Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Tum Apisak on twitter posted 3dmark fire strike numbers for lakefield
tum apisak on twitter
graphics score 11xx
physics score 52xx
For comparison
A lakefield-u with 64 Execution Units and 1100mhz and gen11 graphics has 2200 graphics score
Lakefield has gen11 aswell and 64EU as well.
So this means that Lakefield gfx is running with 550MHz.
So this means Lakefield is in terms performance per watt a real monster.
They throwed a massive number of transistors at this die just to be able to have a low clock rate.
Lakefield has the same fire strike graphics score as a core i-5 8250u.
This is amazing.
Both days are with volume.
Both days around 4%
But there is no "special" news around.
I searched for news, but I can't find anything (beyond the usual).
Some small bits and peaces.
On both days the market is going up but Intel is one of the spearheads.
I thought yesterdays investors meeting was good news.
They talked about a lot of products and they are a little ahead of schedule(compared to the leaked DELL-Roadmap).
The Lakefield Foveros processor is able to boost to 27W power envelope.
This means with a good cooling solution you are able to play games with it.
The CPU + GPU part is a lot stronger then Nintendo Switch.
But in low power mode it is consuming 2.6 mW.
next Supercomputer of DOE
The next supercomputer of the DOE will be disappointing.
There are really some disturbing details in the conference call, in the leaked roadmaps and inside the coreboot github repository.
Perhaps we are realy hitting the end of moores law.
The resistance of BEOL is rising with every shrink.
So this will have a direct effect on the FO4-metrics of the chip.
So the pipeline stages in itself have to be redesigned to fit into the next process node.
But a handcrafted and handoptimized pipeline could be superior compared to a automatically generated pipeline stage for a smaller node.
If Intel is able to create a 10nm process which is able to build Xeon dies with 28 cores and 4ghz of frequency then it is possible that they are the only company which is able to build this processors.
The CPU has to be handcrafted to supplement the metrics of the process.
It seems to be easier to build a mobile CPU.
So they killed SoFia a couple of years ago and now Sofia comes back with Lakefield Foveros.
But now they have a new partner in crime with Microsoft.
Microsoft said they don't want to create yet another smartphone which don't have unique features - instead they want to wait until a new form factor is starting and they want to be in front of the wave.
So foveros lakefield and the 5g modem(remember they only stopped the development of the SMARTPHONE modem) will be a new foundation for the foldable Andromeda / Santorini devices.
Lakefield already has an IPU (Image Processing Unit) important for the camera. This was one reason why Sofia was a lame duck. Because camera on SoFia-Phones was mediocre.
I am courious about the amount of ram inside foveros lakefield.
Is it
[] 4gb
[] 8gb
[] 16gb
But if you look at the tweakers.net-roadmap there is a disturbing detail hidden.
The first 10nm 6core cpu will be in january 2021.
So the hiring of Raja Koduri and Jim Keller is because they need other calculation pipelines / not so "demanding" pipelines inside their portfolio (just to fill the fabs).
10nm++ will be the last process for big monolithic single-thread-heavy cores.
Perhaps we are now seeing 18inch 450mm wafers just to lower the cost of one die.
So a Cloud-CPU will have 28cores on one die but inside one package there are maybe 10 dies. And so we will see a 280-Core package.
The core with the higher single-thread-performance is the winner in this race.
Intel will get additional tailwind from three technical factors:
1. Intel gets back the silicon vendor leadership which Samsung had in the last 3 years. Flash and DRAM business is slowing down and in contrast to this Intel is growing(small). But it is enough to get back the leadership. So ETF with focus on PHLX semiconductor Index has to rebalance the portfolio.
2. The Global Select dividend ETF has stocks with the biggest dividend companies. Automotive companies are cutting there dividends. But Intel is keeping them where they are. So Intel is coninously rising in the Global Select dividend Ranking.
This fonts are currently reranking/rebalancing.
3. Intel has built-up (not voluntarily) some production ready products (chip designs) but was not able to deliver (because of trouble with 10nm)
So we will see the release of innovative products over and over.
(compare it with the jump from 28nm-TSMC-nvidia-chips to 14nm-TSMC-nvidia-chips)
How will this pieces all fit together:
This is the year of the 10Gbit Ethernet
There is a big gap between the speed of NVME drives and the 1gbit ethernet.
And we have fiber-internet-acess with 1gbit but inside the local the network this all goes down the drain.
And I see more devices with cords back again.
Fast NAS with 10gbit ethernet and NVME inside.
Raja Koduri and Jim Keller are speakers at this event.
NDA until tomorrow.
Intel briefing today at Robert Noyce house.
.@intel briefing today at Robert Noyce's house! #Intel pic.twitter.com/AN5GyXinuy
— Legit Reviews (@LegitReviews) December 11, 2018
In other words (if it wasn't entirely clear) according to my source' connections that work at Intel, Intel is now planning to "mostly skip 10nm". (RIP 10nm.) https://t.co/fvo7btg9W7
— witeken (@witeken) June 22, 2018
Tiger Lake 10nm++ possibly EUV
2021??
Here they are talking about 10nm 10nm+ and 10nm++
https://www.anandtech.com/show/11722/intel-reveals-ice-lake-core-architecture-10nm-plus
Afaik
10nm immersion __ Cannon Lake
10nm+ immersion __ Ice Lake
10nm++ EUV ________ ??
Our build to order service partner called us today that we should order early and ahead of time because there will be shortages in the next couple of month.
I think that Diane Bryant is a done deal.
But she has to be quiet for some month.
Her work at google ended August 31.
So now one or two month between old and new job.
And then she is able to start the new job.
My next Workstation around the corner
https://skinflint.co.uk/asus-ws-c246-pro-90sw00g0-m0eay0-a1857698.html
https://skinflint.co.uk/intel-xeon-e-2146g-cm8068403380116-a1859196.html
Perhaps not the Asus MoBo but a Supermicro.
But they are not listed yet.
The power of AVX512 on my 3DPM benchmark:
— 𝐷𝑟. 𝐼𝑎𝑛 𝐶𝑢𝑡𝑟𝑒𝑠𝑠 (@IanCutress) July 23, 2018
On Intel's 10nm dual core CannonLake, non AVX512 scored 382; AVX512 code scored 4171.
On 18-core Core i9-7890XE, jump from 4111 to 30750.
To confirm, AVX512-enabled code on a 2.2GHz 2C outperformed non-AVX code on a 3.4GHz 18C.
Samsung 7nm in 2019
If Intel 10nm and Samsung 7nm is comparable with each other then Intel is the winner of this race.
Computex is starting in 3 weeks and we will see devices with 10nm Intel process.
There is a delay between "we will ship first products in December 2017" and what actually happens.
But remember that Intel had to fight with Spectre and Meltdown.
I mean look at the stock price - this is absolutely nice.
There are first laptops with 10nm CPU out:
Lenovo Lapto with 10nm Intel CPU
And there are first NUCs with 10nm CPU out:
Intel NUC with 10nm CPU
The flagship phones from Samsung and Apple are both with 10nm.
The Iphone X is with A11-CPU with TSMC 10nm-Process
The Samsung S9 ist with Exynos 9810 with Samsung 10nm-Process
Perhaps the Samsung Galaxy Note 9 in September is with a 7nm CPU.
But currently Intel has the lead.
Real men have real process steps
I mean look at the gpu business.
They had 28nm for 4 years. Then there cam 14nm.
No step in betwwen.
The next real step is 7nm.
Intel is now 14nm for four years.
Next stop 10nm then 10nm+ then 10nm++
10nm is broken.
Ice Lake is based on 10nm+.
But I think they have to produce at least one sku of cpus in 10nm to see how it works.
I think this 10nm process is strong.
There was a massive jump from Geforce 980ti(28nm) to 1080ti(14nm).
I think there are two possible scenarios.
First scenario:
There is a quiet period before fiscal numbers.
So every quarter there is only a Window of 60 days where a CEO is able to sell stocks (perhaps even less because of some other events).
During February March April May there was the acquisition of Mobileye.
So he wasn't able to sell stocks in this timeframe (because of the same reason, he had insider knowledge....).
Compared to other leaders this selling of stocks is peanuts.
I mean Lerry Ellison, Elon Musk, Jeff Bezos they all sell stocks in completely different regions.
I mean look at the stock price.
There is a negligible price-difference between the day Krzanich was selling the stocks and now.
Second Scenario:
All industry-players said that they are working close together to fix this.
And with a micro-code-update it is possible to fix this mess.
But in a first press-release some researchers said that all CPUs are broken.
I think that this was one of the hidden security issues that the NSA actively used.
And Intel already had some "remedy" integrated into silicon which was hidden.
So now with a micro-code-update the "hidden Plan-B" is saving the silicon out there.
If the NSA is behind this - this is the "golden handshake" for Mr Krzanich.
I mean look at T-Mobile USA.
T-Mobile Germany wanted to sell T-Mobile USA.
USA said you are not allowed to.
One quarter later they had the best earnings numbers ever.
What a coincidence.
https://xenbits.xen.org/xsa/advisory-254.html
Hardware Virtualization with Xen is not vulnerable to Meltdown
Performance impact of the patch on average is said to be 5%, with some going as far as 30%.
Yup, that's also how I understood it from the article I have linked to. I also got it wrong initially, since many headlines claim that it's integrated. The GPU is not integrated as IP but just a separate die on the interposer.
Intel had a cross-licensing-agreement with nvidia for some patents.
There was no subsequent agreement with nvidia.
So I think we have the confirmation here that Intel is cross-licensing the GPU-patents with AMD now.
Perhaps Apples next SoC could be fabbed at Intel
https://www.eetimes.com/document.asp?_mc=RSS%5FEET%5FEDT&doc_id=1332527&page_number=2
Intel is able to fab the LTE-modem and the Apple-CPU+GPU on one die.
So Intel could be the first source.
And one out of TSMC or Samsung could fab CPU+GPU without the modem second source.
Here is a complete pinout of the connector:
https://gzhls.at/blob/ldb/b/f/4/d/692e53af35741006efc7208e07718c3e0029.pdf
There are 2 PCIE 3.0 x1 lanes available to the outside world.
It should be possible to connect a GPU to this compute cards.
sorry I linked to the wrong compute card this one is with an Atom processor.
But Intel builds a Compute card with Core M Y
https://skinflint.co.uk/intel-compute-card-cd1m3128mk-a1710646.html
In germany it is in stock in some shops:
http://www.pcking.de/eshop.php?action=article_detail&s_supplier_aid=8188511&rid=geizhals
Here is the Mobo of the Core M Y version:
Wanna know what the inside of an Intel Compute Card looks like? Here ya go. This is a Celeron "Apollo Lake" based model. (c) WinFuture.de pic.twitter.com/E7rDGfSLBf
— Roland Quandt (@rquandt) June 1, 2017
They already hit the form factor with the 14nm Core M Y CPU:
https://liliputing.com/2017/06/zotac-pi225-fanless-desktop-pc-thats-smaller-2-5-inch-hard-drive.html
Intel has Coffeelake, or 14nm 6 core mobile chips coming in few months. On the low power side, they should release 10nm 5.2W Core M Y chips end of this year.
$$$ Money Money Money $$$
I own shares of both companies.
INTC since 2013 and NVDA since Feb 2016.
The plus with NVDA is astonishing.
But I don't think that NVDA is in autonomous driving in a very good position.
INTC made better progress in autonoumous driving.
INTC is certifying a real-time-os, certifying an FPGA, an Tensor-Flow-Processor Nervana and bringing it together with sensor hardware and camera hardware.
INTC has a lot of knowledge with the realsense technology.
In terms of real-time-requirements and delivering reliable response times and certification of this approach -> INTC is far ahead.
Intel A.I.-Processor Nervana
https://www.engadget.com/2017/10/17/intel-ai-deep-learning-nervana-npp/
Intel very close to 40$
Looks a little like a breakout.
i7-8xxx Coffee-Lake seems to be a good product.
And there is still a small probability that we will 10nm-intel-mobile-cpus this year.
I think this is still not priced in.
German website computerbase describes the technology behind autonomous driving inside the new Audi A8.
https://www.computerbase.de/2017-07/audi-a8-hands-on/
Sorry this is only in german language but so far the most detailed look inside the black boxes.
Audi don't have the green light to use it on the streets so far.
In germany there was the election one week ago.
But it seems to be a done deal when the new ministry of transport is up running.
All the source-code-certification-stuff and questions regarding the car insurance are close to being completed.
So there are SAE Level 3 cars on the road in February 2018 in germany.
Waymo - a part of the Alphabet Foundation is working together with Intel
https://www.theverge.com/2017/9/18/16328284/intel-waymo-partnership-self-driving-car-chrysler