Everyone go home, the Skylake Xeons are low clocked nuclear

>>59605125
>Moore's Law is dead, what a disappointment!
Yeah everyone knows that already, anon.

I don't particularly know why you give a crap about the TDP of Xeons you're not going to buy anyway.

File: Screen Shot 2017-03-27 at 18.31.02-fullpage.png (1MB, 1110x5110px) Image search: [Google]

1MB, 1110x5110px

Ayyy Intel housefires

Install Gentoo

File: lepepes_41.jpg (32KB, 499x499px) Image search: [Google]

32KB, 499x499px

>>59605585
Oh how the tables have turned

File: 1426355327521.png (75KB, 198x198px) Image search: [Google]

75KB, 198x198px

>>59605585
Turbo keks


>"Based on AMD's technology briefings, we know it has more granular control over clock rate. And it's notable that the 95W Ryzen 7 1700X we're reviewing today uses less power under our stress test than Intel's 91W Core i7-7700k. It takes a significant underclock to put the Kaby Lake flagship in the first place."

>>59605585

Why is Delidlake so shit? It burned down my entire city.

>>59605125
I fucking knew it! They cut the LLC to fit the more cores in, the 10 core Broadwell-E has 25MB of cache.

Skylake as a CPU itself will be a headache, what will push it is 3D XPoint support Altera FPGA support.


But it will lose to Naples in practically everything, power, I/O, mem bandwidth, mem capacity, core count and clockspeed.

What the fuck is Intel doing.

>>59606082
The only place where current Xeons have an advantage is single core Turbo, these 2.2GHz SKUs can turbo a single or two cores to 3.9 without issue.
I don't think AMD has a granular Turbo like this with Naples, but we'll see.

In everything else it loses badly, which is an amazing thing to see again.

File: 1490031269751.jpg (19KB, 140x140px) Image search: [Google]

19KB, 140x140px

>>59606082
Cannonlake server cancelled

This is all Intel has until 2019

File: single socket holocaust.jpg (28KB, 550x550px) Image search: [Google]

28KB, 550x550px

>>59605585
The King is back.

>>59605585
Megabibline Ebition Subercocked bntel debigned binal thermal bolution

>>59605585
kek, it's Netburst all over again.

>>59605585
Wow, so out of the big three Intel is the housefire now?

What happened to those ADVANCED BEST IN THE WORLD fabs?
Or is the arch just pure shit no amount of fab expertise will help?

>>59606449

No. This is what new architectures are supposed to do.

Instead Intel sat on their ass and did nothing but raise the rent.

I'm just sorry it's not a bigger deficit in power consumption and performance.

>>59606365
The funniest thing of that all is
>Intel designed thermal solution

>>59606468

The architecture is ancient now. They've milked it all theu could. The problem is they won't have a new design until 2021.

Meanwhile Zen picks up steam.

>>59606491
Next year should be great, there's way more fixes to be done on Zen than there is on Skylake, so perf/watt will improve next year considerably.

>205w tdp

Holy shit how are they going to cool this in a 1 or 2u rack?

>>59606542
BIGGER
INDUSTRIAL
FANS

>>59605125
>18C
>3.2GHz
>130W
Are you kidding? That sounds amazing, too good to be true in fact.

>>59606542
4 Delta fans with a a PWM splitter per CPU.

Don't blame us if your server suddenly starts floating

>>59606564
Probably the turbo

>>59605125
>L3 size

What the fuck happened

>tfw the moment intel releases a new i7 I'm returning my 7700k to amazon

they wont even see me coming kek

fuck those 70ºC stress test loads on a fucking 160€ CLC

>>59605585
DELID THIS

>>59606668
That's still pretty damn good, a 6900K boosts up to 3.7 and has a 140W TDP right? This Xeon supposedly boosts up to 3.2 at 130W and on top of that has 10 extra cores (more than double).

>>59606696

Should've gone ryzen - it takes obscene volts to push ryzen 7 that hot even on air (admittedly huge dual towers like D15, silver arrow etc, but its still air).

>>59606724
I can still go, don't get me wrong.

The only loss would be the 2nd hand mobo I bought. Can't return that.

I will still probably wait for the new i7 and see how it does in benchmarks.

>>59606722
http://ark.intel.com/products/95441
Probably means something stupid like this.

>>59606690
2.75 MB per two cores.

>>59605125
So they don't even have any functioning 32c dies, and top TDP is 205w?
I'm sure AMD is shaking in their boots with their 32c/64t 180w Opteron for $5000.

>>59606770
http://ark.intel.com/products/96900/Intel-Xeon-Processor-E7-8894-v4-60M-Cache-2_40-GHz

Broadwell Xeons have 2.5MB L3 per CORE, a lot of industrial workloads will suffer from this decision.

>>59606765
Probably, it's too good to be true otherwise.

>>59606722
3.7ghz is the peak single core turbo, anon. It doesn't do that on all cores.
These parts are all super low clocked.

>>59606501
How do you get rid of baked on goyim?
The Thermal Solution

>>59606540
I'm more interested how SMT will work on 32 cores, current SMT at 8 cores can push the 1800X to reach the 10 core Broadwell-E in certain workloads, it's clearly more powerful.
If they can gain another 2% increase in SMT perf next year, after all it is their first try at it, scaling it to that many cores is gonna look sick.

>>59606542
WATER RACKS. Safest choice.

>>59606542
Dump it into space and how it works.

>>59606082
>What the fuck is Intel doing.
It's not what they're doing, it's what they didn't do. For 5 years. They sat on their ass. This is the results. AMD and ARM will tag team their stupid ass.

>>59606817
>Ryzen+
>All the hype of Zen except infinity fabric actually works this time
Desirable

>>59606846
IF has no issues when all the cores are stressed, meaning inter-CCX communication is minimal.

This is honestly only a problem with gaming, and I think this can be solved with better CCX load balancing, faster memory or just AMD making SKUs that run the fabric at full speed for the gamer market, enterprise doesn't need it because it will consume more power and performance impact will be minimal.

>>59606846
Theres literally nothing wrong with the infinity fabric in Ryzen.

>>59606833
>it into space
anything would fucking overheat since there's nothing to give away the heat to.

File: Overclocked ryzen 1700 IBT.png (548KB, 957x1073px) Image search: [Google]

548KB, 957x1073px

>>59606741

Just for you anon i've run a stress test on my shiny new 1700 just to give a point of reference.

>>59606899
Do those power sensors work? Because that's pretty impressive if that's a overclocked 1700

>>59606899
>>59606919
There's some shit about +20c on all sensors for some reason.

>>59605585
So 32 cores at 2.7-2.9GHz base? Intel is literally finished.

File: file.png (106KB, 1149x717px) Image search: [Google]

106KB, 1149x717px

>>59606899
This is just shameful.

It's running under an AIO that costs 160€, for fucks sake.

Completely stock btw.

>>59606919
The total package power measurement is accurate.

>>59606931
The 1700x and 1800x give their readout an additional 20c to allow themselves thermal margin to utilize XFR when applicable.

>>59606963
IBT is more stressful than AIDA from what I remember from my Nehalem days.

>>59606963
That's shitty TIM for you.

>18 cores
>3ghz
> low
In in for a 6c 12t skylake chip.

>>59606997
Well, I didn't set AIDA64 to also stress FPU. It goes to ridiculous temperatures if I do.

In fact, hold on 2 minutes and I'll stress it for you.

>>59607000
And I can't even buy a delidding tool where I live. Oh well, I'll return it when a new intel processor comes out, hopefully with better factory TIM.

File: silver arrow.jpg (19KB, 300x300px) Image search: [Google]

19KB, 300x300px

>>59606919

The 1700 is binned for delicious low power draw, not clocks. I'm struggling to keep that chip stable at 4ghz right now.

>>59606963

I'm using a silver arrow on the standard fan profile set in the bios (i.e not silent or turbo).

>>59606997

Said 1700 also reports similar temps after 8hrs of prime95.

File: file.png (118KB, 1317x713px) Image search: [Google]

118KB, 1317x713px

>>59607019
>>59607000
>>59606997

I wont let it run longer than that, don't want to fuck the cpu up until they release a new one.

>>59607043
HOLYYYYYYYY SHIT what the fug intel.

File: file.png (262KB, 1131x794px) Image search: [Google]

262KB, 1131x794px

>>59607037
I'm running my 7700k under this.

A fucking 280MM CLC.

File: 1489183230528.jpg (273KB, 1032x774px) Image search: [Google]

273KB, 1032x774px

>>59607043
>>59606963
Literally a meme

>>59606224
Wew lad time for Intel to lose marketshare

>>59607052

Well I do live in a country thats on the cooler side of things but ryzen is blatantly designed to do mobile and server deploymewnts where power draw (and thus temps) matter. Desktop usage cases don't give a fuck about those particularly.

Perf/watt is the mother of all IC design targets, I'm so glad AMD is back in the game.

MORE POWER

File: file.png (597KB, 537x736px) Image search: [Google]

597KB, 537x736px

>>59607111
I would consider this ambient to be rather ok for cooling, to be quite honest.

>>59607052
CLC is pleb tier shit

>>59607159

That is effectively the ambient temps that this 1700 >>59606899 faces. In fact right now because the heating isn't on its 18.4c in the room.

>>59607191
I agree with you. It's still better than the Noctua NH-D15 during the first hour or so of stress testing, until the liquid heats up from constant stressing though.

I don't mean to start a "what's the best cooler" war, I am just trying to prove a point that my 7700k is a fucking firehouse.

more like firelake amirite

File: self intoxicated, twisted detonator.png (314KB, 1131x1006px) Image search: [Google]

314KB, 1131x1006px

>>59607043

>meanwhile, on actual nuclear reactor cpus...

Oh Intel, you so crazy.

>>59607253
You hit 4.7ghz under 1.45v?
Thats pretty impressive for Vishera.

>>59607052
>7700k
That thing is shit why didn't you get a 1700

>>59607279
When I bought my 7700k ryzen wasn't out. I've also had bad experiences with AMD in the past.

If you follow my posts back a bit into the thread you can see that my plan is sending back the 7700k as soon as a new, properly cooled i7 comes out, or if that doesn't happen, get ryzen (or the equivalent whenever it happens).

>>59607273

Its not impressive - those 'e' chips are what happens after a few years of process maturity. To think the 8370e clocks over 4ghz out of the box at a 95w envelope. My gut tells me that particular 8320e is good for 5ghz on roughly 1.48v but my motherboard could never handle the load to test that.

>>59607314
Ryzen literally came out a month later, and if your bad and experience is bulldozer then that's an exception from typical AMD

>>59607314

Even when the new i7's drop you'll need a new motherboard as the skylake/kabylake socket is dead.

>>59607368
Is it really? Oh well.

>>59607357
I am aware. The problems I had with AMD all were GPU related. I only had an AMD CPU when I was rocking an athlon whatever a bunch of years back.

>>59607401

>Is it really? Oh well.

A socket change a year keeps the goyim in fear.

>>59607401
>AMD
>GPU
Where do I even begin with this, the AMD that makes CPUs is completely different from the AMD that makes GPUs

Also if you had problems with a AMD GPU that wasn't a failure that required a RMA its your own fault.

>>59607441
1151 has been going on for a while now, though.

AMD has a really impressive core, but does it have an equally impressive presence, services and support as Intel? These things take time, they don't take skill or mathematical knowledge like building ICs, they take personnel, oiling up middle manager hands with cash donations.

>>59605125
Holy shit Intel is actually bankrupt if this is true.
(or at least seriously fucked for the next 2-3 years)

>>59605125
Where's the 32 core part that was promised?

>>59607660
>presence
well ryzen 7 outsold the intel extreme series in the first week of launch vs broadwell-e having 6 months or so to gain marketshare

>>59607660
AMD should hire a decent third party company to do that for them until they get at least a few profitable quarters.

>>59607743
It probably burned their lab down.

>>59607743
There was no PSU with enough amps to support it so it's taking time strapping it on a diesel generator

>>59605125
How close to 3 GHz can binned Ryzens hit at 50W?

> Hewbrew sweating intensifies

>>59607517
It's not literally every year. It's every two years. LGA1151 launched in late 2015. It's being replaced by LGA2066 in late 2017. Servers are moving to LGA3647

File: 1473745001007.png (55KB, 1509x905px) Image search: [Google]

55KB, 1509x905px

>>59607911

>>59607911
Some of them can hit 3ghz at 50w easy, and those aren't even particularly good samples. BitsandChips got one that had pretty incredible scaling at low power levels.
Safe to say that some 32c/64t Naples Opterons will be clocked at 2.5ghz or slightly higher.

File: 1490633251804.png (49KB, 849x621px) Image search: [Google]

49KB, 849x621px

>>59607911
3.2ghz you mean

>>59605125
>compare 6154 to 6136
The 6136 is the most impressive part there, but it looks to be practically golden considering how everything else is far more inefficient even at lower core count with lower clocks.. Probably comes with a Rolex and Intel might be lucky having 2 of these chips per wafer so supply will be minimal.

Either that or just a clockspeed typo, which would make more sense.

Or even worse, these are actually TURBO frequencies and not base frequencies.
If these are Turbo ones I feel sorry for Intel

>>59608007
If these are turbo clocks Intel might as well consider not releasing Skylake-EP/EX and just release FPGAs, because nobody looking at either performance or power will touch them.

>>59607989

If i'm remembering right that isn't far off The Stilt's own results.

>>59608073
Considering the mobo situation these tests should really be done in bulk when the platform stabilizes, there were BIOS updates from MSI that actually lowered voltages required, it was a few days within launch.

>>59607989
>8c/16t
>stable 3.6ghz
>under 75w
>stable 3.8ghz
>under 100w

A refresh of Ryzen is going to be ridiculous.

File: rtPj3Rl.jpg (430KB, 2160x3840px) Image search: [Google]

430KB, 2160x3840px

>>59607314
>as soon as a new, properly cooled i7 comes out
Hope you like waiting until 2021/2022.

>>59608199
More like top end Ryzens on a HP process would be ridiculous, rest can use LPP because low power chips have plenty of uses and are a fantastic marketing point.

>>59608007
They're half turbo half baseclock, a few SKU leaked before this. 6130 2.1ghz bclk 3.4ghz turbo, 6150 2.7ghz bclk 3.4ghz turbo, 8160 2.1ghz bclk 3.2ghz turbo, etc. They'll probably never run at this turbos with all cores (even though they're labeled all core) due to thermal restraints. It's like that feature on Nvidia GPUs where you can specify a TDP and it'll never exceed it even though it can potentially turbo higher.

>>59608258
I thought base clocks means and has meant never going lower than base clock while in P0 power state.

I think even the current Broadwell's stick to that.

>>59608007
>compare 6154 to 6136
It might be to do with L2 cache. Skylake X is supposed to have 1MB of L2 cache per core, I think the big difference in power consumption might be to do with disabling 1/2 or 3/4 of the L2 on some SKUs.

What I really want to know is whether L3 cache is still fully inclusive. I'm guessing not, but Intel has gone full retard recently, so it's hard to know.

>>59605125
> L3/core drops 45% (2.5 MB -> 1.375 MB)
> only 6 more cores than Broadwell-E

can this really be right? enormous unforced error if true.

>>59608312
Yeah baseclock is minimum, the 8160 isn't going to go below 2.1ghz ever unless it's idle, but it'll never hit that 3.2ghz with all cores, even though they market it as capable of hitting 3.2ghz with all cores. While the 8168, if it has the same 3.2ghz turbo, would actually run at 3.2ghz because it's TDP is higher.

>>59608367
There's a 32 core part as well, possibly under the "Orichalcum" branding.
Lmao

>>59608367
4 more cores, Broadwell has a 24 core SKU with 60mb of L3

Intel tried to play moarcores against the moarcore king. Naples has 32 cores and 64mb of L3.

>>59608363
forget being "smart" or not, is fully inclusive L3 even really possible on large core count processors?

increased likelihood of L3 collisions mean that L2 lines start having higher probability of being left forcibly unused.

>>59608367
>>59608412
>>59608413
Broadwell-EP peaks out at 22, only the wallet-raping EX goes to 24.
Similarly the Skylake-EP will probably max out at 28 and leaving 30/32 for Skylake-EX.

>>59608363
>Skylake X is supposed to have 1MB of L2 cache per core,
Where did you hear this? Skylake desktop only has 256kb per core

>>59608529
Gold/Platinum Xeons will have 1mb of L2@core.

>>59608564
That would explain the smaller L3, but can other cores pull from other core L2? No small amount of workloads will suffer from a much smaller LLC pool

>>59608596
Who the fuck cares it'll be brutally assraped by Naples anyway.

>>59608363
>>59608529
>>59608564
If Skylake-E really has 1MB L2/core, Intel is very fucking lucky that AMD is getting scheduler debouncing patches pushed out for Ryzen.

Thread migrations will be very costly for caches that large.

>>59608609
I'm liking Naples as well but lets not be fanboy here.

>>59608596
Other cores need to pull from what's logically in other L2 caches or you just don't have a coherent memory system.

The point being raised elsewhere is that you just can't realistically have 1MB L2/core and 1.375MB L3/core if L3 is inclusive of L2.

>>59608637
Naples has more coarz, higher clocks, more I/O, more PCI-E lanes while using less power. I guess Intel has Optane.

>>59608564
>>59608596
>>59608642
Going from 256kB L2 + 2560kB L3 => 1024kB L2 + 1375kB L3 will be a pretty large performance booster for Intel on most loads.

The downsides are substantially higher power, less area efficiency, and very slightly higher latency on hits.

>>59608754
Wouldn't sticking to expected cache sizes and clocking higher be do almost the same thing? On top of it most software isn't written for this new memory hierarchy so it's certainly not a drop-in replacement for systems running Haswell or Broadlake

>>59608754
while Ryzen has 512kb L2 + 2048kb L3

>>59608931
Zen has its own cache issues, nor are all caches on the MCM shared across, so it's not directly comparable.

AMD in a year or two should start making dies for specific markets instead of sticking with one, cost and yields are nice, but efficiency is lost.

>>59608754
>>59608929
>>59608981
All software has different architecture preferences, but on Haswell/Broadwell, L2 has about 3x lower latency than L3.

The trick is finding workloads where the ~32kB L1Ds have lots of misses but where most of the working set fits comfortably in the 256+kB L2.

>>59605125
>200W
fug, who would datacenter cope with that kind of heat?

>>59606709
criminally underrated post

>>59606887
There's radiation. Just make CPU's that can withstand couple thousand C° so that energy used matches radiarion.

>>59609237
Datacenters can deal with much more, the real issue is how much is the room cooling cost when you got a huge rollout of 1/2u racks

>>59609480
Is anyone seriously gonna put a 205W TDP monster in a 1U rock?

>>59609509
>rack*

Am I the only one that misses configurable TDP options on the mobo? Why do now only server motherboards get it.

>>59609509
yeah, some people will but not everyone can.

most datacenters sell some slice of cabinets or Us with an amperage or wattage limit, usually on the order of a few kilowatts per 42U cabinet.

There are plenty of 1U servers with 1+kW redundant PSUs, but most 1U systems aren't rated for 2x ~200W CPUs, and you can't really fill up a lot of cabinets completely with such machines.

>>59609583
2U seems like it'd do fine, but 2U seems like a waste of space without GPUs or a massive array of SAS/SATA drives

>>59605559
>>Intel is dead, what a disappointment!

Fixed for you, retard.

>>59609903


Intel isn't dead, but it has been fucking around (read: Jewing) way too much and it's biting them in the ass.

Retarded management doesn't help.

>>59605559
You really bought into Intel's bullshit about Moore's law? They only hit the limits of their own architecture. Mostly because they were too lazy and greedy to come up with a new one. Swear to god you Intel shills just keep getting more and more desperate.

>>59606542
180W naples 32/64 are 1U compatible. Don't fret. It's possible.

>>59610330
Whose to say there aren't 150W parts? Possibly clocked at somewhere around 2.2.

>>59607743
It is a monolithic cpu, so they need to make a perfect 32 core die.

Not going to happen all that often, obviously,

>>59610417
Maybe it's not gonna be out on launch day and released when yields aren't shit, kinda like Broadwell-EX 24 core.

>>59610387
What matters is powers/flop, not total package power. If you could make 1kW chip with 1000 cores, it would sell. At that density you could even splurge for a custom ice slurry cooling.

>>59605125
>those puny caches and absurd tdp
kek, intel is finished

Some people might buy Skylake-E for AVX-512. I'm not sure what kind of application needs this instead of GPU, though.