Power9 IBM, Intel, AMD, finished, etcetc

File: Screenshot from 23-03-2017 14:25:42.png (767KB, 1727x969px) Image search: [Google]

767KB, 1727x969px

Some more neat stuff, huge logic and size difference between a SMT4 and SMT8 core.

File: Screenshot from 23-03-2017 14:27:07.png (279KB, 1715x966px) Image search: [Google]

279KB, 1715x966px

Much shorter pipeline, should be a IPC monster.

File: Screenshot from 23-03-2017 14:28:26.png (232KB, 1725x937px) Image search: [Google]

232KB, 1725x937px

Frontend numbers.
Wide as hell core.

File: Screenshot from 23-03-2017 14:30:52.png (197KB, 1740x964px) Image search: [Google]

197KB, 1740x964px

>FP128!
>making Javascript/Python unslow
>botnet

Insane on-chip bandwidth, and 17 metal layer design.
For reference AMD's zen is 11 metal layers while Kabylake is 12.

File: Screenshot from 23-03-2017 14:32:41.png (534KB, 1728x959px) Image search: [Google]

534KB, 1728x959px

>>59539987

Forgot pic.

File: Screenshot from 23-03-2017 14:34:17.png (321KB, 1725x983px) Image search: [Google]

321KB, 1725x983px

PCIe 4.0!

FOUR DOT ZERO

File: Screenshot from 23-03-2017 14:35:35.png (84KB, 1726x906px) Image search: [Google]

84KB, 1726x906px

Finally, internal perf

There's a bunch more stuff in the PDF, if you're interested I suggest a read, it's good stuff.

No one?

File: IBM_Lotus_logo.png (3KB, 337x82px) Image search: [Google]

3KB, 337x82px

>>59539919
i could never understood ibm pricing strategy. POWER was many times better then intel, why can't they make ARM model?
I wish i had reasonably priced power, risc cpu's

Will Apple switch back to PowerPC now?

Intel macs are boring.

>>59540021
This IPC or wat

>17 metal layers
Ooh baby.
eDRAM slice looks like their IP applied to bulk. I think this is a 14nm LPP chip. Frequency scaling apparently tops out at 4ghz too. More than a little interesting.

>>59539919
So what's the use case for this?

File: 1486195380210.jpg (51KB, 666x629px) Image search: [Google]

51KB, 666x629px

>8-way SMT

I am not that knowledgable.
Could someone please explain to me what kind of product IBM is producing and why it is hyped?

will this beat the i7 7700k in gaming performance?

I've been keeping tabs on this one. Mainstream SKU line when?

>>59540021
It looks like Intel is going to fucked from bellow by AMD's cheaper and more power efficient Zen core, and fucked from above by IBM's monstrous Power9, and tucked internally by Krzanich. The next couple of years won't be fun for Intel's share holders.

POWER9 sucks for games lol IBM is bankrupt and finished LOL I can't play CS:GO on this omg how useless.

IBM fan boys BTFO, i7 7700K the only processor you need.

>>59540640
Neural-net processors.

Next gen CPU for data crunching numbers.

>>59540791
IBM BTFO! How will they ever recover?

>>59540640
IBM currently only works in the business sector making databases and mainframe type servers. Tons of reliable, redundant hardware to ensure unquestionable accuracy and reliability for stuff like credit card transactions. They've essentially exited the consumer market with their last real attempt there being Cell.

The above seems like it would be a more capable general purpose design, but its strengths would really be in servers. The consumer desktop market is harder than ever to try to make a stake in. Powermacs were essentially the last holdout, and the final generation of mac pros were pretty dated even on release and cell was a complete failure. I really doubt IBM is all that interested in making a comeback there.

>>59540591
POWER 8 had it too

>>59539999
>17 layers

This thing will cost more than those top secret gubbiment CPUs and radiation hardened ones.

>>59539941
>shorten fetch to compute by 5 cycles

holy shit

>>59540981
From what I remember IBM dropped the consumer market because they think it wasn't worth it fucking around with retarded consumers.

Keep it to professionals, seems to have worked for them.

>>59541102
>From what I remember IBM dropped the consumer market because they think it wasn't worth it fucking around with retarded consumers.

IBM sold their server business to Lenovo for fuck's sake, they think that most businesses are retarded.

They literally only do business with Fortune 100 and governments now. That's where most of the money is anyway.

>>59541166
>IBM sold their x86 server business to Lenovo
FTFY

>>59541193
thanks, forgot

IBM is basically playing chicken with HP for big iron right now, they want the Superdome Integrity market.

Everybody's already running linux so running a new processor isn't that big of a deal, these workloads can translate really easily.

>>59539919
I suspect it will still be expensive and power consumption will be too big, with will greatly limit its chance against Skylake server and Zen server CPUs (Naples).

I haven't touched any Power machine since Wii.

>>59542402
Except POWER doesn't compete with x86.
It's in the Big Iron segment and its only competition is Itanium (basically dead) and SPARC (soon to be dead).

>>59542898
It wants to, actually, to grab more marketshare.

In its big iron niche, I agree, but the OpenPower and scaleout P9 is actually meant to go into mainstream severs, Xeon E5-2600. They already are trying to push Power8 into this market, which kind fails though because bad Linux ecosystme support and high power draw of the beefy platforms (Centaur memory chips etc).

>>59542898
Power9 seems scalable so if IBM's gonna enter the regular datacenter/server market Intel and AMD are this is its chance.
It has Nvidia's support as well

>>59540243
can i play crysis with it or what? give some input for retards pls

>>59540591
>>59541001
what kind of workloads can deal with that gracefully, i.e., not having the threads murder each other over scraps of L2 and L1D?

Tyan POWER 8 servers start at $3000. I wish them luck, but it's out of reach for most people.

>>59545335

Chump change for their market.

File: 1481443067547.webm (413KB, 640x360px) Image search: [Google]

413KB, 640x360px

Oh great! The next gen CPU banks can use to enslave their customers!

Money architectures powered by obscurity.
Or the "illuminati" as smart people would say.

>>59545346
True, but they wont make a sizable dent in the market because of cost. The main reason ARM flourishes is because they're cheap. It certainly isn't because they're better than Intel.

>>59545265
Databases larger than you've ever seen.

>>59545538
I'm pretty sure those just need as much I/O and as you can give them at low power, they shouldn't need either memory extreme capacity or bandwidth

>>59545428

They don't give a shit about market. The marketd they compete in have zero competition. Power architrcture servers are serious number crunching machines that operate in picosecond intervals.

These are what are used in high frequency trading machines, powering the NYSE, NASDAQ, and scientific computing. And the industry thanks them for charging so little.

>>59545382
I keked

>>59545642
Good for you, but it's the truth that banks utilized POWER racks en masse when the masses didn't even know what "Apple" is. Because let's face it: If it weren't for Apple, there wouldn't be a Power arch in the first place.

Think of it as you want, I'm just telling you what I know.

>>59545595
They do have competition. Intel takes a sizable chunk out of their market every year.

>>59545690
I was talking about the cactus in front of the bubble blower

File: 1481650843887.jpg (177KB, 730x457px) Image search: [Google]

177KB, 730x457px

>>59545755
Yeah, whatever floats your boat, m8

>>59545750
No dude this isn't fucking Xeon tier shit. This is well above that.

>>59545798
god damn i want to go swimming here right now and drink beers floating in the water. FUCK

>>59540791
newfag

File: 1401038213896.jpg (62KB, 450x557px) Image search: [Google]

62KB, 450x557px

>there will be non-x86 Unix workstations again in your lifetime

>>59545850

>tfw my idea of a perfect holiday is floating around in a inflatable donut in water like this knocking back beers

>>59546140
Fuck, I'd buy one even if it was ARM. I'm sick to death with x86.

>>59540655

Run CS at 800x600@144? :^)

>>59545750
Yeah, but those are mostly systems that only run on exotic architectures for legacy reasons (x86 wasn't that good before) and can be readily moved to cheaper x86 boxes.
But POWER keeps going for the same reason mainframes still exist - they have a well-defined niche where nothing else can be used. POWER's niche just so happens to be highly reliable systems like telecoms and factories use.

>>59540278
they made plenty of money raping their customers who all moved on basically

if they marketed an enthusiasts board for like $1000 including the cpu they'd sell them like hotcakes

>>59547300
>they have a well-defined niche where nothing else can be used

more like IBM has completely locked them in to jew them indefinitely

>>59545428
These aren't consumer chips for cheap-ass consumer systems, they're for high-end database boxes and other critical systems run by large companies where $250,000 is a tinke in an ocean of piss. Marketshare doesn't mean shit unless you're a megacorp pushing cheap trash in volume to make up for the fucking terrible profit margins.
>>59545750
Intel mostly cannibalized the remnants of Sun, really. IBM iron was never something you'd fill a datacenter with in most cases, whatever they did lose to Intel they already lost decades ago anyway.
>>59546140
For good reason, after about 1998 they were overpriced archaic trash that did little a good NT workstation couldn't do for a quarter of the price with far more versatility.
>>59547195
So is ARM, please stop wishing for the market to become even shittier than it already is for the sake of some contrarian fantasy.

>>59547593
Sounds like Intel

File: 1qaiafbo1.jpg (229KB, 540x720px) Image search: [Google]

229KB, 540x720px

>>59547195
Buy a Raspberry Pi :^)

File: 1c63370dc3580207c1d3360eec793bb44c888d9f8c3d81be25d94c347cb8f62d.jpg (16KB, 439x293px) Image search: [Google]

16KB, 439x293px

>>59540340
they wont, because power is very power consuming.... and you know how much they like macbooks.. thats only one reason

>>59541102
this

>>59547856
pi works great

it's a nice little machine

>>59540278
PowerPC does have this model to an extent, and you nor anyone else ITT knows that because they're useless shit for anything outside of embedded systems, just like post-2000 pure RISC in general
>>59547568
DEC did that in the '90s with the Alpha
it was shit and nobody bought it even though it could run Windows and x86 applications

likewise no "enthusiast" is going to pay out the ass today for something that does absolutely nothing in a home environment you can't already do with a PC running Linux, a low-end system p from a few generations ago, or a PPC development board

RISCtards need to learn from history and stop projecting their hipsterism on the rest of the market, fucking nobody but a few contrarians who've barely written a line of assembly in their life would buy into that shit

don't get me wrong, it's still cool to be into this shit and exciting to explore different ways of doing things, but the bitter circlejerk on this board about inconsequential shit deeply abstracted away from them in 99.9% of what they do gets really old and makes it hard to actually discuss these things

Why dont Nvidia get the x86 license from IBM

Apple should move back to powerpc intel a shit

>>59548206
they're going to go arm sooner or later

wait and see

>>59548280
Weren't there some rumors that they are going to use custom chips like they do for the iPhone?

>>59539919
IBM and Apple are the only tech companies I respect (maybe through in AMD in there too). hoping for a PowerPC revival in Macs but that'll probably be another 5+ years off

>>59549345
Modern PPC/POWER is shit for consumer computing, fuck that.

I want another dual CPU power pc workstation

>>59549345
>hoping for a PowerPC revival in Macs

this will never happen

more likely macos will go over to armshit

>>59549110
yeah i'm sure they will, they are at least a generation ahead of any other arm makers

Looks neat, OP. Sounds like the hype is actually pretty real.

>>59540446
It says @ constant frequency, so what do you think?

>>59551412
It didnt tell if it using STM/Hyperthreading or not

>>59539964
the symmetric approach looks interesting, would be good to know if they really use a unified register file as this is insinuating (i.e. 'everything can be a vector'). branch prediction looks underpowered here, hope they can catch up with Intel. Perhaps they finally implement some TAGE variant?

>>59539964
also, what's the lsu latency and bandwidth to the vsu? This is where Intel really shines with 64 bytes/cycle.

Holy fuck this thing is a monster
AMD single handedly revived the server processor industry
>>59542898
Did Fujitsu really drop their SPARC efforts? I read that their new SX is going to be ARM, but I haven't seen they actually killing off their SPARC chips

File: roman.jpg (90KB, 400x358px) Image search: [Google]

90KB, 400x358px

>>59545988
you just outed yourself my dude

>>59540021
How many cores per socket for each? I feel like it's comparing 16 cores vs 24 cores.

>>59539919
how the fuck do i get a power9? i want to program in that sweet ass power assembly

>>59552281
Buy one from a reseller online in 10 years

>>59540340

Power hasn't been a consumer architecture for a decade now, it's all data centre and supercomputers now.

>>59552655
Nintendo rehashed old PPC chips until 2013

>>59552256
>AMD single handedly revived the server processor industry

Lol wat?
AMD are literally nobodies in the server market.
There are more ARM cores in servers than AMD right now.

>>59552670
>I have no idea about the impact Naples will have
wew lad

>>59552678

>>"speculation"

Nobody has any idea, it's currently vapourware.
IBM certainly don't give a shit, AMD doesn't compete with them.

>>59552678

POWER9 has 48 channels of PCI-E Gen4 per chip.

Naples has 8 channels of PCI-E Gen3.

>>59552678
>>59552670
how the fuck can he? he has never tested the chip himself you fucking shill.

>>59552692
>Nobody has any idea, it's currently vapourware.
It isn't, Ryzen is already out and proved that the basic building blocks are sound
>>59552710
Naples has already been showcased in public back in December at the Radeon Instict event
>>59552709
Yeah, it doesn't directly compete with most of IBM's market
However the market hasn't moved this much since 2013

File: Intel_Itanium-1-.png (110KB, 308x342px) Image search: [Google]

110KB, 308x342px

>>59552731
>but marketers said it would be good!
man I can't remember the last time I heard that one, anyone want to remind me?

>>59552731
>Ryzen is already out and proved that the basic building blocks are sound

MOAR CORES doesn't necessarily make a good server chip.

>>59552750
But 128 PCIe lanes, 16 DIMMs per socket, 8 channel memory and low power consumption sure as hell do.

>>59552744
Ryzen has already been tested in server workloads, at which it excels
>>59552750
MOAR COARZ coupled with great single thread performance and ridiculous efficiency does
Unless their fabric has some ridiculous issues AMD is set to dominate the server market within a few years, even if it does 8c/16t chips are going to take on most of the single socket market

>>59552762

No NvLink, no high bandwidth connectivity, only PCI-E gen3.

>>59552709
>Naples has 8 channels of PCI-E Gen3.
That's Ryzen, Naples has 128 PCIe Gen3 channel per chip

>>59552782
Infiniband is supported, nobody cares about GPUs in a full x86 environment, there's a reason Nvidia only has around 150 million of Tesla revenue per quarter, that's a small number of cards.

>>59552782
128 lanes of PCIe Gen3 are a lot of bandwidth senpai, without getting married to Nvidia's NVLink and still being able to use bandwidth hungry accelerators

>>59552782
Seriously you nvidiots need to look at reality, not much can make use of a fp64 card, this is pretty evident by Intel having 15 times the revenue of Nvidia.

Both Nvidia and AMD are literally insects for Intel

>>59552773
>Ryzen has already been tested in server workloads, at which it excels
you mean benched for basic bitch cases under ideal conditions that don't reflect the real world? itanic did that too, and these are shitty consumer chips suitable for low-end servers at best, we won't really know how good it actually is until the real shit actually starts shipping and making it into production setups, period

but who gives a shit anyway, why the fuck are we discussing gayMD vs. jewtel shilling in this thread? the original starting point in the argument is still true, AMD hardly "redefined" the market single handedly, this shit was in the works for years

>>59552861
and someone on /g/ is suggesting that seismic analysis doesnt reflect real world workloads


time to disconnect from the internet

>>59552875
That shits sounds like something a uni would do, gaming is a proper real world workload for 99% of people

File: x5000-system.jpg (57KB, 470x350px) Image search: [Google]

57KB, 470x350px

>>59552655
>Power hasn't been a consumer architecture for a decade now
um try again sweetie :)

>>59552861
IBM had the foresight of knowing that Zen was actually going to bring competition to the server market, unlike Intel
If AMD hadn't went all in IBM wouldn't had needed to continue developing POWER, instead they could had redirected their efforts towards their quantum machines or their Synapse processors and keeping POWER8 around since it would had been good enough (just like Krznich though Broadwell-E was going to be good forever)
But yeah, keep believing Zen is a lie and has nothing to offer
>>59552887
The AmigaOne uses a chip designed in 2007 senpai

>>59552886
Nobody playing games will do it on a 32 core fucking Opteron, likewise no normal person will buy a $5000 CPU, jesus christ people what the fuck are you doing on this board

>>59552901
>Nobody playing games will do it on a 32 core fucking Opteron
People do this since Bulldozer Opteron's were decommissioned, just like people do with Nehalem Xeon's

>>59552886
if by uni you mean
a)every eng that wants to built a house
b)every goverment
c)every oil company
d)every company that deals with static analysis in general
e)every scientist that deals with seismic prediction
f)everyone that deals with structural analysis in general..

the list is pretty big..

>>59552910
Secondhand server CPUs are good for a few things but not really for gaming, that prefers clockspeed over a bunch of low clocked cores.

>>59552898
>AmigaOne uses a chip designed in 2007
The newly released AmigaOne X5000 uses a chip designed in 2010, actually

>>59552875
>and someone on /g/ is suggesting that seismic analysis doesnt reflect real world workloads
not necessarily, no, it depends on their methodology. somehow I find it doubtful that they were feeding it a real "seismic analysis" workload to a shitty single-socket demo box and rather doing it the way most other popular benchmarks do and giving it some tiny datasets to play with for a short amount of time

that's pretty much just an empty appeal to reddit-cool science, the context of the calculations isn't the end all be all no matter how big the words are

>>59552898
>IBM had the foresight of knowing that Zen was actually going to bring competition to the server market, unlike Intel
they don't give a rat's ass about what's going on in commodity systems and never will, whatever share they had in those markets were already lost to the likes of Intel, Sun and company decades ago anyway

they simply did what they have always done: build a ridiculously fast chip for markets where cost is no object, flagship POWER systems have always been like this

>But yeah, keep believing Zen is a lie and has nothing to offer
this isn't what I'm saying at all, I'm saying that you've got a big corporation's cock so deeply lodged in your throat that you can't see reality in a way that doesn't lob huge amounts of praise on them, it's fucking retarded and I wish /g/ would stop doing it

>>59552971
pretty sure their naples video covered this up quite a lot..

it wasnt for sure a big ass workload but it was an eye opener
they had 3 demos 2 with 1 billion grids and one with 4 billion on which the intel crashed and burn

>>59552940
workstation/server chips have a pretty wide variety of configurations and aren't all big 8+ core monsters with low clocks, most high-end consumer chips usually have similar workstation/server counterparts with a similar or slightly lower frequency and a few extra features enabled

>>59553008
These were mostly memory capacity/bandwidth workloads, which the higher DIMM count and twice the memory channels insanely helps.
It is somewhat cherrypicking, but then again these memory bandwidth/capacity workloads are far from rare in this sector.

>>59552971
>they don't give a rat's ass about what's going on in commodity systems and never will, whatever share they had in those markets were already lost to the likes of Intel, Sun and company decades ago anyway
Except it doesn't, HPC is a major field where POWER gets major competition from x86_64 systems, nowadays just from Intel
But back in 2012 they got major competition from both vendors, when Piledriver got quite some design wins for major supercomputers
Naples is, unsurprisingly, also attacking the HPC market which is pretty obvious from their Radeon Instinct event

Why do you people keep going to the HPC market? HPC doesn't make even 15% of the server market, datacenter, SMB, cloud, webserver are far bigger markets.

HPC is nice for profit margins, but it is by no means a large share of the server market

>>59548100
>it was shit and nobody bought
so what, this means that it can't happen in the future?

you probably thought trump wouldn't win because businessmen can't be president, too

>>59553011
actually as i said seismic analysis is one of the most taxing workloads on a system regardless of what it needs or not

they pretty much showcased that their 2p commercial server compared to what intel offers is pretty much a better deal mainly because of MOAR cores and MOAR memory out of the box

>>59553026
Because it's the most interesting, nobody really gives a shit about some SMB middle manager who thinks he needs a 16 core Xeon his shitty powerpoint presentation.

>>59553008
>it wasnt for sure a big ass workload but it was an eye opener
for what though, peak performance under ideal and unrealistic conditions?
naples is surely not a non-event, it's pretty exciting, all I'm saying is that you don't take marketing as gospel before the thing actually starts seeing use in the real world where things aren't nice, clean and controlled and the work done is much different from little predominantly cache-resident benchmarks

>>59553014
that's indeed true, but it's not POWER's only market, arguably not even their most significant one, they've never really been dominant in it either

>>59553049
as i said here >>59552921

if you think anything of this magnitude is unrealistic you should probably go and RMA yourself

>>59539919

Too bad it's only avalible in expensive as fuck server systems and has just about no desktop software support.

This doesn't affect me. I dont own a server park and I dont have a team of programmers to port the software I need.

But that's good for the server market I guess. afaik power cpus are rarely used outside of niche applications though, while ARM and x86 dominates.

>>59553053
>if you think anything of this magnitude is unrealistic
what magnitude? the magnitude of big cool words the marketers are using to get you to suck them off and hail them as your lord and savior? the label doesn't have some inherently magic quality to it.

>>59552790
>there's a reason Nvidia only has around 150 million of Tesla revenue per quarter, that's a small number of cards.

They own the HPC market. Most supercomputers under construction right now have a GPU component, and overwhelmingly that's nVidia.

>>59553070
lets get something clear here

seismic analysis is just a sub of structural analysis that involves everything i actually wrote and some more

so no there is nothing unrealistic about that whoever thinks that because they showcased one workload over a dozen of them on a 30min presentation needs to get his brain checked

>>59553080
Their revenue directly contradicts that, 150 million USD revenue on Teslas is around 30k high end fp64 GPUs sold per quarter.

That number is literally nothing.

>>59552887

10 year old, outdated POWER architecture, that's being produced by a company bought by Apple nearly a decade ago, and is only still under production because of outstanding government contracts.

>>59552948
>The newly released AmigaOne X5000 uses a chip designed in 2010, actually

Which is an ARM processor, and seven years is a fucking eternity ago for the ARM architecture.

>>59553086
you don't seem to understand what I'm saying
you're just appealing to the label, saying that because it's "seismic analysis" it's "real" work reflective of exactly what you'd see in a production environment
what I'm saying is that just because they're using calculations common to this use case doesn't really mean dick, shitty benchmarks like linpack and livermore loops do this too and they're a terrible gauge of real world performance
the fact that some tuned benchmark running on a tuned system by marketing is performing well on some small dataset problem means fuck all in the real world where datasets are big, shitty programmers rule and systems are rarely meticulously tuned and maintained, because they're for doing something productive instead of generating pretty bar graphs for fanboys to jerk it to and feed their narrative of corporate cock sucking and crediting them for everything good in the world

>>59553097
you're thinking of the X1000, see >>59552948

>>59553106
no it's not, what made you think that? it's PowerPC e5500

>>59540340
That would be awesome, Apple's hardware has been standard PC crap for over 10 years now.

>>59539964
These sort of look like bulldozer cores, eh?

Where it was really small cores that could combine together.

>>59540836
You mean a learning computer?

>>59552861
Wowie. You're actually as delusional as you are retarded. Comparing Naples to Itanium (not Italic, you clueless fucktard).

Holy shit.

>>59539919
It's too expensive and power-hungry to finish anything or anyone. Cheap shit always wins.

>>59549345
>PowerPC revival in Macs
POWER and PowerPC are two different lines, PowerPC is basically dead.

File: 1489958525427.jpg (266KB, 844x986px) Image search: [Google]

266KB, 844x986px

>>59553944
hey Raj, if you actually bothered to read beyond a first grade level you'd realize that's not what I'm doing at all, I'm comparing your mindless cock sucking of marketing to the similar cock sucking the Itanic received (albeit at a far greater, industry-wide level) in the '90s while it was trashing benchmarks under optimal conditions, even though when it actually made it into the real world it was average at best and fucking pathetic on average

>(not Italic, you clueless fucktard)
>"Italic"
...wait, who are you calling clueless again?

>>59539941
Here's hoping their compiler can deal with that. Instruction scheduling for maximum IPC was the downfall of things like the Itanium

>>59540655
since there are no games for POWER arch, we will never know.

>>59549393
>more likely macos will go over to armshit
What makes ARM so bad?

>>59556920
Well, there are, granted they are console games. But I really don't think there are all that much platform-specific stuff left in games nowadays, so most games would just have to be recompiled if such a machine were to come out.

>>59539919
I wonder how fast is monero mining on this