AMD Ryzen mobile

>>60680266
Let's hope this will mean ultraportable quad core laptops. Been holding off buying a laptop due to every quad core laptop are heavy and clunky.

>>60680350
Well, they're targeting the ultraportable gimp meme, which I personally ain't interested in, I'd rather see a 30W APU that can really stretch its legs

>>60680266
would jizz on / 10. seriously that looks gorgeous.

>>60680266
>the CPU portion (4 cores) should take a less than 80mm^2,

Source me nigga

Did they say it'd also come with SMT On?
Or was that the Vega core bits?

>>60680447
His dad works at Nintendo

File: ISSCC%208.jpg (239KB, 720x540px) Image search: [Google]

239KB, 720x540px

>>60680447
No source, just a well informed estimate.

Ryzen 4 core CCX are 44mm^2, but that's without any interconnects, buses or memory controllers.

The full Ryzen 7, which is two CCX, are 192 mm2, but those include 32 I/O lanes and GMI links, which will be mostly removed in mobile chips.
So a 80mm^2 for the full CPU portion of Ryzen mobile is a good estimate.

File: 4HUNFk0[1].png (610KB, 1329x719px) Image search: [Google]

610KB, 1329x719px

>>60680350
>Let's hope this will mean ultraportable quad core laptops
Looking good m8

>>60680490
4 core 8 thread.
Assumed 12 Vega CUs(768 shaders) using AMD's slides.

>>60680379
Good. I spend most of my time travelling for work. I'm looking for an ultraportable personal laptop to run a windows vm for my personal coding project so a quad core laptop is needed. Kind of sucks to bring the company laptop that weighs 3kg and is still a dual core. Rather leave it at home.

I'll make a estimate for 15W 4 core clocks.

2.3-2.5 all core boost, with single core near 3.8

For 15W that's quite impressive.

>>60680490
>Did they say it'd also come with SMT On?
No reason for it to be disabled.


This is actually really good to see, a single CCX, away from the limitations and latency bound problems of IF.
$5 says the mobile chips are going to have a higher IPC than desktop Ryzen.

File: Screenshot from 02-04-2017 21:00:52.png (39KB, 826x775px) Image search: [Google]

39KB, 826x775px

>>60680599
Threadripper will have higher IPC than Ryzen, much less a single CCX chip.

See pic

>>60680266
Nigga that's kawaii :3

File: mMBmP2A[1].png (17KB, 493x288px) Image search: [Google]

17KB, 493x288px

>>60680592
We don't need to estimate senpai

Source; https://forums.anandtech.com/threads/my-non-exacting-ryzen-power-consumption-numbers.2502431/

>>60680516
I've been hearing 1 CCX chips will have 6MB of L3 instead of 8MB, just a rumor though.
That should make the chip even smaller and more power efficient, more room for Vega

>>60680681
So I wasn't off at all, somewhere close to 2.5

That is.. assuming they stick to REAL 15W TDP, not like Intel mobile "TDP/SDP" that can turbo up to 30+ watts on 15W parts.

>>60680697
The single CCX bears 8MB L3, which is slap bang in the middle of the die, so it's unlikely they're going to be sacrificing it, unless I'm missing something here.

>>60680790
There were Raven Ridge leaks on sisoft(easy to forge though) with 4MB L3

>>60680833
May have been defective L3 and lasered down to 2C/4T? It's possible.
Though if the rumours of AMD hitting 80% yields on these CCXs are true, then a 2C/4T would be severely limited in supply.

>>60680871
It was a 4 core, and at 3.1GHz clocks.

Still, that sounds far from a ULV APU, so it's likely forged or something far in the future, or maybe a AM4 APU

>>60680891
It would certainly be odd as to why they would be lacking in L3 for any reason other than defect, would any enlightened individual care to educate me as to why you would sacrifice a significant amount of L3 for what is seemingly no benefit?

http://www.cendirect.com/main_en/tech-specs-Z5M90AA%23ABA-JO9196EM.html
http://www.cendirect.com/main_en/tech-specs-Z5M89AA%23ABA-JO9194EM.html
http://www.cendirect.com/main_en/tech-specs-Z5M88AA%23ABA-JO9182EM.html
http://www.cendirect.com/main_en/tech-specs-Z5M87AA%23ABA-JO9181EM.html

They're not even waiting for APUs to put into laptops.
>1800X in a laptop
REV UP THOSE FAN ENGINES

>>60680912
Likely a early ES, as to why they'd cut out L3 if it's not a defect, the market for these chips aren't as big in L3 requirements as the Ryzens are, so maybe there's a perf/watt benefit to disabling it.

Alternatively, it might be another way for product segmentation if yields are indeed true, with such high yields you're gonna have to artificially disable chips to offer lower end parts instead of waiting to stockpile them directly from the wafer.

>>60680935
>>60680935
>1800X in a laptop
Is this like those stupid "ACER Predator" things that have no reason to exist other than for the company to have something to brag about?

I know there are some more sane options on that list, but come on now.

>>60680977
But with artificially disabling cores, wouldn't it be possible for end users to simply renable these cores? Like back with Phenom and pencil modding?

Assuming yields are true, and segmentation is required, surely they could laser the L3 right down the middle and produce 2, 2C modules?

>>60681000
If it's downclocked it's still more efficient than a 1700 at the same clocks.

Anyhow, OEMs have stuck 120W Xeons into laptops before, professionals don't give a fuck, they want a mobile workstation.

File: Screenshot from 31-05-2017 23:01:01.png (54KB, 812x316px) Image search: [Google]

54KB, 812x316px

>>60681016
You can only enable them if they're software locked, if they're fused off there's no way to get them back.

And yes, there's 2 core parts, most probably for tablets/2-in-1 and IoT

>>60681055
2.1/2.5 seem a little high for tablet use, I'd be more willing to wager low end laptops, maybe even "ultra"books, whatever that meme means now.

>>60681055
Fanless, I guess.

Those frequencies are fucking impressive though

AMD based laptops when? Think big names like Thinkpads, Dell precision/latitudes/hp probooks will adopt the new AMD stuff?

>>60681160
If you're into gaymer shit ASUS just released a Ryzen 1700 and RX580(at 65W) into their ROG laptop

It's gonna take a while for those though, Intel's dick is too strong

>>60681160
Dell, HP, ASUS would certainly be key customers to AMD. I guess once we get closer to Ryzen's mobile release, we may see something strong come out from one of them.

>>60681180
>RX580(at 65W)
>65W
How in the fuck

File: AMD%20Embedded%20GPUs%20Launch%20Presentation-06_575px.png (907KB, 955x537px) Image search: [Google]

907KB, 955x537px

>>60681234
Cherrypicked voltage bins + lower clockspeed.


AMD has been doing this for a while, giving gaymers scrap chips while industry gets the nice ones, of course since gamers don't respect AMD anyway, AMD has no reason to respect gaymer manchildren that still bought Tesla/Fermi over far superior RV700/Evergreen cards, it's a vicious cycle.

So I'm not really surprised AMD gives no little to no amount of shit for gamers.

http://www.anandtech.com/show/10710/amd-announces-embedded-radeon-e9260-e9550

>>60681299
Nvidia also saves up nicer bins for enterprise, but they're nowhere close to 100% perf/watt difference like AMD shitbins vs nice stuff, more closer to 20%

File: 006919_wp_20170531_11_05_21_pro.jpg (276KB, 1919x1080px) Image search: [Google]

276KB, 1919x1080px

>>60680266

>>60681234
Better bins with lower clocks. AMD never gives best bins to gaymen, saving them for embedded/laptops, mostly because
>AMD has no reason to respect gaymer manchildren that still bought Tesla/Fermi over far superior RV700/Evergreen cards

>>60680266
THINNN

>>60681512
Heh, you like little baby

>>60681512
ITS SO FUCKING BIG
BIG
AMD
SOCKET

>>60681701
Over 50% of the pins of the socket are unused, and shares the same layout as the 32 core EPYC chips.

It's either:

Forward compatibility with zen+/zen2/zen3 over the next few years, more than enough pins for even 7nm
24 core EPYC consumer version

>>60681771
>24 core EPYC consumer version
That's too much cores.

>>60681851
Intel shat out their 18 core part(which won't release for a few months, but hey) it's only fitting AMD one ups them.

>>60681871
>just add another die in package lmao
Intel are soooooooooooooooooooooooooooooooooooo fucked.

>>60681883

"Hey Lisa I just got word from our spy that someone over at Intel said that they MIGHT look at releasing their 28 core for HEDt an-". . . "Say no more Jeff, 4 DIES, ONE PCB, SINGLE SOCKET HOLOCAUST!"

>>60681851
>t. jewtel corelet

File: AMD-EPYC-CPU-740x416.jpg (26KB, 740x416px) Image search: [Google]

26KB, 740x416px

>>60682143
That's exactly what Epyc is, though.

Can I build a tiny but powerful computer with this?

O-kay, AMD is raping the ever living shit out of Intel. Now Pajeet needs to have good fucking launch drivers for Vega instead of usual abortion.

>>60682183
Will AMD sell this to you and not OEMs?

Very fucking unlikely

>>60682188
AM4 APUs are Q1 2018, anon.

>>60682195
And these mobile chips are Q3

Notice the two quarter difference?

>>60682188

They will. Many AM4 mobos have display outputs for the exact reason that they are compatible with APUs. You can get older gen APUs that were made for the AM4 socket, don't see why they wouldn't sell APUs standalone to make money.

>>60682220
Yes, AMD wants Q4 laptop sales.

>>60682233
They will, but they come much later than mobile BGA parts

>>60680266
The thought of decent battery life with RX 560 tier performance and a 4c/8t CPU, especially if it's cheap, is fucking drool worthy.

>>60682379
>RX 560 tier
~550. It's 768 shaders but at lower clocks and DDR4 is a big bottleneck.

>>60682401
DDR4 is very unlikely to be a bottleneck for such a small GPU, doubly so with bigger L2 caches, tiled binning and HBCC which are all designed to reduce bandwidth requirements.

>>60680266
As interesting as this APU seems, no one will really care for it because it will
>Cost more than the Ryzen 3
>Perform worse than a Ryzen 5 1500X
>Have an iGPU that's barely better than the older gen Xbone

>>60685575
It's a laptop CPU. Still better than Intel integrated iGPU garbage.

>>60680521

>4c/8t Zen core + Vega GPU for a Zen APU
>the laptop is THAT FUCKING THIN

>Apple is looking to dump Intel
>rumor has it that Apple is eyeing the Zen APU for its next-gen macbooks

Sweet lord. I hope this pans out to be true, the stock price for AMD would shoot through the roof. Especially if the Zen APU manages to deliver some god tier performance in CPU and GPU together.

>>60686626
it will happen. there is no doubt about this. no other company has a product like this, apple can not let this go.

>>60680935
I mean the 1800X uses similar power to the i7-7700HQ, so why the fuck not?
And since they're soldered, easier to get the heat out. I bet it feels cooler on your lap than a 7700HQ laptop.

>>60681021
This. I know many engineers that would buy Xeon or Gaming laptops even though they don't game because they needed the faster CPU and the CUDA.

>>60681234
Polaris is very efficient at lower clocks.

>>60681771
40% of pins on AM4 are unused as well with the 1800X.
I think we're going to be getting 12core on AM4 next year, and 24core Threadripper.
A good deal of pins are unused even with the 32core Epyc that's essentially the same socket. It looks like 48core Starship is going to be on the same socket.

>>60682165
Yes, and there is technically no reason why Epyc can't run on X399 from what I can see. It's the same socket.
Maybe they would need more power connectors, but I don't think so. They should be able to do 4 pins each die instead of 8 pins and just clock them ~20% lower.

>>60682188
They will. But it might take 3-6 months. :/ Lots of macbook and surface pro orders to fill first.

>>60685575
4 cores per CCX though. It could potentially have better CPU performance than the Ryzen 4 cores. And it might have dual graphics support.

>>60686626
Yeah. I bought my 600 shares back after the dip. Confident.
Imagine how cheap that APU is going to cost to make with the absurd yields GloFo is getting. Probably over 90% profit margin and they are going to completely dominant the laptop market with them.

>>60680542
I'm pretty certain it's 4-11 CUs max.
They might be manufacturing 12 but disabling 1 for yields

>>60686734

Well, a 2xCCX Zen die (1700/1800) chips have an 80% yield. A single CCX die yields must be nearing the lower to mid 90s. Which is unbelievably good.

The only issue is going to be Vega that will make up the APU portion. Factor in uncore and memory controller, and still the die is probably ~80-85% the equivalence of a 1700/1800 die size. Which probably means that yields for a Zen APU will again be in the 80s, likely in the upper 80s.

So if a 14nm 300mm^2 wafer has a yield of say 600 dies and we assume a perfect 85 middle of the road yield per wafer, you're looking at 510 dies at the maximum and (if you assume a solid low of 80) 480 dies per wafer at minimum.

Considering that not only Apple but also other AIOs would want this, the demand for this is going to be insane. RX Vega might do poorly, but Zen APU w/ VEGA core will do excellently. It probably costs fuck all for AMD to make these Zen APUs and with Apple's market presence and other AIOs factored, they're going to undercut Intel so god damn massively.

All that said though, cautious optimism with a positive hope.

>>60686836
Right. And at probably $4000 per wafer cost to AMD, that's $7.80 per die. Plus assembly and shipping, two bucks more? If they can sell them to OEMs for $100 for the top 2c/4t 11CUs SKU, that's 90% profit margin.
It's likely something that'd run Doom at 45-60fps, 1080p, on medium-high settings.

>>60686924

Doom is Vulkan, which benefits massively from Async Compute, which Vega APU core will deliver in spades. So yeah, I can see that happening.

>>60680266
Part of that left over space is PCIe, MC, and i/o.

But anyway, 210mm^2 puts it at 60% larger than a G4560 for something that has 2x the CPU performance and probably 3x the GPU performance.
That's pretty good, but AMD should undoubtably be looking to make one with less transistors.

>>60686734
>And it might have dual graphics support
We all remember how well the last one turned out. Especially from a value proposition, where a $125 APU and a $100 R5 250 got worse performance than a $65 Athlon and a $150 R7 260X

>>60687105

But that's because the Bulldozer core is pure garbage. Zen is a different monster and can deliver on the performance spec.

>>60687150
No, I was specifically talking about the GPUs and Dual Graphics. DG made no sense when you could get the same CPU with the iGPU disabled for half the price of the flagship APU, which meant that you could pair it with a much more expensive GPU and still save money compared to a DG pair.
Even if RR comes in at around $125 for the flagship APU (which I can guarantee it won't at this point), there are no sub-$100 GPU that can be DG'd with Vega. And if there were going to be super cheap Vega dGPUs on the horizon, then it won't happen in another two years, when RR would have been succeeded by another APU generation.

>>60687193
It made sense when someone only had a $300 budget for a PC that can play some games, but then could upgrade later.

The problem wasn't what you said, but what GPUs you were limited to dual graphics with.
If the top end APU can dual graphics with any Vega GPU, like an RX680(tentative), that is what I'm hoping for.
Being limited to dual graphics with a shitty R7 250 is why it sucked before.
One of the big features in Vega is that it's not locked to 4 clusters in its compute structure, so if dual graphics comes back it could/should be much better.