INTLEL BTFO

Something like that would be $4-500 easy.
Though it'd be great if they included an HBM stack on the Vega core like the Incel Vega.

Except one Intel core= two amd cores.

The y could replace two of the cores on the chip with a single Vega core and not overly increase complexity. That'd still allow a max 16 core chip with an iGPU.

just imagine how good their yields are that they can literally price it for such a low price

where were you when Intel got BLOWN THE FUCK OUT THIS UNIVERSE

>Isn't that price for the 18-core Incel too?
just checked on Mindfactor: it's 1868€ there.

Attached: 1529687324143.gif (312x234, 2.85M)

Even with the new process they must still be pushing 70+% yields.

No.

Last node had 80% full fat yields and 99+% die usage. Its not out of the question that AMD is likely getting similar yields with the 12nm node.

Attached: 1525272814701.jpg (480x472, 23K)

Attached: 1528288212273.png (1452x892, 1.28M)

AMD's 32-core Ryzen Threadripper 2990X appears on 3DMARK

overclock3d.net/news/cpu_mainboard/amd_s_32-core_ryzen_threadripper_2990x_appears_on_3dmark/1

Maybe in terms of power consumption

INTEL IS FINISHED

>Threadripper APU
why not just plug in an RX 550/GTX 1030/anything that runs solely off PCI/E power if you need grafics to setup your server
>using threadripper for not server usage
wtf are you doing?

top kek

Nice try goy, the year is 2018, not 2011

LOL, no they couldn't. That'd require a major reworking of the architecture, AND the fabrication process, it's simply not going to happen. At least not this generation.

It isn't 2011 anymore, get with the times, Grandpa

Attached: delicia.jpg (395x558, 31K)

12nm isnt really a new node, just an improvement of the 14nm one they were using. Yields shouldn't be any worse. Also its the same die we see in the ryzen chips and the prices for those didn't increase either compared to the first gen.

>66516061

That's actually a really really interesting idea.

16c/32th Vega 44, on a single die. That's would be fantastic!

8c/16th Vega 22 on a HEDT is meh, but would probably sell well.

>That'd require a major reworking of the architecture
No more than the APUs currently available compared to the regular desktop Zens did. The experience gained during the RR development would certainly be helpful as well.

>joking few weeks ago that TR2 32c will cost 1500€
>it's actually 1509€
TOP JEJ

Attached: 1519246488467.jpg (640x632, 76K)

>Dat pic
REEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE

>Spending 1500 euros on a cpu
I shiggy diggy

>TR1 specs
Hmmm.

why didn't you stay on /v/?

>12nm isnt really a new node
exactly.
I don't know why people get confused by some number.
the 12nm is most marketing less 12nm.
The pitch and the rest dimensions are _the same_ as it was with 14nm.
What has changed was prolly some ASIC libraries, masks and slightly the methodology in order to increase yields or get better quality in produced chips.
As a result, the yields are in the worst case the same a it was last year with the 14nm node.

Attached: 1447161347847.jpg (1846x1923, 607K)

lol, called it, similar price as the 7980XE

Intlel's Chernobyl Edition™ CPU still costs ~300€ more in germany if the leaked price of 2990x is real

nice job

2990X will probably cost roughly 1800 burgers

PLS DELID DIS

Attached: 1527629778452.jpg (679x758, 54K)

That includes 20% VAT.

hell yeah it does

Attached: putin adidas.jpg (503x401, 38K)

Who the fuck tries to delid a soldered chip?

that puts the price at 1400 burguers
holy fuck the 2920X and 2950X will be dirt cheap

Intelshills, so they can repaste it

Attached: hx082DG3TVg.jpg (1280x720, 103K)

goy boy

Attached: incelqualitybait.png (722x431, 85K)

Are you blind? Read that again.

At least the tampons have leak protection

It was sarcasm, you're not fun

>2 Zen
>1 Vega (what like 32 CU's?, I'm pushing it from the Vega and RR Die Shots)
>1 HBM

Are you always this stupid or is today a special occasion

underrated post

mask compatibility was actually a big thing, you can keep using them. zen+ probably only required a respin of the base layer and some of the first metal layers

Jesus how new are you

Attached: 1528222249083.png (1196x676, 1.05M)

Oh snap!

Attached: 1525359495425.jpg (600x600, 24K)

kek

So many posts and not one brainlet asks how the two additional cores are wired to PCI-E lanes and DIMM slots...
> pro-tip : they're not
> tfw 1950x owner.. inb4, intel shill
This is a hard pan until they define how the new cores are wired and until detailed benchmarks highlight the performance hit. The new cores literally have no direct wiring to I/O.

Don't forget your chiller, bro.

Attached: 8348_13_intel-core-i9-7980xe-7960x-cpu-review.png (615x773, 27K)

TR4 1950x very cheap

Attached: TR4.png (1452x772, 602K)

so 882 burgers?

PLS BUY INTLEL OUR 16 CORES ARE DEFINITELY BETTER

Attached: 8379_09_amd-threadripper-vs-intel-core-i9-clock.png (616x773, 25K)

Magic.

Attached: 1528575394592.jpg (948x747, 74K)

yep

Not magic. They two new dies (16 cores) worth likely and literally have to siphon all of their I/O from the two dies that compromise 1950x. All of the PCI-E i/o has to go through another die. All of the DIMM slot access has to go through another die. All access to hard drives does as well.

Devil's in the details on this one. It is nice that the 1950x is dropping in price. However, with a 1700 accessible at $170 a hard sell to justify a leap to $800 for 8 more cores. All together, AMD is making incredible progress and I am loving the price drops. However, at $1,500 and given how its wired/gimped this is a no go for me until more details come out.. shocked they announced it and pricing yet no detail as to how in the world this thing functions. I'd assume someone buying one would have knowledge enough to question this.

Isn't this slower than the last gen Threadrippers because of the extra latency in the interconnects and dies?

We already know they're connected through the xgmi only. Anything more would require new designs which isn't going to happen while tr is epyc's left overs. First leaks have shown it to perform decently under render and encode type workloads.

>Anything more would require new designs
of package or socket, not the CPU. A new package that routes once channel to each die shouldn't be a problem at all, apparently they just aren't willing to invest the work for a niche product.

But how does epyc work? Why can't they rename their 32 core epyc line into a 32 core TR line? They don't want to copy intel?

TR4 and SP3 are mechanically the same socket but the pins are routed differently. Epyc and TR use the same package for the dies but the TR4 socket appears to be only able to route memory and PCIe lanes to two of the four positions for chips on the fabric.

I guess it was done that way to save costs for the mainboards? No idea. If what AMD published is true threadripper wasn't really planned, some engineers just went ahead and thought it would be a neat halo product for their desktop line that could be done for cheap with little to no additional developement needed. They probably weren't expecting the platform to actually be a significant success, not to mention that they'd ever push the fucking 32 cores into it.

That's what I'm saying, just get Zen1 epyc and make it TR2, with socket and all. The price is already higher, but at least it would function better.

that would require new mainboards, which is something AMD definetly doesn't want to do. They are coming back from nothing, building an ecosystem is worth a lot to them, more than a bit more performance for their HEDT platform that is better than the competitions one already anyways.

also another point, if there are any epyc products that for some reason have two broken memory channels per one die/to many broken PCIe lanes per one die they can just sell it as 48/32 core threadripper. That wouldn't be possible if DRAM/PCIe access was spread out evenly across all 4 dies.

Fair enough. I didn't think they should actually do it, but it's nice to know why. Thanks.

Look pretty much the same?

Almost

We still don't know, by 4 die EPYC benchmarks it shouldn't be any worse, but rumor it's that TR will only do dual channel memory, which implies that only two dies will have their memory controller enabled

quad channel over two dies. And that's not a rumor, anandtech has it in their article

Watch this shit have like 15c difference between dies

Attached: chipthumb.jpg (600x503, 91K)

Using the whole
>SHILL BANKRUPT
>SHILL SUICIDE

Shill
amdflaws.com

> We already know they're connected through the xgmi only. Anything more would require new designs which isn't going to happen while tr is epyc's left overs.
Which makes it a shit tier Frankenstein beyond linear, non-branching, and or highly cache-able work loads.

> First leaks have shown it to perform decently under render and encode type workloads.
Highly cache-able straight forward workloads.
As one can expect. $800/$750 for a proper 16 core.
or $1500 for a seriously neutered epyc. I guess this works out for render/encoders.

I'll wait until all of the cores are properly fed.
I can paint a number of scenarios for workstation class loads where this would perform like dog shit because it has absolutely no direct access to chipset periphs, pci-e, or dimm modules. If you have a process running on one of these cores that needs to access said I/O, you're fucked for performance. Incredibly fucked if you have shared resources across to the dies that are actually wired. NUMA is already hard enough to navigate, i have no desire to enter into another freak show.

>Shill
>amdflaws.com
>Defunct "flaws" that were software patched weeks ago.

Attached: 1303896696205.jpg (640x512, 63K)

>published by some cutthroat group that was prosecuted for stock manipulation before
>without giving the companys invovled the usuals 3 months to fix it before going public

ITS TOTALLY LEGITIMATE!!

Attached: 1448325429005.jpg (960x720, 107K)

$750 USD

Attached: Screenshot_2018-06-27_16-27-04.png (1000x391, 115K)

>3.4 GHz base, 4.0 GHz boost
Nice, that's higher than what they showed in Computex.

They wont do this because it will cut into profit. Threadripper is a purpose built gimp platform for an Epyc processor. I'm fine with that given the price decrease that comes along with it. However, wedging 2 more dies into this platform that aren't properly fed is where the buck stops for me. TR is fine at 2 die 1950x. However, I want a completely different platform if they intend to scale beyond this. It's time they bring the antiquated server technology they keep hiding stupid margins behind down to the desktop like RDIMM vs DIMM and the ability to go beyond 128GB ram. Epyc by comparison supports up to 2TB. Also, if the board can't fit the components which it obviously can't. it's time to target e-atx or go even further up in size to something that can. In essence, there's diminishing returns now in this platform/approach.

and XFR is 4.2 GHz I think

>nazi CPU intercommunications
oy vey!

ITT schizophrenic pajeet having an outburst

topkek

Any source to that claim? I thought that would happen but I haven't seen anything relating to that as yet.

Most all tech sites were forced to update their articles about the AMDFlaws website that popped up. The financier behind it all has tried this with other companies in the past. The people behind it were literal Israeli intelligence agents.

Real World Tech and most other reputable sites stated right away that it was obvious stock manipulation.

>DAMEGE CONTROL

JUST!

Attached: 1505147990486.jpg (329x329, 57K)

PLS DELID DIS

Attached: 1527629778452.jpg (679x758, 54K)