AMD low-power x86 core "Jaguar"

[Dark Kaeser]

NOTA: ainda sob NDA por isso é uma leak, preview

AMD launched its low-cost AM1 platform last month and we must note that it was a rather unusual launch. Since it's AMD's entry level socketed platform, there was no hype and oddly enough AMD's NDA is still in effect, despite the fact that AM1 products have been available for weeks.

Luckily we managed to get a bit more info from an early AM1 adopter and Fudzilla regular Andrea Sibaldi. Sibaldi decided to pick up an Athlon 5350 and an Asrock AM1B-ITX board, both of which are available in Italian retail.

http://fudzilla.com/home/item/34390-heres-our-first-crowdsourced-amd-am1-review

De relembrar que apesar da designação Athlon, se trata de um APU "Kabini" versão socket (AM1 ou FS1b).

 

[Dark Kaeser]

reviews do Athlon 5350

http://anandtech.com/show/7933/the-desktop-kabini-review-part-1-athlon-5350-am1

http://www.pcper.com/reviews/Processors/AMD-AM1-Athlon-5350-Reviewed-Low-Power-Low-Price-APU-Masses

 

[||Red_Evil||]

isto para um media center parece-me mesmo muito bem!

 

[Dark Kaeser]

AMD Athlon 5350 / 5150 & Sempron 3850 / 2650

here's the initial Ubuntu Linux benchmarks of all four AMD AM1 APUs that are available at this time: the Sempron 2650, Sempron 3850, Athlon 5150, and Athlon 5350.
...

In this article are the results from these four APUs when running Ubuntu 14.04 LTS but with the manual upgrade to the Linux 3.14 kernel and Mesa 10.2 RadeonSI graphics stack as outlined in the earlier article.

http://www.phoronix.com/scan.php?page=article&item=amd_athlon_sempron&num=1

Em breve o Michael espera fazer um comparativo entre os drivers open source (Radeon SI) e os proprietários (Catalyst), se bem que os comparativos que o Michael publica no Phoronix acabam por ter sempre o problema da escolha dos "bench" escolhidos.


DDR3 Memory Scaling Performance With AMD's Athlon 5350

For especially the Radeon R3 Graphics performance but also in other workloads with frequent memory accesses, there's much greater performance potential in going with DDR3-1600, which is the highest-rated system memory frequency supported by the Athlon 5350/5150 and Sempron 3850 while the Sempron 2650 dual-core APU is capped to DDR3-1333MHz.

http://www.phoronix.com/scan.php?page=article&item=amd_am1_ddr3&num=1

 

[blaster_00]

O CPU não chegou a 30º durante os benches... isso está mesmo a pedir passive cooling ou só uma ventoinha de 12cm para o sistema todo.

 

[Dark Kaeser]

Overclocking The AMD AM1 Athlon & Sempron APUs

AMD Sempron 2650 - The dual-core APU clocked at 1.45GHz by default (100 x 14.5) could be bumped to 1.50GHz with the maximum multiplier value of 15. Increasing the APU frequency to 105MHz yielded a 1.575GHz clock speed for the two AMD Jaguar cores. Bumping the APU frequency also led the system memory to running at DDR3-1399MHz (the Sempron 2650 is capped to DDR3-1333MHz over DDR3-1600MHz with the higher-end AM1 APUs.)


AMD Sempron 3850 - With not being able to increase the multiplier any higher, the quad-core 1.30GHz APU could just ramp up to 1.365GHz when increasing the CPU ratio to 105MHz and the AMD DDR3 system memory was then running at DDR3-1680MHz. Pushing past a 105MHz ratio -- as with the other APUs -- led to stability issues.


AMD Athlon 5150 - The multiplier on the Athlon 5150, which stock runs at 1.60GHz for the quad-core APU, also couldn't be increased and thus when going with the 105MHz APU frequency led to a 1.68GHz frequency for the APU and system memory.


AMD Athlon 5350 - Lastly, for the highest-end AM1 APU right now, the Athlon 5350 (2.05GHz quad-core), its ratio could be bumped from 20.5 to 21. With that bump plus a 105MHz frequency yielded a clock speed of 2.205GHz and DDR3-1680MHz for the system memory.

The Athlon 5350 was about 7% faster with OpenSSL with its light overclocking, which is similar to the 5~8% gains out of the other AM1 APUs.

http://www.phoronix.com/scan.php?page=article&item=amd_am1_overclocking&num=1



AMD Athlon's R3 Graphics: RadeonSI Gallium3D vs. Catalyst

Building on the results from last week that compared the out-of-the-box Ubuntu 14.04 performance to that of manually upgrading the key graphics driver components, the tests for this article included:

- Ubuntu 14.04 in its out-of-the-box configuration with the Linux 3.13 kernel and Mesa 10.1.

- The same Ubuntu 14.04 installation but then upgrading to the Linux 3.14 kernel where there's proper dynamic power management (DPM) support enabled for the Radeon R3 Graphics so they can re-clock to their highest performance state.

- The above installation but besides upgrading to the Linux 3.14 kernel also using Mesa 10.2-devel and xf86-video-ati 7.3.99 using the Oibaf PPA. Besides the new Mesa and Git, one item to point out is that ColorTiling and ColorTiling2D are both enabled by default at this stage. However, HyperZ isn't yet enabled by default for the RadeonSI hardware, but with many of these tests being done at lower resolutions, it really wouldn't benefit these particular test-cases too much.

- The Ubuntu 14.04 LTS configuration on the same hardware but switching from the open-source driver to closed-source. The fglrx package was installed from the Ubuntu Trusty archive to provide the fglrx 13.35.5 / OpenGL 4.3.12798 driver.

http://www.phoronix.com/scan.php?page=article&item=amd_am1_catalyst&num=1

Os closed source driver Catalyst continuam a apresentar melhor resultados que os open source.


mais uns testes em Linux

- AMD Sempron 2650 Benchmarks and Review – Dual Core AM1 SoC

- AMD Sempron 3850 Benchmarks and Review – Quad Core AM1 SoC

- AMD Athlon 5150 Benchmarks and Review – AM1 APU

- AMD Athlon 5350 Linux Benchmarks and Review

 

[Dark Kaeser]

Mais uns teste do Michael do Phoronix.


- 13-Way Low-End GPU Comparison With AMD's AM1 Athlon

Overall, when running the brand new Ubuntu 14.04 release and using the binary AMD/NVIDIA drivers, the Radeon R3 Graphics (also branded as the Radeon HD 8400 series) its performance was close to that of the discrete Radeon HD 6450. The Radeon HD 6450 PCI-E graphics card still retails for $40~50 USD so considering that a Athlon 5350 APU can be found for about $60 or less, that's a pretty good deal for the GCN-based graphics paired with the four low-power Jaguar cores.

Compared to the NVIDIA hardware and their Linux driver, the Athlon 5350 was competing against most prominently the GeForce GT 220 and generally ahead of the GT 610 and GT 520.

http://www.phoronix.com/scan.php?page=article&item=amd_am1_discrete&num=1


E fica aqui um teste interessante da influência que a quantidade de memória alocada tem na performance do iGPU

- How Much Video RAM Is Needed For Catalyst R3 Graphics?

For those curious or unsure how much video memory should be allocated when using the Catalyst Linux graphics driver with an AMD APU, hopefully this data was of some value. With 512MB or greater, most Linux games at 1920 x 1080 should run just fine and not have any detrimental performance issues in coming under video memory pressure. Fortunately, most AM1 motherboards default to 512MB of video memory, so by default you should be in fairly good shape.

http://www.phoronix.com/scan.php?page=article&item=amd_kabini_vram&num=1

Devo dizer que estou a ficar surpreendido pela quantidade de testes que o Michael tem feito com estes SoC da AMD, até agora tem sido dos mais completos.

 

[uwannacookie]

Já vi videos de Gaming (BF3) no quad-core 5350, e o desempenho é abismal, apesar de 4x. A jogar a 1080p em médio com uma HD6850 consegue-se ~60FPS.

As manufacturadoras deviam pensar em fazer boards com dual socket kabini, e fazer uma PCI-E 8x, nesse caso. É um bocado ridiculo, mas gostava que existisse. Ter 8 núcleos a 2.05Ghz, 8GB de ram, e adicionar uma R9 270X, gostava de ver os resultados. 8 nucleos com TDP máximo de 50w. Isto tudo, num formato micro ATX.

 

[ObscureAngel]

uwannacookie, performance abismal no sentido de bom ou mau?

 

[uwannacookie]

Para o custo do CPU, conseguires jogar BF3 em high numa gráfica de média gama a fps superiores a 30, acho excelente!

 

[ObscureAngel]

Epá eu não sei..
Isto porque tive a ver reviews e parece que coisas com o cinebench e tudo os resultados são ligeiramente inferiores a um Pentium G (haswell) que também tem o mesmo preço.
A termos de gaming, não sei porque nunca testei com a HD 7850 esse tal Pentium G, mas tenho um cá em casa para os meus pais.

 

[Dark Kaeser]

Ainda não morreu!?

Mysterious AMD A9-9820 APU with Radeon R7 350 iGPU spotted in Chuwi's latest AeroBox mini PC

The unannounced A9-9820 8-core/8-thread APU appears to be part of AMD’s ever-expanding mini PC processor portfolio introduced in late 2019, and it might only be available in Japan for the moment, as it will be powering a Chuwi AeroBox mini PC featured on the OEM’s Japanese site. There is also an integrated Radeon R7 350 GPU that is based on the older Polaris architecture, possibly a refreshed Oland model.

https://www.notebookcheck.net/Myste...-Chuwi-s-latest-AeroBox-mini-PC.457390.0.html

Ah e usa ainda DDR3...

 

[miguelbazil]

Chuwi... Há empresas que aproveitam mesmo tudo. Às tantas, o processo de fabrico e falta de procura tornou isto tão barato, que vão tentar fazer dinheiro com algo estupidamente barato.

 

[Dark Kaeser]

Mas o é estranho é onde eles vão arranjar 8c Jaguar, uma vez que supostamos a AMD apenas vendia SoC quadcore.

 

[Nemesis11]

Eu ando um bocado fora do mundo de jogos e consolas, mas aquela caixa não é muito parecida ou mesmo igual a uma das Xboxs?

É que se sim, faz sentido ter "Cat" Cores.

Outra coisa, no mercado PC, nunca foram vendidos APUs com 8 "Cat" Cores, certo? Nem com Polaris, certo?
Isto será um dos APUs de consolas convertido para o mercado de PCs? É que não estou a ver a AMD a fazer novos APUs com "Cat" Cores, nesta altura do campeonato.

 

[SideWalker]

Aquilo é mesmo Polaris? "R7 350" é nomenclatura de GCN, ainda. E batia +/- certo com o GPU da xbox one original.

 

[Dark Kaeser]

Sim os cat cores só tiveram 4, a exceção era mesmo as consolas, mas se bem me recordo o IP dos SoC das consolas pertence às respectivas empresas, Sony e Microsoft.
A não ser que houvesse outra coisa que me esteja escapar .

 

[Nemesis11]

Sim os cat cores só tiveram 4, a exceção era mesmo as consolas, mas se bem me recordo o IP dos SoC das consolas pertence às respectivas empresas, Sony e Microsoft.
A não ser que houvesse outra coisa que me esteja escapar .

Eu não quero acreditar que a AMD vá fazer novos processadores com Cat Cores. Pelo menos com os Cat Cores existentes. Não faz grande sentido. Até para os Chromebooks eles usaram Bulldozers com diferentes clocks.
Poderá ser um CPUs das consolas com um GPU externo?

 

[Dark Kaeser]

Boa pergunta, não me recordo se foi público algum detalhe dos SoC das consolas, mas não me parece que fosse monolítico, devia ter algum tipo de interposer, se não de silício algo mais simples de substrato.

 

[SideWalker]

São monolíticos.
Mas o da Xbox one tem um monte de esram e o da ps4 usa gddr5 e não ddr3.
Ou seja, o exacto soc de uma das consolas não deve ser. Mas um derivado custa a crer que compense sequer o trabalho de o fazer... Muito estranho, esse chip.