Processador ARM for server

Da AMD só se ouviu dizer que o K12 não era prioridade, o Ryzen atrasou uns meses, e neste momento também não sei se lançar a 14nm seria boa ideia, esperar para ver se quando os 7nm estiverem prontos a AMD volta ou não.

Já vi noticias que o Jim Keller estava contra o projecto K12 e outras que dizem que ele saiu da AMD porque a AMD abandonou o K12. Isto é contraditório.

Em relação ao primeiro processador ARM da AMD, foi apresentado em 2015 como um Octa-core A57 e só uma board chegou aos consumidores, em julho de 2017 e só com a versão quad-core. O fabricante dessa board satisfez as pre-orders e desde então a board esteve sempre em Out-of-Stock.
Em 2017, este cpu não é competitivo em nenhum ponto e tem uma arquitectura antiga. O poder ter 14 portas sata e 2 portas 10 Gbit é interessante, mas não foi posto em prática.

Acho que por agora, ARM na AMD, está abandonado.

E970KLQ.jpg


https://www.phoronix.com/scan.php?page=news_item&px=AMD-ARM-In-My-Hands
http://www.lenovator.com/product/103.html
 
Também já nem valia a pena produzi-las, até porque essa board teve problemas com o PCI-e descoberto coma entrega das primeiras unidades, o que fez com que a produção fosse suspensa, foi depois retomada, mas ainda sem a PCI-e funcional,

But precisely with the expansion slot and there is a problem that recognizes the developer himself . At the moment, it found that the cause of the problems lies not in hardware. After a thorough review, the developers LeMaker Cello found that defects in the hardware is not present, the more that all other interfaces, including SATA and the Ethernet, are working properly. It seems that all the problems associated with PCI Express, have a soft nature and can be solved by issuing the appropriate firmware upgrade over what the creators of the board, together with AMD and are now working.
http://www.startlr.com/mainboard-amd-lemaker-cello-has-problems-with-bus-pcie/
 
Estes gajos dos National Labs americanos :n1qshok:

The Los Alamos National Laboratory’s High Performance Computing Division now has access to 750-node Raspberry Pi clusters as part of the first step towards a development program to assist in programming much larger machines.

bitscope-cluster-module-zwlj6pz3.jpg


The platform at LANL leverages a modular cluster design from BitScope Designs, with five rack-mount Bitscope Cluster Modules, each with 150 Raspberry Pi boards with integrated network switches. With each of the 750 chips packing four cores, it offers a 3000-core highly parallelizable platform that emulates an ARM-based supercomputer, allowing researchers to test development code without requiring a power-hungry machine at significant cost to the taxpayer. The full 750-node cluster, running 2-3 W per processor, runs at 1000W idle, 3000W at typical and 4000W at peak (with the switches)

“It’s not like you can keep a petascale machine around for R&D work in scalable systems software. The Raspberry Pi modules let developers figure out how to write this software and get it to work reliably without having a dedicated testbed of the same size, which would cost a quarter billion dollars and use 25 megawatts of electricity.” Said Gary Grider, leader of the High Performance Computing Division at Los Alamos National Laboratory.
https://www.anandtech.com/show/1203...has-750node-raspberry-pi-development-clusters
 
Eu olho para isso e vejo o sentido que faz. É um corte de custo brutal, e mantém a disponibilidade de mais cores. O preço típico do PI é estupidamente baixo, e consumo também. E o CPU sem um OS pesado lá em cima é capaz de bastante até.

E depois de ter visto a torre de 40 PI's, sempre pensei "até que ponto poderemos escalar isto de forma útli?". Estou bastante curioso de saber como se porta em termos de rede, no entanto, porque fazer comunicar 750 PI's e garantir throughpt da rede para toda a coumincação que tem de acontecer, não é do pé para a mão.

Espero agora pelo próximo que arranjar um do género com 1000 nós :D
 

Faz mesmo muito sentido. Barato, consome pouco, ARM 64 bit, boot de todos os nós a partir de um SD card ou por PXE. Muito interessante mesmo.

Uma entrevista ao responsável da Empresa que criou este cluster:


E depois de ter visto a torre de 40 PI's, sempre pensei "até que ponto poderemos escalar isto de forma útli?". Estou bastante curioso de saber como se porta em termos de rede, no entanto, porque fazer comunicar 750 PI's e garantir throughpt da rede para toda a coumincação que tem de acontecer, não é do pé para a mão.

750 nós a 100 Mbits vai ter uma performance miserável. No entanto, como a parte da performance não tem qualquer interesse para o caso, não há qualquer problema.


A questão é saber se a Marvell está interessada e tem $ para continuar a vender processadores ARM para servidores. Tanto a Cavium como a Marvell têm grande ligação à área de networking e processadores para servidores ARM pode fugir ao interesse deles. Penso que o ThunderX2 "Vulcan" não está em perigo, porque está meio apresentado e há produtos anunciados. Vamos ver se lançam o ThunderX2 "antigo" e se vai haver um ThunderX3.
Além disso, há rumores que se a Broadcom não conseguir comprar a Qualcomm (a questão ainda não acabou), que vai tentar a compra da Marvell e se a Broadcom desistiu do Vulcan anteriormente, não deve ser agora que vai ter interesse.

Mas falando do ThunderX2 "Vulcan":

Isto foi o que foi apresentado em 2013, quando foi anunciado o projecto (NOTA: Pode ter mudado desde então):

iUw4ezs.jpg


kuBiMKm.jpg


5BHOY7E.jpg


The original idea Broadcom had with the Vulcan chips was to be the first to market with a 16 nanometer FinFET processor. The homegrown ARMv8 core was to have a four-issue, out of order, superscalar core, something akin to a Power7 core, but actually inspired by the NetLogic XLP MIPS-based network processors that Broadcom had acquired several years earlier. Each core was to have four threads – more than the two of a Xeon core but less than the eight of a Power or Sparc core. The Vulcan core had 32 KB of L1 instruction and 32 KB of L1 data cache plus 256 KB of L2 cache. The core could issue for instructions per clock cycle and could have 180 outstanding instructions in flight; it also supported both 32-bit and 64-bit Arm processing modes. (Qualcomm ripped out the 32-bit support in its “Falkor” core on the Amberwing chip to squeeze more cores on the die. Interestingly, the Vulcan core had two Neon floating point units, which together could process four double precision fused multiply adds (FMAs) per clock. The Vulcan cores were intended to run at speeds in excess of 3 GHz.

To make the original Vulcan SoC, Broadcom was planning to gang up the cores and segments of a distributed L3 cache using a coherency fabric that had “multi-terabit” capacity. This coherent fabric also linked out to DDR4 memory controllers on the die as well as to the Inter-Chip Interconnect (ICI), which appears to be Broadcom’s own implementation of NUMA, derived from the NLP network processor family. The Vulcan SoC had a switch hanging off this coherent fabric, which linked on the other side to a multi-terabit non-coherent fabric that in turn had various I/O and networking coprocessors and accelerators as well as system management controllers on that fabric. The Vulcan chip also sported full virtualization of the cores, memory, I/O, and accelerators.

The architecture was similar in many ways to the Xeons of the time, with a much wider table lookaside buffer (TLB) for the L2 cache. The idea was to make the highest performing ARMv8 core and get to 90 percent of the performance level of a Haswell Xeon core and then have more cores than the middle-bin parts that represent the best bang for the buck and the bulk of the shipments in the Xeon line.

Não sei se o processador actual ainda terá suporte para ARM 32 bit e duas unidades NEON. O primeiro porque não vai ser usado, o segundo porque NEON é mais pensado para o espaço embedded. Alias, a ARM anunciou as instruções SVE para o mercado HPC e não só.
Outro ponto interessante é ele originalmente suportar SMP4. Num sistema com dois sockets, daria 256 threads de forma concorrencial. Não sei se ainda terá esse suporte.

Slide apresentado na SC17:

sYTCtIr.jpg


Shah did not divulge the sizes of the L1, L2, and L3 caches on the Vulcan ThunderX2 and said that they would be revealed at some future time. The exact nature of the floating point was also not discussed, but we don’t think any of these have changed since Cavium took over the Vulcan project. What we can see is that the Vulcan die tops out at 32 cores, with variants with 24 cores and 28 cores activated, which is holding toe-to-toe on the core count with the new Skylake Xeon SP processors from Intel. The Vulcan ThunderX2 has eight memory controllers, two more than the six of the Xeon SP, and delivering more bandwidth to the processors. The DDR4 memory controllers on the Vulcan chip have two DIMMs per controller, for a total of 16 memory sticks per socket, and at the moment, the NUMA scalability is limited to two sockets and 32 memory sticks. In theory, that can mean 2 TB of memory per socket, using very expensive 128 GB sticks.

Benchmark Stream, onde ele ganha com facilidade a um Xeon Gold, devido a ter mais controladores de memória:

KfHa9Mj.jpg


Benchmark SPECcpu, usando GCC e onde o Xeon Gold usa o ICC e o GCC:

cavium-vulcan-thunderx2-skylake-xeon-spec.jpg


As you can see, the 32-core Vulcan ThunderX2 is within spitting distance of the 20-core Skylake when using the GNU compilers, and with Arm and Cavium working on tuning up the Allinea compilers, they expect to be able to goose performance by another 15 percent or so further on these two SPEC CPU 2017 tests.

Esperam um avanço de 15% na performance, quando os compiladores da Allinea.

Comparação em alguns benchmarks com ele a 2.2 Ghz e usando GCC contra o Xeon Gold a 2.4 Ghz e usando o ICC:

MthJp1F.jpg


De referir que este processador da Intel custa 3 mil $ e tem um TDP de 150W. Não há dados ainda deste Cavium.

https://www.nextplatform.com/2017/11/27/cavium-truly-contender-one-two-arm-server-punch/

Imagens do servidor HP Apollo 70 que usa este cpu Cavium:

f6KuE3n.png


ArF0jRr.jpg


https://www.nextplatform.com/2017/11/22/hpe-aims-hpc-servers-storage-enterprise/

q5VTRVu.jpg


ghw13h4.jpg


http://www.lemagit.fr/actualites/45...ent-dun-serveur-Proliant-ARM-Cavium-ThunderX2

Video deste servidor na fase de testes:


Isto é um chassis 2U com 4 servidores dual processor. Isto é, 256 cores em 2U. Pode levar Gpus FirePro s9170 ou dois discos 2.5 na parte de trás em cada servidor. Não falam de outros fabricantes de Gpus. No interior tem dois slots M.2 e 3 discos 3.5 por cada servidor.

Algo interessante é que a Argonne anunciou que ia comprar um Cluster com 32 nós para testes com ARM. Quem é a Argonne? A mesma instituição que tinha previsto um supercomputador feito pela Cray e Intel de 180 petaflops em 2018 e que foi adiado para 2021 com 1 Exaflop. Parece-me um aviso.
32 nós cabem num bastidor, por isso é coisa ainda pequena.

Argonne National Laboratory is collaborating with HPE to provide system software expertise and a development ecosystem for a future high-performance computing system based on 64-bit ARM processors.

To support this work, Argonne will install a 32-node Comanche Wave prototype ARM64 server platform in its testing and evaluation environment, the Joint Laboratory for System Evaluation, in early 2018. Argonne researchers from various computing divisions will run applications on the ecosystem and provide performance feedback to HPE and partnering vendors.

https://insidehpc.com/2017/11/argonne-install-comanche-system-explore-arm-technology-hpc/

Ao mesmo tempo foi anunciado um supercomputador com este processador da Cavium, feito pela Cray, para o Met Office Inglês. Chama-se "Isambard"

“This is an exciting time in high performance computing,” says Prof Simon McIntosh-Smith, leader of the project and Professor of High Performance Computing at the University of Bristol. “Scientists have a growing choice of potential computer architectures to choose from, including new 64-bit ARM CPUs, graphics processors, and many-core CPUs from Intel. Choosing the best architecture for an application can be a difficult task, so the new Isambard GW4 Tier 2 HPC service aims to provide access to a wide range of the most promising emerging architectures, all using the same software stack. Isambard is a unique system that will enable direct ‘apples-to-apples’ comparisons across architectures, thus enabling UK scientists to better understand which architecture best suits their application.”

http://gw4.ac.uk/isambard/

Também foi mostrado um servidor do "Project Olympus" da Microsoft em formato OCP:

yFBBNPr.jpg


Uz551h1g.jpg


7lv7D1W.jpg


Vb1Tz5l.jpg


oydXebo.jpg


f4rDV4z.jpg


A plataforma suporta processadores até 195W, mas não se sabe se é o TDP deles.

https://www.servethehome.com/cavium-thunderx2-ocp-platform-details/


Eu sei que isto é uma thread de ARM para servidores, mas esta noticia é capaz de interessar o pessoal do Fórum.

Apareceu um score no Geekbench de um Asus TP370QL, com um Qualcomm 835 a correr Windows 10. Supostamente, é um Portátil de 13 polegads. O score é este:

piY4E5R.png


O resultado é miserável, mas como este benchmark, em Windows, só existe para x86, ele está a correr x86 usando binary translation, onde a performance vai sempre sofrer impacto.

http://winfuture.de/news,100673.html
 
Á partida a Marvell poderá estar interessada, uma vez que estes chips também devem dar para servidores SVN e infraestruturas de rede, especialmente agora com o investimento em 5G, e à partida a parte do Storage, dependendo de isto não estar ou não ir consumir muito mais recursos.

A Broadcomm actualmente assenta o seu negócio essencialmente em modelos de licenciamento, no qual os chips para servidores não se enquadram, pois ainda estão na fase de consumirem demasiado em I&D.
 
- Deep Dive Into Qualcomm’s Centriq Arm Server Ecosystem
https://www.nextplatform.com/2017/12/06/deep-dive-qualcomms-centriq-arm-server-ecosystem/

- Battle For Datacenter Compute: Qualcomm Centriq Versus Intel Xeon
For today’s episode of Battle for Datacenter Compute, we are examining how Qualcomm and one of its server application partners, Cloudflare, have stacked up the new Centriq 2400s against the past two generations of Xeon chips from Intel.

In these tests, Cloudflare tested a single-socket server using the Centriq 2452 with 46 cores running at 2.5 GHz against a two-socket Broadwell Xeon E5-2630 v4 with ten cores running at 2.2 GHz per socket and another two-socket Skylake Xeon SP-4116 Silver with twelve cores running at 2.1 GHz. The Xeons have HyperThreading simultaneous multithreading turned on, so that is 40 Broadwell threads versus 48 Skylake threads versus 46 Amberwing threads. The two Broadwell chips cost $1,334, the two Skylake chips cost $2,004 together, and the single Amberwing chip costs $1,383. They are in the same ballpark on clock speeds, thread count, and price. Krasnov correctly presented the performance metrics per core and per system, so we can see the differences in these two aspects of the systems immediately.
https://www.nextplatform.com/2017/1...r-compute-qualcomm-centriq-versus-intel-xeon/
 
Foi apresentado o AppliedMicro X-Gene3 e em parte o X-Gene3XL

APM-X-Gene-3-block-diagram.png


O X-Gene3 é um processador de um só socket com 32 cores, 8 canais de memória, 32 MB de L3 e 42 lanes Pci-Express3. Não tem aceleradores integrados nem Lans. Em simulações parece ser competitivo no SpecInt com o Intel Xeon E5-2680 v4. A questão é que o X-Gene3 só sai no segundo semestre de 2017 e nessa altura a Intel terá no mercado o Skylake-EP e a AMD o Zen.

O X-Gene3XL sairá depois e tem 64 cores e suporte para 2 sockets.

Source:
http://semiaccurate.com/2016/04/25/appliedmicros-x-gene-3-aims-for-intels-e5-xeons/
http://www.linleygroup.com/cms_builder/uploads/x-gene-3-white-paper-final.pdf

É interessante. Não sei é se não virá demasiado tarde.

Entretanto a Applied Micro foi adquirida pela MACOM e a sua divisão de Computação vendida a um Fundo de investimento liderado pelo Carlyle Group, e agora foi anunciado que passou a designar-se Ampere Computing e que é liderada por, nada mais nada menos, que a ex-Presidente do CA da Intel, Renée James

Startup to relaunch AMCC's X-Gene 3 as its first product
the ARM-based server SoC Ampere is pitching today was originally designed by Applied Micro Circuits Corp. (AMCC) back in 2015. Known as the X-Gene 3, it was AMCC’s third-generation ARM-based server SoC, running at 3 GHz, made in a 16nm FinFET process at TSMC.
Meanwhile, after leaving Intel in the summer of 2015, James joined Carlyle in early 2016 as an operating executive. The Carlyle Group rebranded the X-Gene ARM server SoC business as Ampere. James became CEO last fall.
https://www.eetimes.com/document.asp?_mc=RSS_EET_EDT&doc_id=1332934&page_number=1
 
Private Equity Amps Up Arm Servers With Applied X86 Techies

The team that Ampere has put together has some heavy hitters on it, starting with Renee James, who was president at Intel and was passed over for the chief executive officer position at the chip giant for Brian Krzanich
Chi Miller, a long time finance specialist at Intel and formerly the director of finance at Apple, has joined Ampere as its chief operating officer and chief financial officer... Rohit Vidwans, who spent 26 years at Intel and lead develop of many generations of Atom and Xeon processors, has joined Ampere as executive vice president of hardware engineering, and Atiq Bajwa, a 30 year veteran of Intel who was the head of X86 architecture, is the chief architecture for the Arm upstart. And Greg Favor, who was a fellow at AMD and part of the K6 and K7 development teams and who was the lead architect at Applied Micro for its X Gene processors, is a senior fellow at Ampere.
https://www.nextplatform.com/2018/02/06/private-equity-amps-arm-servers-applied-x86-techies/
 
Benchmarking An ARM 96-Core Cavium ThunderX System

A Phoronix reader granted us remote access to a FOXCONN C2U4N_MB system featuring two Cavium ThunderX 48-core SoCs. For those curious about the potential of a modern 96-core ARM platform, here are some basic benchmark results.
...
The Cavium ThunderX configuration at 96 cores with the FOXCONN C2U4N_MB motherboard also had 4 x 32GB DDR4-2133MHz memory, 250GB Samsung 850 SSD, and was running Ubuntu 16.04 with the Linux 4.10 kernel and GCC 5.4 compiler.
https://www.phoronix.com/scan.php?page=article&item=cavium-thunderx-96core&num=1
 
Isto não é bem um servidor. :D Mas acho que é interessante na mesma. :)

qBXn52R.jpg


zpGsUQO.jpg


Gigabyte ThunderXStation using Cavium ThunderX2 Launched

Gigabyte ThunderXStation Tech Specs
Here are the key specs we received on the workstation/ server:

  • Form Factor: 4U Tower
  • CPU: Single or Dual socket ThunderX2 ARM64 Processors
  • Memory: Up to 16 DDR4 Channels
  • PCIe: 6 x PCIe 3.0 Slots and 2 x OCP x16 slots
  • Network: 2 x 1/10 GbE QLogic NIC
  • Graphics: Nvidia GeForce® GT710 with dual monitor support
  • BMC: ASPEED AST2500 with IPMI management SW
  • Storage: 4 x NVMe + 2 x 2.5″ U.2/SATA III combo bay
  • Firmware: AMI Aptio V ® UEFI and AMI MegaRAC™ BMC
  • OS/Tools: CentOS 7.4 with gcc/LLVM/gdb tool chain, OpenJDK 9.0
The actual motherboard has a few interesting specs beyond those mentioned above:

  • Single or dual CPU operation
  • Integrated chipset
  • 8x DIMMs per CPU using DDR4 2666/2400/2133 MHz DIMMs
  • ASPEED AST2500 BMC with:
    • 1x Management LAN 10/100/1000 Mbps
    • Integrated VGA, 1920×1200
  • 3x PCIe Gen 3.0 slots per CPU which an be configured as (1x 16 lane + 1x 8 lane) OR (3x 8 lanes)
  • 1x Type-1 OCP PCIe Gen 3.0 x16 slot per CPU
  • 2x NVMe PCIe Gen 3.0 x4 ports with onboard SAS slimline connectors per CPU
  • 2x SATA III onboard connectors
  • 2x U.2 or SATA III 2.5″ drive bay, supports any combination of NVMe and SATA III SSDs
  • 1x RS-232 port
  • 4x USB 3.0 ports
  • 2x 8-pin GPU power connectors

https://www.servethehome.com/gigabyte-thunderxstation-using-cavium-thunderx2-launched/

64 cores / 256 threads, logo faz SMT4. E este é o ThunderX2 antigo "Broadcom". Os cores não são pequenos. Nice..... :)
 
Eu sei que isto é uma thread de ARM em servidores, mas visto que isto é uma noticia importante, coloco aqui:

Apple Plans to Use Its Own Chips in Macs From 2020, Replacing Intel

Apple Inc. is planning to use its own chips in Mac computers beginning as early as 2020, replacing processors from Intel Corp., according to people familiar with the plans.

The initiative, code named Kalamata, is still in the early developmental stages, but comes as part of a larger strategy to make all of Apple’s devices -- including Macs, iPhones, and iPads -- work more similarly and seamlessly together, said the people, who asked not to be identified discussing private information. The project, which executives have approved, will likely result in a multi-step transition.

The shift would be a blow to Intel, whose partnership helped revive Apple’s Mac success and linked the chipmaker to one of the leading brands in electronics. Apple provides Intel with about 5 percent of its annual revenue, according to Bloomberg supply chainanalysis.

Intel shares dropped as much as 9.2 percent, the biggest intraday drop in more than two years, on the news. They were down 6.4 percent at $48.75 at 3:30 p.m. in New York.

Apple could still theoretically abandon or delay the switch. The company declined to comment. Intel said, “We don’t comment on speculation about our customers.”

For Apple, the change would be a defining moment. Intel chips remain some of the only major processor components designed by others inside Apple’s product portfolio. Currently, all iPhones, iPads, Apple Watches, and Apple TVs use main processors designed by Apple and based on technology from Arm Holdings Plc. Moving to its own chips inside Macs would let Apple release new models on its own timelines, instead of relying on Intel’s processor roadmap.

"We think that Apple is looking at ways to further integrate their hardware and software platforms, and they’ve clearly made some moves in this space, trying to integrate iOS and macOS,” said Shannon Cross an analyst at Cross Research. “It makes sense that they’re going in this direction. If you look at incremental R&D spend, it’s gone into ways to try to vertically integrate their components so they can add more functionality for competitive differentiation."

Stand Out
The shift would also allow Cupertino, California-based Apple to more quickly bring new features to all of its products and stand out from the competition. Using its own main chips would make Apple the only major PC maker to use its own processors. Dell Technologies Inc., HP Inc., Lenovo Group Ltd., and Asustek Computer Inc. use Intel chips.

By using its own chips, Apple would be able to more tightly integrate new hardware and software, potentially resulting in systems with better battery life -- similar to iPads, which use Apple chips.

While the transition to Apple chips in hardware is planned to begin as early as 2020, the changes to the software side will begin even before that. Apple’s iPhones and iPads with custom chips use the iOS operating system, while Mac computers with Intel chips run on a different system called macOS. Apple has slowly been integrating user-facing features over the past several years, and more recently starting sharing lower-level features like a new file management system.

‘Marzipan’ Platform
As part of the larger initiative to make Macs work more like iPhones, Apple is working on a new software platform, internally dubbed Marzipan, for release as early as this year that would allow users to run iPhone and iPad apps on Macs, Bloomberg News reported last year.

The company has also previously released Macs with ARM-based co-processors, which run an iOS-like operating system, for specific functions like security. The latest MacBook Pro and iMac Pro include the co-processors. Apple plans to add that chip to a new version of its Mac Pro, to be released by next year, and new Mac laptops this year, according to a person familiar with the matter.

Intel has dominated computing processors for more than a decade, taking market share from Advanced Micro Devices Inc., its only rival in the market. Intel also designs and builds modem chips for some iPhone models so that they can connect to cellular networks and make calls. While Apple is down the list of computer sellers by unit shipments, it’s third when measured by revenue last year, highlighting the premium status of its products.

Apple’s decision to switch away from Intel in PC’s wouldn’t have a major impact on the chipmaker’s earnings because sales to the iPhone maker only constitute a small amount of its total, said Kevin Cassidy, an analyst at Stifel Nicolaus & Co. A bigger concern would be if this represents part of a wider trend of big customers moving to designing their own components, he said.

In 2005, Apple announced a move to Intel chips in its Macs, an initiative that put former Intel Chief Executive Officer Paul Ottelini on stage with Apple co-founder Steve Jobs. It was a partnership that shook up the PC industry and saw Apple shift away from chips co-developed by IBM and Motorola.

Apple’s current chip designs made their name in thin and light mobile devices. That would indicate Apple will start the transition with laptops before moving the designs into more demanding desktop models. Apple has to walk the fine line of moving away from Intel chips without sacrificing the speed and capabilities of its Macs.

A decision to go with ARM technology in computers might lend it credibility where it has failed to gain a foothold so far. Qualcomm Inc., the biggest mobile phone chip provider, is working with PC makers to introduce new thin and light laptops based on its chips in another attempt to steal share from Intel. Microsoft Corp. is supporting that effort by providing a version of its Windows operating system for ARM technology-based chips.

Intel’s dominance of the market has been based on its ability to use leading manufacturing technology to produce processors that are more powerful than those of its competitors. Its would-be rivals haven’t yet produced designs that have displaced Intel’s products when it comes to crunching data quickly.

Apple’s custom processors have been recently manufactured principally by Taiwan Semiconductor Manufacturing Ltd.
It’s decision may signal confidence that TSMC and other suppliers such as Samsung Electronics Co. have closed the gap on Intel’s manufacturing lead and can produce processors that are just as powerful.

https://www.bloomberg.com/news/arti...an-move-from-intel-to-own-mac-chips-from-2020

Apple vai mudar de Intel para ARM, possivelmente antes de 2020. O projecto tem o nome de código "Kalamata".
Vai ser possível correr aplicações Ios em MacOs, talvez ainda este ano. O projecto tem o nome de código "Marzipan".

A Apple "só" representa 5% do volume de vendas da Intel, mas o problema não é esse. O problema é que a Apple é vista como um "Market Leader" que faz com que outras empresas sigam os seus passos. As ações da Intel estiveram a cair 9% com esta notícia.


Visto que acabaram de sair Portáteis e 2 2 em 1 com Windows para ARM, não querem renomear esta thread para ARM for Computers? :)
 
E porque não abrir um tópico? Parece ser algo que a médio prazo terá pernas para andar, por isso deve dar para manter um tópico próprio não?
 
HPE, Arm, SUSE, and three leading UK universities establish one of the largest Arm-based supercomputer deployments in the world to advance digitisation of UK economy

The three supercomputer clusters at EPCC, University of Bristol and University of Leicester will in total run more than 12,000 Arm-based cores, hosted by HPE Apollo 70 HPC systems. The clusters at each university will be largely identical, consisting of 64 HPE Apollo 70 systems, each equipped with two 32 core Cavium ThunderX2 processors, 128GB of memory composed of 16 DDR4 DIMMs with Mellanox InfiniBand interconnects. The operating system is SUSE Linux Enterprise Server for HPC. Each cluster is expected to occupy two computer racks and consume a total of approximately 30KW of power.
https://news.hpe.com/academia-and-industry-collaborate-to-drive-uk-supercomputer-adoption/
 

Esse PDF parece bem interessante. Ainda só o vi na diagonal. :)

Noutras noticias, o ThunderX2 parece estar disponível para toda a gente no mercado a partir de hoje.

B8CZl25.jpg


Uma imagem do htop. :D São 2 sockets X 32 cores X 4 threads (SMT4).

Mas a parte mais interessante, para mim, é esta:

We go into the history of Cavium ThunderX2 in our full review, but we can say this, the new generation is on par with AMD EPYC 7000 series and Intel Xeon Scalable in terms of performance while topping out at a $1795 price tag.

Se a performance andar à volta dos de topo da AMD e Intel, no Epyc 7601 e nos Xeons SP Gold, este preço é mesmo muito bom.

Cavium ThunderX2 Key Specs
Here are the key specs from the press release:

  • Single chip system on a chip (SoC) server CPU
  • Core and socket level performance comparable to highest end Xeon Skylake Platinum CPUs
  • Second generation of full custom Cavium Arm core
    • Quad Issue, Fully Out of Order
    • Full SMT support – 1, 2, 4 threads per core
    • Up to 2.5 GHz in normal mode, up to 3 GHz in Turbo mode
    • 3X single thread performance compared to ThunderX®
  • Up to 32 cores per socket delivering > 2.5-3X socket level performance compared to ThunderX
  • Cache:
    • 32 KB L1 instruction and data cache, 256KB L2 per core
    • 32 MB distributed L3 cache
  • Advanced server class RAS features covering memory, CPU, cache, CCPI2 and PCIe interfaces
  • Advanced power management
    • On-chip management engine for dynamic voltage and frequency scaling across the chip
    • Full Turbo mode support
  • Single and dual socket configuration support using 2nd generation of Cavium Coherent Interconnect with > 2.5X coherent bandwidth compared to ThunderX
  • System Memory
    • 8 DDR4 memory controllers per socket
    • Dual DIMM per memory controller, for a total of 16 DIMMs per socket
    • Up to 4 TB of memory in dual socket configuration
    • 33% higher memory bandwidth and memory capacity compared to Xeon Skylake Platinum CPUs
  • Flexible IO:
    • Integrated 56 lanes of PCIe Gen3 interfaces, x1, x4, x8 and x16 support, 14 integrated PCIe controllers
    • Integrated SATAv3, GPIOs, USB interfaces
    • 16% higher IO bandwidth compared to Xeon Skylake Platinum CPU

https://www.servethehome.com/cavium-thunderx2-hits-general-availability/

Parece que eles têm uma review dele pronta a entrar online e só estão à espera do OK da Cavium. :)
 
Back
Topo