ti and now it beats both by a lot.

Its weird how Nvida cards just degrade so badly...

>>61186455
They are releasing something big in a month.
If they manage to enable TBDR, of course.
>>61186468
Kekler dies due to Maxwell being a major change in uarch.

File: 1497629707710.jpg (52KB, 568x612px) Image search: [Google]

52KB, 568x612px

>>61181554
IT'S NOT A GAMING CARD, RX VEGA WILL BTFO GTX 1080Ti FOR CHEAPER PRICE AND BETTER EFFICIENCY, J-JUST WAIT!!!!

>>61186508
You should probably stop shitposting and watch some Kanter and his video about tiling in Maxwell or something.

>500mm2 GPU
>Cost less than 300mm2 GPU

TOP KEK, AYYMDPOORFAGS

>>61186537
What? It's 484mm^2.

>>61186443
what, HBM latency is lower than GDDR5. The only tradeoff with HBM is it's more complex to manufacture and capacity limits are more of a thing.

>>61185684
They've already been doing that. My workplace has had a few Instinct machines for awhile now.

>>61186607
Is ROCm any good?

>>61186624
Good enough that we're heavily weighing our options.

File: 1482175164780.png (341KB, 1005x742px) Image search: [Google]

341KB, 1005x742px

>>61181554
>Lisa said wait till end of July
You guys aren't waiting. Bad.

>>61182162
>Dice literally got PAID
[citation needed]

>>61186667
Well that's unexpected. AMD/RTG are the last i've expect a working software ecosystem.
But well EPYC launch with
>EPYC ecosystem is ready
was also a pleasant surprise.
>>61186703
He's lying, since the idea of Mantle itself belong to DICE' lead programmer.

>>61182803
So since my computer needs to survive sub zero temperatures in the trunk of my car for hours at a time I cannot buy amd?

NvidIA it is

>>61183786
I love all the fags on here who pretend they know half a thing about anything to do with CPU/GPU architecture that wasn't just gleamed from some review site which was told to those losers by a marketing jew who half heard the jist of it from the pajeet they have designing the thing. BUY AMD STOCK, NVIDIA IS FINISHED NOW WITH THEIR RINGBUS MEMORY CORES.

>>61186828
How did you go from "water cooling is danger to components" to "AMD can't withstand sub-zero"??

>>61181554
>Review explictly explains who the card's marketed towards
>Explictly says to stop panicking and stop shitposting everywhere
>Comments on video are full of the exact same retardation seen elsewhere

He tried, but you just can't win against retarded gaymur manchildren.

>>61185471
uhhh... don't buy no-name bronze rated 1000w PSUs then? quality trumps power capacity every time. a quality PSU should be able to exceed it's own power rating easily. my platinum PSU literally says you can exceed the rating by 200w and only loose some efficiency.

>>61181819
Anon are you retarded?

DICE was the one that created Mantle, not AMD. AMD just implemented drivers and did marketing for it. Johan Andersson, DICE's technical director, literally said in an interview that he basically did everything and then went shopping for partners to put it on a platform. After he experienced development on consoles he wanted a similar level of control on PCs, so he designed a spec for a new graphics API and asked for hardware vendors to join him. AMD was the only one who said yes so they got a free "feature" they could advertise.

>>61185439

>It's fucking 4096ALU chip with high clocks. Like the fuck, you can't make big and powerful GPUs sipping nothing unless you magically invent a way to throw fuckton of low clocked shaders since more shaders is inherently more efficient than clocking less shaders higher.

It's not doing any more work in games than a 1080 (actually less since it's slower in games) but somehow the efficiency is completely shit. I bet the front end is the limiting factor and the shaders are mostly idling in games.

>>61186885

Of course they say that because they want to get free review samples from AMD in the future and AMD has been shitlisting reviewers left and right lately.

>>61186915
No shit, since TBR and primitive shaders are disabled. FE shipped with slightly modified Fiji drivers. I guess it's stable enough.

>>61186929
>AMD has been shitlisting reviewers left and right lately
Source?

>>61181554
I've waited this long. I can wait a bit longer until AMD release the RX cards and have their driver issues worked out.

If Nvidia is still better after that I guess I'll buy a Geforce.

>>61186944
Have a wild guess why our (((friend))) Paul Alcorn suddenly included Ryzen into his "best gaymen CPUs" list.

>>61186962
Because it factually is.

>>61186962
I asked for a source, not something pulled out of your ass.

>>61186974
The last one had only (((Intel))) processors and caused gigantic shitstorm.

>>61186993
Yea, because they factually aren't

>>61181554
>>61181572
Why don't they just release the drives together with the hardware? why are they so retarded and enjoy sabotaging their own marketing so much?

>>61187014
Because they have very small software department that can't keep up.

>>61187014
AMD is secretly run by masochists

>>61187014
Because they have to launch both gaymen drivers for RX Vega and ROCm compute kernel for instinct at the same fucking time with a relatively small software team.

File: 1496695346022.jpg (118KB, 879x758px) Image search: [Google]

118KB, 879x758px

>BUT IT'S JUST THE DRIVERS GUYS
>WAIT FOR DRIVERS
>IT WILL RISE THE PERFORMANCE TO %200

>>61182787
Moving the goal post general?

>>61181554
Why are you treating gamersnexus with such authority? They aren't particularly respected nor are they as well informed as others.
Wait for anandtech or toms to do an actual thorough review.

>>61187014
probably because this is not a gaming card, and gaming on it is a back burner issue.

>>61187048
Maybe they should stop being fucking retarded jews and just hire better developers. Raja even said drivers are harder to make than the hardware and yet it's a mostly EE company

>>61182878
They are second-tier tech journalists, plain and simple. I'm not sure why /g/ feels the need to circle jerk their articles.

File: JEWISH-INDIAN-MAN.jpg (20KB, 236x354px) Image search: [Google]

20KB, 236x354px

>>61181554
>Just wait for drivers
>it's not a gaming card
>Just believe in AMD
>I swear it will be better than a 1070

>>61187094
They need money for more (and of better quality) code monkeys. Money comes with EPYC. I guess we have to wait and see.
>>61187131
Yes, you wait for gaymen drivers.

>>61183971
>>61183980
Ultimately when Nvidia makes these mistakes they are overlooked, when AMD does they are exaggerated.

>>61181819
um... try closer to 100-150% more performance.

granted this was a shit cpu with a good gpu going from 20-40fps average to 70-90fps average

If you had the hardware, its likely that it gave you far less of a boost, but fuck me was that a massive fucking boost it gave to shit hardware.

>>61187158
Because it's nVidia! They released drivers that fried fucking video cards and got away with it. Also 3.5.

>>61182054
amds gpu division during bulldozer ran at a net 0 income if not loss.

You can argue that their gpus on the apus made them worthwhile and ill agree but net loss is net loss.

>>61187176
>falling for the 3.5 meme
It was 4gb just 500mb slower ram.
Still 4gb.

>>61187190
Last 500mb were slower than my fucking system memory.
Fuck Huang.

>>61187146
Companies work on debt, they should just take the plunge

>>61187190
You know, nVidia lost class action suit versus them.

>>61183845
no, pro cards perform universally worse than their gaming equals in both pro applications and games.

The reason you get a pro card is dedicated dev time if shit breaks and a guarantee that if you have an approved system, shit will not crash.

The reason nvidias are faster is this.

nvidia makes one monolithic die that they use for the highest end gpu, usually the top end quatro
nvidia uses failures from that to make a lower end quatro
nvidia then makes a titan out of the ones that are further failures
then nvidia puts out a ti under top titan at the time.

when gpus are equal, the gaming driver/card pulls ahead due to clocks or optimizations for games,

case in point, for a long time nvidia's top gpu was a 5000$ quattro that was 20% bigger than their titan, but only 5% faster.

>>61183169
that was titan 1, current titans have both double and half precision gutted to the point that a 560 amd is near 50 times better than it in said precisions.

>>61187358
Titan 1 was a nice card for it's niche. Too bad Huang hates people that want compute for cheaper.

>>61183020
if the people looking at it are right, the driver is in fallback mode right now, leveraging none of the improvements in the gpu.

This may really be a wait for the drivers situation, and not the usual 'fine wine' shit

>>61187378
At least it's stable enough to perform benchmarking.

>>61184054
not drivers, firmware that keeps the fact its mcm away from the developer's ability to fuck it up.

>>61187405
From the way it seams, so much needs to be dealt with on the driver side that fall back is literally telling everything 'treat me like a fury'

If anything this is fantastic, as even if the drivers fucking blow out the gate and shit is broken, the gpu will still function at least at fury x levels for older games.

Its honestly really nice to know that when they are going to have a massive driver bullshit pile to deal with, the card will still function adequately.

>>61187080
Anandtech and Tom's are the same entity, and http://www.purch.com/experiences/intel/

File: baby.jpg (7KB, 205x200px) Image search: [Google]

7KB, 205x200px

>>61187080
>gamernexus aren't respected because I don't like their seeing facts.

>>61184198
honestly for the gpu, I think they are going to go a standard brainbox that may have some shaders and shit on it just so they can sell a 1 chip version, then mcm shaders/compute units alone and attach them to the die. depending on how small they can make the computer units and how mcm able it is, you could see something like 1 brainbox and 10+ modules all connected to it, or possibly 4 standard modules+hbm

the gpu side is interesting now that we have seen the cpu side and what infinity fabric should be able to do.

>>61187500
t. Steve Burke

>>61187515
What will (possibly) happen in Navi is the idea of Sideport turned up to 11.

File: amdfag.jpg (3MB, 8624x3896px) Image search: [Google]

3MB, 8624x3896px

>>61187500
It's FAKE NEWS. AMD ALWAYS WINS

So, RX Vega confirmed? I'd buy it if it was $400, but I doubt it will be that low. I don't even know what's the situation with drivers.

>>61187605
>I don't even know what's the situation with drivers.
Vega is cosplaying Fiji right now.

>>61187555
https://www.twitter.com/NVIDIAGeForce/status/723633222450593797

>>61187500

Gamersnexus are pretty retarded at times. My favourite was them slapping a 9590 on a board not really designed to handle it (the msi 970) and they werel eft scratching their heads why the 9590 was posting lower minimum fps compared to its lower clocked (and still 8 core) siblings when the 9590 is flatout - from a clockspeed point - the superior chip to the likes of the 8370.

>>61187443
>this is fantastic
Really? It's fantastic? Literally months and months ago when the first AotS benchmarks leaked (GTX 1080 level) the excuse was the same. So even if this is true, AMD has done absolutely no work on drivers since then and the actual launch is like 1 month away. And that's "fantastic" to you? It's a fucking disaster and odds are you won't see your supermagic driver any time soon either, considering performance hasn't budged from where it was fucking months ago.

>>61187768
Big Maxwell was also delayed several times yet no one complained.
Why?
BECAUSE IT'S NVIDIA HAHA

Unfortunately for AMD, Nvidia are rolling in money and able to spend it on R&D.

Heck they are selling auto companies 5 cent tegra chips for hundreds of dollars.

>>61187851
Cool yet their revenue is still peanuts compared to Intel and AMD targets Intel's most profitable markets.
GPUs are irrelevant.
Meme learning is almost irrelevant.

>>61187880

This.

To give some context and to show how (relatively) little the gpu market is worth to AMD even during the bulldozer years it was the cpu side of the company that was keeping AMD in the black. The gpu division has effectively been a money pit.

>>61187880
What about the 100million AYMD gpus in consoles?

>>61186362
It depends on what amd wants to do. remember, polaris as amd said, is there 100-300$ solution, I don't think amd will forgo the 300-500$ market completely.

as for the gaming card, I fully expect 1080 minimum at worst but fitting in a halfway between 1080 and 1080ti as what I expect. a fairly small 15% boost through drivers making the card work properly would already put it there.

now on the topic of cheaper, we could just be getting bottom binning and everything worthwhile is carted off to the pro market or accelerator market. If that's the case, literally anything amd sells to us would be considered recoup, and even 1$ would be better than just tossing the die, we could very well see amd undercut the fuck out of nvidia with a 1080 equivalent at 1070 prices, or a 1080ti competitor at 1080 prices.

Also remember the deus ex demo? where they only used 2gb of vram but were able to pull off smooth gameplay? this would allow amd to use 4gb of vram and effectively halve their hbm cost and further drive the price down with a very minimal downside currently. imagine if there was a 1080ti card, but you could get it with half the ram and next to no perceivable down side for 50-70$ less and this the base card was already at 1080 prices.

amd is doing some interesting things that will let them drive the prices down hard. and this isn't taking into account that amd may make a loss leader card hoping to retake market share so their apis get more widely used, just so nvidia can't directly fuck with amd as easily through gameworks.

>>61187933
Very slim margins.
>>61187923
At least Vega will make nice iGPU IP, well, the moment they enable FUCKING TBR.
LIKE HOLY SHIT.
THAT'S THE MAIN FEATURE
DO IT RAJA

>>61181658
That's a joke, right?

>>61187946
>a fairly small 15% boost through drivers making the card work properly would already put it there.
It'll be much much bigger. The difference between kekler and maxwell was what, 30% with the same tflops rating due to tiling afaik.

>>61187965
When people said it uses Fiji fallback mode it means it really uses Fiji fallback mode. It cosplays overclocked Fury X.

>>61187799
Big Maxwell (i.e. 980Ti) came out right in time to kneecap its competition (Fury X) and was a good product from the start. Its launch was so well executed one might even think they knew exactly what they were doing and when to do it.

The issue with Vega isn't the delay anyway, it's the performance which hasn't changed from supposedly "early" boards and Fury drivers from 2016. That was a plausible excuse back then when a release date wasn't even announced, it's not so plausible today with FE out in the wild and RX coming in 28 days. This is really quite shitty, if AMD doesn't get their shit together I dread to think about how NVIDIA will price Volta. Even if this is a software issue it's going to take AMD a long time and multiple releases to fix, if their track record is anything to go by.

>>61188029
Big Maxwell means GM2xx family. Small Maxwell was GM107.

>>61187974
no, fucking god no.
nvidia fucked the drivers on the older cards or just refuse to optimize them, look at halo wars 2 beta that was dx12, most of the driver control was out of nvidia's hands, and the 780ti was still playing ball with the 1060-1070 range, but in every other game it's relatively shit.

its shit like this that means you can not judge performance upticks based on nvidia implementations.

that said, I said I still believe, with everything vega isn't currently doing that is in there, 15% should be a reasonable jump.

>>61188050

IIRC in the last few days Nvidia has actually delivered on the fermi DX12 drivers they promised over a year ago.

>>61188076
And how's the performance?

>>61188045
Pretty sure 980Ti is GM200 though, what are you on about?

>>61188092
Titan X (Maxwell) was before 980ti. GM2xx family was delayed several fucking times. Guess why.

>>61188104
The problem with Vega doesn't have anything to do with delays, why the fuck are you talking about delays? Do you think they'll delay Vega even more despite announcing a release date or what?

>>61187001
best cpus for the money are hard...
amd isn't better than any of the cpus on the list if you look only at games on fresh boot systems.

but the issue is if you are only getting an i3 or an i5, you likely aren't getting a 120/144 hz monitor, and the extra cores/threads amd provides will be far greater overall improvements then the higher highs,

for the money you go this list

the pentium, whatever one clocks the highest and is less then a 1500
the 1500
the 1600
the 1600x (notable exception on its inclusion because it's only 50$ more and negates any necessity to overclock)
the 1700

and if you only game on the system or demand 144, the 7700k oc to 4.5+, if you don't oc, the 1700 would be better even if you don't oc that either.

>>61188124
Vega was already delayed, anon. It was late 2016-early 2017 on older roadmaps.

>>61186954
I got a 1060 6gb on a 280x rma, I was going to get a version of vega... but fuck me, normal computing the fans don't spin, in games I can't demand more and make it go under a stable 60fps, and because it uses so little power, It does not heat my room up...

I may very well wait for navi unless amd has something they can show me that demands my attention.

>>61187452
editorial wise on toms they are intel biased, however on the review side, they are fairly good with some head scratching choices on why they pick some games in particular.

>>61188137
I know, but that's not the issue. The issue is the underwhelming performance we're seeing on a product which has already been launched and the fact that there hasn't been any improvement compared to leaked results from 2016 which have now turned out to be accurate. This can only mean 2 things:
>AMD has done absolutely 0 work on drivers for 6-7 months - unlikely, especially since the Vega driver versions are actually recent and it's not plausible for them to sit on their hands doing nothing for half a year
>this is close to what Vega is really capable of (in current games)
The performance is meh in DX12 and Vulkan too where drivers are thinner, it's not like you can shove the blame on AMD's DX11 drivers being shit (again).

>>61182819
depends on the application, seeing as these card classes, pro cards, are certified to not crash yo see what card performs better for you in x application and get it.

>>61188268
Have a wild guess why nVidia delayed big Maxwell and look closely what Vega brings to the table.

>>61185076
>and you can CF 480/580 anyway.
not in an itx case you can't

>>61188291
NVIDIA had no reason to release big Maxwell for gaming since 970/980 were already cleaning house as AMD had just rebranded R9 200 into R9 300, essentially releasing nothing ew. They released Titan X as a halo product and 980Ti in order to compete with the impending launch of Fury X, once that was actually necessary.

So what are you saying? That Vega has been kept back on purpose despite AMD being slaughtered by Pascal for over a year? It's irrelevant anyway, the performance of the launched product speaks for itself.

>>61187176
they fry gpus yearly.

>>61187226
yes and no, the 4gb was not what they lost it over, it was other specs where they couldn't fuck with the meaning they lied about.

>>61188357
I am saying GM2xx family was delayed due to motherfucking tiling requiring driver implementation you mouthbreathing moron. Same with motherfucking Vega.

>>61188384

ROPS and cache iirc.

>>61188389
Vega has had literally half a year since the first leaks for tiling implementation, they're still not done despite literally releasing a Vega card with a "gaming" mode already? Is that your grand theory?

>>61188414
>Is that your grand theory?
Yes you cockmunching retard. It took nvidia several month to bring tiling online you moron.
>"gaming" mode
That does fucking nothing.
3/10 bait made me reply.

>>61188076
>IIRC in the last few days Nvidia has actually delivered on the fermi DX12 drivers they promised over a year ago.

It's easy to deliver on that when you have to implement all of it in software as none of your cards support it in silicon.

>>61188473
>Fermi
Thermi had a proper hardware scheduling.

>>61187500
>>61187741
>putting duct tape between an m.2 ssd and it's cooler and wondering why it overheats
my favorite

>>61188442
>It took nvidia several month to bring tiling online you moron.
NVIDIA had GPUs with tile based rasterization in the gaming market before anything based on GM200 though.

>>61188570
I am saying nothing about motherfucking GM200 you brainlet. I am talking about GM2xx family.

>>61186898
there was a 550 watt psu that could be pushed to an 850 watt for everyday use and not suffer.

brands matter far more then you may realize.

>>61185467
the 7xxx sold like shit, the 2xx sold like shit, the 5000's and the 4000's relative to how good they were, sold like shit, the 6000's that brought the cost of performance down, sold like shit.

amd has sold like shit for fucking their entire time of having ati, for honestly no good reason.

>>61185530
they dont recoup, but the ati purchase was more about having a complete package then it was about being the best, their gpu research and development improves their apus.

>>61187500
It's not because I don't like their results, it's because from the very onset (they title themselves gamersnexus) they position themselves to look at tech from a very particular perspective. That's fine, but they make broad generalizations about tech in general from that limited and flawed vantage point.
Also, they employ questionable methodologies and display a lack of expertise that other review sites don't.
>>61187452
They are not the same entity. Purch is a conglomerate that owns both, but they each have discrete staffs with different resources, aims, and skillsets.

>>61188652
Also chipsets for Opterons.

>>61188268
>tertium non datur
There are a lot more reasons why AMD hasn't shipped the gaming driver for the Forntier Edition.

There is a very good chance the driver isn't fully stable yet and that they didn't bother testing and optimizing it on the Frontier Edition.
The last thing the driver team needs 1 month before the reveal of RX Vega is to test and support the driver on a different product that nobody is even going to seriously use for gaming.

>>61186402
likely 1070 for 1080 performance cut down
cheaper than a 1080 for the in between 1080 and 1080ti, and if there is a card that has 16gb of ram 1080ti prices for not getting dicked by to little vram in the next 3-5 years.

>>61186537
if you look at the dies as garbage they couldn't sell to enterprise, you sell them as loss leaders to consumers to recoup market share.

they could very well be cheaper then you are imagining.

>>61184926
its not optimized for any shit really.
Even a P2000 win in majority of scenarios.
Its a useless card.

>>61188593
What the fuck are you talking about then? Maxwell v2 i.e. 970/980 came out like 1 year after the last Kepler card (780Ti), that's quite standard for new architectures, what delay? Pascal came out like 1 year after the last Maxwell card too, that's simply standard. Where is this delay? You know, don't even bother replying, you're obviously not capable of holding a coherent conversation.

>>61188663
It's highly unlikely for AMD to release official performance demos and promo videos featuring Vega playing games if they're completely missing the implementation of essential features and performance is much lower than the finished product (what's the reason to "promote" and demo something if you're just making it look bad?), but believe what you will. It's not really a different product either, it's not like RX Vega is going to have a completely different GPU than what they've used for FE either. I guess we'll see in a month if the same Vega GPU is magically 30-40% faster in order to beat a 1080 Ti while drawing the same amount of power.

Raja claiming everyone who measured the die size so far has gotten it wrong.

https://twitter.com/GFXChipTweeter/status/881370308694822913

Closest perfect square to 500 is 484. Up to you if you want to believe Mr. "SLI 480's".

File: 1360368220235.jpg (80KB, 720x720px) Image search: [Google]

80KB, 720x720px

>>61188816
>you're measuring it wrong
AHAHAHAHAHAHA
Just fire the streetshitter, this is getting embarrassing.

>>61188807
What they demo and what they ship to customers are two different things.
Right now the Frontier Edition essentially runs in legacy mode like a Fury X because that's a well tested and stable driver that's ready to ship.
The RX Vega driver that takes advantage of the new architecture will only be shipped to customers when RX Vega releases.

>>61188903
They better have that shit 90% finished by now.

>>61188921
We don't know that. At least they demoed HBCC working.

>>61188632

You couldn't be more wrong. They were neck to neck for years in terms of sales.

Nvidia only began to oversell ATI/AMD when Kepler family came out and strengthen that lead with Maxwell (mainly due to the 970).

>>61188936
Wut? Fucking thermi outsold evergreen.

>>61188816
Raja, wtf are you talking about? Do you mean that HBCC shouldn't be included into die size measurement?

>>61189047
Those previous measurements did not include the HBM modules.

>>61189047
What? Die size =/= package size.

>>61189085
>package size
There was significant shrinkage.

>>61189047
>Die size: 25.90mm x 19.80mm (GPU only, not including memory stacks) Area: 512.82mm2

The streetshitter claims it's close to 484.

>>61188936
no, with the 4000 and 5000 amd was just about 40-45%, and everything after that was downhill despite being better or better priced, usually both then nvidia.

Not to mention amd made net losses most years, some only turning 10 million in profit, while nvidia's shittiest years are 10 times what amd was in proffit.

>>61189359
I honestly think Huang can release literal shit with nvidia stickers plastered all over it and it'll still manage to outsell AMD's best.

>>61189414
Thermi vs Evergreen?

>>61189414
that was the 200 and 400 line against the 4000 and 5000, since then nvidia hasn't put out abject shit, just usually shit value.

>>61188807
Don't ask what happened to GM104 and GM100

>>61189602
They promised Q2 2017 for Vega so...
I guess that was the only way.

>>61189636
that seems to be frontier edition
just an attempt to not piss off investors
I have no idea why else they would release a card with half broken drivers

>>61189602
inb4 they go with Catalyst

We will see how it really performs in about a 5 months to a year when they fixed the trainwreck that are Radeon launch drivers