Wait For Raven Ridge Thread
Where the FUCK is it, AMD?
This Bristol ridge processors are NOT what we need.
>>62321441
Vega is not ready yet.
It causes all the delays.
Well, probably.
Holiday season.
>>62321581
I keep seeing this posted, but never really any source or reassurance that it will come this holidays
>>62321638
Lisa Su said so at least 3 times in the last 4 months.
>>62321656
She also said Vega would be disruptive.
But it's a dud.
R600-tier trash.
>>62321731
wot
Vega is shit.... for games(supposedly because of shit drivers)
You can see that they are all sold out BEFORE reaching the market because Miners and researchers are eating them up like crazy
>>62321800
>researchers are eating them up like crazy
What?
Radeon is useless for anything because no CUDA.
ROCm is immature as fuck.
Oh, wait, AMD is partnering with Baidu for ML shit.
I guess chink slave labor will write the libs for AMD.
But niggers still need to port fucking TensorFlow.
WE WANT RAVEN RIDGE
WE WANT APUS
WE WANT RYZEN + VEGA LAPTOPS
REEEEEEE
>>62322093
You will get one Q4 2017.
Chinknovo will probably refresh AMD-powered chinkpads somewhere late Q1 2018.
>>62322116
That would be amazing honestly.
I need muh low powered penguin terminal.
AMD respects muh freedoms in regards to graphics drivers and xwayland + gpm support.
3 premium laptops are all that's needed for Raven.
11-13" zenbook (around $600)
13" Dell XPS high end $1000
13" HP Spectre $1200+
>>62322184
We need a thinkpad X**(13") with Oled.
>>62322184
You will have these, RR targets premium laptops first.
>>62322174
Ye, AMDGPU+Radeonsi is bretty fucking good nowadays.
They've just pushed DSBR implementation for Vega.
>>62322219
>OLED
I too love burn-in.
>>62321731
It's not a dud, they just fucking shipped it with the drivers half finished and with a key new uarch feature disabled.
Wait for them to actually fucking implement primitive shaders before evaluating what Vega is or is not.
>>62322256
Well they've recently pushed DSBR implementation into Radeonsi.
I guess they're working on it.
What do you guys expect of Vega mobile? Is GTX 950 levels too much of a stretch?
>>62322289
They also said in the same commit that it's and I quote: "tuned for Raven Ridge, don't know if it works for Vega 10"
I wouldn't be surprised at all if most driver effort went into getting the APUs in shape, they'd make AMD far more money and market.
>>62321638
>>62321656
Yeah, that info is official, their guys keep repeating it now and then, when speaking somewhere, or in PRs.
>>62322116
OEMs have been slow when launching AMD latops though, so the selection this year will likely be limited though.
>>62322342
You mean iGPU or mobile Vega11 (read non-shit bins)?
>>62322349
APUs still use Vega, just less ALUs and smaller GPU in general.
Vega iGPU will perform nice given DSBR and at least somewhat sizeable L2$.
>>62322396
I mean the RR iGPU
>>62322381
That's why RR targets only premium laptops for launch.
Mainstream shitboxes are probably Q1 2018 along with Ryzen Mobile Pro and AM4 RR.
>>62322411
It's 704 ALU tops.
I don't know anything else.
If L2$ is anything sizeable it'll perform very well.
DSBR saves a LOT of memory bandwith by tiling and fetching into L2 instead of main memory.
>>62322413
AMD can target what it wants, but I'm afraid it will all be decided by OEMs. And you know how the bastards love to shit on AMD and take advantage (shit laptop parts, high prices).
>>62322396
APUs use Vega but the way memory management works on cards and APUs is completely different(CPU fighting GPU for resources) DBSR would also have a much bigger impact on memory bandwidth constrained platforms like laptops.
>>62322447
Did you miss AMD's little Computex presentation or EPYC launch day?
OEMs are selling their assholes to AMD.
Literally.
>>62322448
It's unified memory.
Though I need more info on HBCC. I wonder if it can interact with CPU pages.
But yes, DSBR will save a lot of BW by fetching shit into L2$.
>>62322184
Man, I need me some $800 great value/$ RR laptop
13"-14"
1080p IPS
NO TOUCH BULLSHIT
Good build quality: backlit keyboard, good trackpad, no flex. Small bezels, good plastic or aluminum.
Top of the line RR APU
Sleek design, but not too thin.
8-16GB DDR4
250GB SSD
8+hours battery life
That's all I want and need right now.
>>62322342
More like a 940MX
>>62322510
I want something with an IBM thinkpad keyboard and a 4:3 screen.
I wonder how many tris/clock will RR iGPU chew.
>>62322483
>OEMs are selling their assholes to AMD.
Of course I did. It was not on Computex, but on ~20th june, btw (I work in this field, so assume I am not clueless).
Servers mean nothing for laptop OEMs, you know. Also, the support shown there was mostly in the form of words - but those are cheap, while time will show if they are also willing to put effort in marketing and selling Epyc servers. They could all too well just put few token systems into catalogue but then doublecross AMD and just making sure only Intel stuff sells because incentives and so on.
But in any case, notebooks are different world.
>>62322615
I think he means the short laptop and Ryzen Mobile demonstration they made.
>>62322615
Since AIB (anything but Intel) is a market in itself right now, don't worry, these are not token systems.
Heck THREE hyperscalers are doing rollouts H2 2017.
For a platform this fresh this is a success.
>>62322539
4 shader engines on 12CU? Way overkill
>>62322652
Vega scales with ALUs and not FF units only.
Well, not now.
But when it will work...
...it would be nice.
>>62322644
>For a platform this fresh this is a success.
We can't be certain until we know there are healthy volumes, really.
>I think he means the short laptop and Ryzen Mobile demonstration they made.
I don't think it was anything noteworthy. We'll get to see later how many design wins they are and more importantly, if they finally make AMD first class citizen - that will be the thing I will be looking for most.
Remember, the best yet APU for laptops (Carrizo) got "premium" laptops with FUCKING SINGLECHANNEL. And not just one module unpopulated, it was with impossibility to use dual channel at all. The same chassis with Intel CPU had two sodimm slots. See Envy x14 or how was that PoS called. The Lenovo gaming laptop was also shit, it had single channel and dedicated GPU. Crap.
I fully expect those shitheads to mysteriously cripple Ryzen Mobile too.
Oh BTW, HP also crippled Carrizo by disabling its dynamic turbo (the ability to go above TDP for limited period of time). By default, the chip can go to 25 W for 180 seconds to be more effective with turbo, but HP activelly disabled that and locked it to 15W all the time.
Of course, their Intel laptops are allowed to do this with their Core 15W U-chips. Don't assume any goodwill from these companies just because Ryzen is good.
>>62322730
Nah man, AMD ain't playing this time around.
Although that Alienware A51 with threadripper/ryzen has some pretty shitty design choices.
>>62322730
I sincerely doubt that is gonna happen with RR.
>>62322787
Anything Alienware is a collection of shitty design choces t b h.
What matters is that Dell of all OEMs licked mommy's butthole clean enough to get early access to jewripper.
>>62322804
I bet that part of that is because of the name "Threadripper" OEMs love that shit on their marketing and PR campaigns
>>62322289
>DSBR
Fucking fat lot of good that will do when Vega10 is still front end geometry bottlenecked.
>>62322804
>I sincerely doubt that is gonna happen with RR.
I'll only believe it when it happens. You know, I happen to hate conspiracy theories shit with exceptional disdain for them. But if I wasn't like that, I would be completely convinced that Intel is paying them for crippling AMD, again.
>>62322837
At least it's something.
>>62322833
So what will stop OEMs from marketing Ryzen Mobile?
>>62322876
Intel blood money
>>62322876
Intel subsidies. For marketing and other expenses, you know!
(Did you know Intel gives them money and helps funding promotion of laptops? They could just attach some strings to those deals, if they didn't already.)
>>62322876
>At least it's something.
Well, it'll help make the results more dramatic when Vega is no longer choking on triangles, so at least there's that.
>>62322876
What? I meant it's good for marketing, the. Name is just a plus.
Intel has historicay brived OEMs to gimp their AMD products, so probably that.
>>62322909
Neither mindshare nor bribes stopped Ryzen. Its selling well and is appearing in OEM boxes.
So what will stop RR? Another billions of dollars in rebates?
>>62322913
They should call the next Crimson release "detonator6".
>>62322781
>By default, the chip can go to 25 W for 180 seconds to be more effective with turbo, but HP activelly disabled that and locked it to 15W all the time.
>Of course, their Intel laptops are allowed to do this with their Core 15W U-chips.
I forgot to point this out, but you can imagine how much did this "help" in reviews (yay short bursty benchmarks that take less than 2 minutes, they absolutely love Intel's 30W turboing). Almost as much as that single-channel did to properly emphasise the APU's GPU performance.
>>62322979
>Neither mindshare nor bribes stopped Ryzen.
That is because it is selling in retail to DIY computer builders and small assemblers that work from retail-bought parts.
You can't bribe every single end-user, while you can easily bribe the 5-10 notebook makers.
>>62323029
Ryzen Pro has several design wins.
No one fears Intel anymore because the likes went a little bit TOO far with price gouging chasing 60%+ margins.
Meanwhile, those 8th gen 15W quad-core i7's are looking good
> turbo up to 4GHz (1.8GHz base)
Too bad they didn't update the graphics, it's the same "meh" Kaby Lake HD 620.
>>62323029
Anyway, don't take me wrong - few people will be celebrating as much as me if AMD makes it big. But it will yet be hard and uphill battle and nothing will come to them easily, more like everything will still be against them.
I'm a huge fan and always sticked with them since 2002 or so. All time Radeon, Athlon, APU since. Writing this from a Richland chip.
>>62323050
Limited ones. for example with HP they only got one line (Elite 705), while Intel is in like 10x as much or more SKUs.
>>62323060
RR will have higher base and all-core turbo at the expense of lower single core turbo.
>>62323072
>Limited ones
Still something, given the platform is fresh out of the oven.
>>62323060
Too bad that turbo lasts for a grand total of 2 seconds.
>>62323060
The dynamic maximum power was raised to 44 W though to make the high turbos and 4 cores possible, from 19 W on previous 15W chips. Like, LOL 15W, excuse my banned acronym.
It will be severely limited once it heats up and runs out of the burst reserve.
>>62323075
We don't know that tbqh. Last leaked I've checked, RR barely boosted up to 3.3GHz and the Vega graphics didn't even beat the HD 620.
>>62323094
AMD already had the 705 design win with Bristol Ridge Pro btw, since last year. Same goes for the Lenovo SFF machine.
>>62323113
And Zen ES samples were clocking at 3ghz.
And Vega in that "leak" was clocked at 300mhz.
Actually working 704ALU Vega iGPU will rape anything Intel silly.
>>62323113
There were no proper benchmarks yet. Just ES clock leaks and entries from dubious tests.
>>62323060
That 15w is already bullshit and I'll tell you why.
A single HT capable core working at 4Ghz is going to eat ~8.25 watts, and the rest of the platform still needs power to function - a few watts for the memory, some power for the IO hub, power for the native connectivity options, as well idle draw for everything else.
That they can guaranntee 1.8Ghz across all cores "within TDP" tells me enough.
More Intel lies for the Intel bribed to spit out.
>>62323229
At least they finally made 4core ULV chips.
And they are coming out before the rape.
>>62323256
Yeah, it's important the switch was made. They might have low sustained clock and up to 44 W power draw when going all out, but at least they are already here - it was getting ridiculous how lowend phones got octocores and laptops were 2C4T. Future successors will be better.
>>62323306
Yes, 10nm dieshrink will do the trick.
Now the grim part: AMD will start fabbing peanuts-sized mobile dies on 7nm LP as soon as possible. So Intel has like half a year worth of fab advantage.
Zen is extremely efficient on both power and area. And since 7nm LP will have comparable density to Intel's 10nm, well, next-gen APUs will be nice.
AMD should also consider 2.5D integration for APUs.
I used to be above lowend APUs, but as I got older and stopped doing shit like encoding, I might actually like the lowend version with 2C/4T of Zen, if it ever comes to market. Maybe they cancelled it.
>>62323435
Why would they cancel it? Its binned RR die with two cores disabled.
>>62321441
Dont give me this thin shit.
Where the thicc workstations with MXM and AMD HBM at?
>>62323443
Nah, there was supposed to be a dedicated chip for that. It is not economical to sell big chips into lowend. Llano to Carrizo were all 250. RR might go down to 170-180mm, but it will still be too big.
The codename of the embedded version was "Banded Kestrel" while RR is "Great Horned Owl".
>>62323479
Ah, Core-m competitor.
According to "leaks" RR goes all the way down to 4.5W. So I guess 2/4 part was canned.
See the slide. But it might have been in concept stage back then - it was not heard of since, and it is missing in 2018 roadmap.
>>62323479
In the same vein, AMD made a halved version of Carrizo for lowend to replace Kabini/Beema, it's called Stoney Ridge.
Or 2/4 chip is a part of RR family.
Who the fuck really knows?
>>62323497
Not as much Core M competitor, but cheap, lowend chip. Intended to fight Pentiums, Celerons. You can't get any decent margin against those by using 160-200mm2 big chips. Intel dualcores or Atoms are tiny.
>>62323533
Intel also has much denser process compared to AMD's 20nm.
>>62323229
Still better than what we currently have
>>62321441
How does a 2-1 laptop works with a screen that slim?
>>62321574
vega is ready the problem is they dont have enough bins for it
>>62321441
I WANT 2 DIMM SLOTS NOT SOME PUSSY ASS 1 SLOT THIN SPAGHETTI LAPTOP
I hope there's at least IGP that easily beats GTX750 Ti, or at least Vega that isn't expensive as fug
>>62327858
How can iGPU be expensive as fuck?
>>62323569
Problem is when the 14nm generation was rolling out from GF and TSMC everyone said 14nm would be just as expensive per transistor as 22nm and like the same or worse for Intel because of their troubles with production. I'm sure the situation has improved now but the original projection was pretty poor
>>62327975
I was talking about Vega in general. That is if AyyMD is definitely making Vega more than just 56/64/FE, and bringing a more affordable desktop range.
I'm conflicted for graphics card choice. H*cking miners, eating up all the stock!
>>62328013
Vega11 is later. It also uses HBM.
Most good Vega10 bins are simply going for Pro (these are going to be pretty popular) and MI25 (these are going to be popular once AMD finally fucking ports most of the frameworks to ROCm).
Consumers get ass.
where's the "once magic drivers come out, vega will btfo everything nvidia has" faggot when you need him? I want to laugh at him!
>>62328041
But "magic drivers" are yet to come out, newshit.
>>62328034
Aw frick. I just want a better card than my 750Ti
Guess I'll just get a used >3.5gb
>>62328055
Get a 1060 or 580, whatever is cheaper for you.
>>62328048
The "magic drivers" are yet to come out for the 10y/o R600 too.
>>62328441
And for nv30 too.
WHERE ARE MY APUS
REEEEEEEEE
>>62321828
CUDA is only for specific research.
OpenCL works fine. There's even Radeon version of CUDA.
>>62329523
Yes, ROCm is nice, BUT it needs more libs. And more frameworks.
>>62328441
All the previous Radeon cores improved more over time than contemporary GeForces. Even the ones that were launched years after GCN introduction (like Hawaii, Fiji). So that should apply even more for Vega, which brings bigger changes. It will probably look way different in 6 and 12 months from now.
>>62331732
At least previous GCN iterations launches with most shit working.
>>62328441
Core uarch features being disabled in launch drivers is not claiming there will be "magic drivers." There is nothing magical about expecting RTG to actually fucking implement the new uarch feature specifically targeted at finally alleviating GCN's front end bottleneck at high shader counts.
>>62329523
No one will use OpenCL when they can use CUDA.
>>62329538
Like he said immature as fuck, and the HIPS compiler is in worse shape than ROCm.
Bristol Ridge looks nice for what it is,
https://www.youtube.com/watch?v=3JTCzS_Uxso
Raven Ridge can't come sooner
>>62332799
What? HIP (which is NOT a compiler, it's a source translator) is part of ROCm and is capable of translating 95%+ of native CUDA source to C++, how is that in bad shape?
The reason researchers buy non-CUDA devices is because for the most part, unless you do DL/NN most BLAS/LAPACK routine libraries are out there already, like ViennaCL and clBLAST, and OpenCL is way more cross-platform than CUDA is, so it's more scalable between heterogeneous clusters. ROCm was not the first framework to offer libraries and routines to OpenCL, there's been plenty of OpenCL development on middleware; ROCm just offers particular focus on integrating drivers, kernel, and libraries to heterogeneous Radeon clusters targeted towards Deep Learning/Neural Nets (along with improvements to BLAS middleware). Only nVidia devices ship with CUDA, so OpenCL is better for real distributed computing, AMD also has pure compute and memory bandwidth advantages over nVidia cards.
will be nice once these come out. thinking of building one with this case: http://www.geeekstore.com/store/p69/GEEEK_A20_Mini_ITX_Case.html
will look awesome next to tv
>>62335202
>low profile gpu
What? What is this 2005?
>>62321441
1. Without CUDA it's useless.
2. The gentile CPU is inferior to the Jewish CPU by definition. Goyim can't into maths.
>>62335295
>goyim can't into math
>as Intel gets spitroasted by the chinks in Nvidia and pajeets in AMD
>>62335295
>iGPU
>CUDA
Has science gone too far?
>>62335295
Macs seem to be doing well without CUDA...
>>62321441
lmao I am not going to trust amd with any apu again.
I got fucked in dry arse, I have 5yo apu netbook and they are not going to update it to new xorg version because fuck up, its 5yo cpu lol!, go and buy new one.
They are even worse than fucking google unable to update nexus 4 with 5.1> because its too slow for new wallpaper and new bar
fuck them all
>>62336839
Even when you use open drivers?
>>62322256
There's more than 1 key uarch feature disabled. I'm keenly awaiting RR when the Vega drivers are decent.
Just imagine if there was a ryzen raven ridge thinkpad able to run win7
Just thinking about this makes me fully errect
A man can dream can't he
>>62337577
Lenovo will probably refresh their AMD chinkpads with Ryzen Mobile somewhere late Q1 2018/early Q2 2018.
>>62337370
I'm aware, but there's no point in implementing any of the rest of them if they don't get primitive shaders working.
>>62337325
Open drivers are shit, I am not dreaming about using yt with 720, but idle netbook without burning itself
>>62340450
Getting them to work is a silly complex task. Don't expect them in consumer drivers any time soon.