i9 Speculation

You're going by the assumption that Intel can maintain those turbo clocks under TDP

It can't, at least not more than 6 seconds.

>>61854637
anon, did you count in performance degradation from more cores on intel "morebingbus" scheme?
it doesn't scale like zen 90%

>>61854686
I'm not sure how I would factor that in or if it would be reflected in any current results. Metal Xeons don't seem to have any problem scaling. It's not like the old ringbus.

There are cinnebench runs on a 18 core Xeon that this is derived from, it scores 3000, naturally because it can't clock all the cores that high unless you want a housefire.

>>61854718
Two possibilities: The i9-7980XE is better binned than the Xeon, or Intel is playing games with their TDP again.

>>61854736
>is better binned than the Xeon
haha, nice joke

>>61854736
Both have the same TDP and all core turbo(3.4) they're the same chips with shit like RAS capabilities and ECC turned off, oh and 4 less lanes, nobody seems to mention the xeons are 48 lanes.

>>61854670
Throttling would be readily apparent on any test that takes longer than a few seconds, then.

>>61854765
And now you know why Skylake-X needs a 480mm rad and pulls 220W+ at stock

Your napkin math is pointless.

LCC vs MCC die.
TDP is different.
You're still feeding 18 cores with 4 channel memory

File: xeon gold.jpg (349KB, 1060x1047px) Image search: [Google]

349KB, 1060x1047px

>>61854752
Well it does look like it's boosting to 3.4 GHz, if it has scaling issues that bad, it's not looking good. Would only be 10% faster on average at twice the price.

>>61854752
Did anyone ever confirm whether Threadripper has ECC support? It was rumoured that it does.

>>61854836
It does, but only UDIMMs, so 1TB limit on memory.

>>61854819
His napkin math was only about 10% off according to
>>61854718


>>61854843
>It does, but only UDIMMs, so 1TB limit on memory
HAHAHAHAHAHA

>Intel: ECC? What do you want hardware raid for? Help, police, this man is trying to run a VM!
>AMD: sure just don't try to make a datacenter

>>61854831
Scaling isn't the problem, power draw is.
If the 18 core can run at sustained 3.4GHz then it would be some 15-20% faster than threadripper, but doing so it would use way more power.

>>61854782
480mm rad is fucking overkill for 220w, a modest air cooler can dissipate 220w assuming you can move the heat away from the die and for skylake-x is just not possible with the TIM under the IHS

>>61854865
It's effectively running at 3 GHz with those results.

>>61854875
Well, your only choice is to delid and risk ruining your chip, or just get a better cooling solution even if it'll only lower temps slightly.

>>61854875
Chances on Intel soldering this time?

>>61854885
>Delid a $2000 chip
>20% failure rate for people delidding
>chip actually costs $2400 average
kek

>>61854894
I doubt it, base clock is only 2.2ghz so they can keep the heat low enough to not kill it and get away with saying the damn thing isn't throttling

File: HOLOCAUST2.0.jpg (1MB, 1912x2632px) Image search: [Google]

1MB, 1912x2632px

>>61854637
BTFO

File: 1502124729070.jpg (224KB, 1280x720px) Image search: [Google]

224KB, 1280x720px

>>61854961
It's 2.6 GHz, not that low.

>>61854977
That would be pretty amusing but would also instantly double the TDP, so probably not happening.

>>61854979
must've remembered an old side or something, but still 2.6 is still pretty low. 2.6ghz would keep all the cores at near peak efficiency to keep heat down and again let intel claim there shit isn't throttling

>>61854686
>it doesn't scale like zen 90%
That's because Intel doesn't use [Infinity Glue]™

File: Naamloos2.jpg (35KB, 600x600px) Image search: [Google]

35KB, 600x600px

>>61855092
>2.6GHz
>2.6% yields
>26% stock drop

>>61855092
That would be peak efficiency, but 2.7 is also Zen's peak efficiency, 3.4 on TR is past the 1st critical point.

That's why the 1950 non X is gonna be way more interesting to me, it'll be 10% slower but use 30% less power than a 1950X

>>61855221
I bet AMD will launch them to shit on Intel's parade when they launch the 14 core and higher.

>>61855221
>but use 30% less power than a 1950X
Would be cool. It's like a 16-core CPU at the TDP of a 7700X.

>>61855469
>K*

>>61854686
I haven't seen any testing to confirm that Intel's bingmesh doesn't scale well.

>>61855221
AMD hasn't launched that yet because they want people to buy the more expensive 1950X, the lower TDP cheaper models are better buys at these core counts.

>>61854736
Given what we've seen about SKL-X in general and the 7900X in particular, it seems to be the latter. We'll all see on Aug 28.

>>61855003
Not sure what you think you're talking about, it's literally what happened.

Thread music
https://m.youtube.com/watch?v=rWlXU2DeYkQ

>>61855531
>Aug 28
They're launching October 18th.

>>61855556
The 7920X launches Aug 28.

>>61854843
128 GB DDR4 UDIMMs don't exist. Shit, I'm not sure if 128 GB RDIMMs (i.e., not LRDIMMs) even exist.

>>61855573
Nobody says they won't exist, if AMD is saying TR supports 1TB with UDIMMs, there's probably something coming out in the pipeline sooner or latter.

>>61855203

>2.6 shekels per successful shitpost

>>61855685
I honestly don't see enough people wanting 512-1024GB desktops in the next 3-4 years before DDR5 arrives for memory manufacturers to bother even trying this scheme.

>>61855685
Not saying it's not possible, but going for EPYC would be a lot more cheaper.

You can get 1TB with only 64GB DImms

>>61854637
I bet over 12 cores it will be soldered
Can't be other way

>>61855963
We'll see if Intel's run out of bullets in the revolver they've been using to shoot themselves in the foot with all year.

>>61855963
I hope, but even at that you're paying way more money for similar performance.

Also solder will only help power consumption marginally, you'll still be looking at 300W housefires.

>>61856120
>implying Vega isn't a 300W housefire

>>61856147
It is, but what's that got to do with Skylake-X being a oven?

File: Corehunter.jpg (79KB, 600x420px) Image search: [Google]

79KB, 600x420px

>>61856014

>>61856120
It's gonna be Broadwell-E vs Ryzen again, you get 5% more performance and worse power consumption at close to double the cost.
But this time you don't get ECC nor more lanes either.

Also TR is a high margin product, it'll get a slight price cut sooner or latter just to fuck with Intel.

>>61856120
Intel isn't looking for power saving here just performance trophy

>>61856682
If AMD wanted the performance trophy it can just re-purpose a EPYC SKU, cut out some RAS features, and ask supermicro to put out a few thousand motherboards for desktop, and Intel wouldn't be able to do shit.

It won't because that's just dickwaving

>>61856739
Speaking of that, the 18 core Skylake is priced so high it literally competes with EPYC on price

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/3


The 32 core 1P EPYC is $2100, and you get all the server goodies, 3 times the I/O, double memory channels, VM encryption and actual decent motherboards.

For what reason would the Skylake-X exist?

>>61856822
>For what reason would the Skylake-X exist?
There's none, it's an abomination. There's not a single processor in that line up that has right to exist, or at least not at their current prices when Ryzen/TR exist. I'm really curious to see how Coffee Lake will be though.

>>61856909
If there's no major regression coffee will be pretty good.

The regression part I'm worried about is Intel's own slides where they compare a mobile 2 core to mobile 4 core, the 4 core is only 30% faster.

And that's Sysmark of all things..