Gráfica Imagination Announces B-Series GPU IP: Scaling up with Multi-GPU

[muddymind]

Novos IPs de GPU da Imagination que ainda se tenta mexer lol:

https://www.anandtech.com/show/16155/imagination-announces-bseries-gpu-ip-scaling-up-with-multigpu

E agora a melhor parte: GPU discreto para desktop e servers!

https://www.innosilicon.com/html/news/43.html

 

[Dark Kaeser]

Aparentemente uma evolução das A-Series introduzidas no final de 2019.
A mesma estratégia de "tiles" também.

Mas o topo passou de 2TFlops para 8TFlops.

 

[Torak]

Wavefronts de 128.
Não serão um pouco grandes, mesmo para desktop....

 

[miguelbazil]

Eles estão a mencionar cloud computing, se calhar isto não é nada direccionado a desktop. Mais para server.

 

[Dark Kaeser]

Texas Instruments integrates Imagination GPU into Jacinto processor

TI will use the BXS-4-64 GPU core in its next generation of Jacinto system-on-chip designs. The latest Jacinto 7 family, launched in January and starting to ship now, includes the TDA4VM and DRA829V SoCs for advanced ADAS and automotive gateway systems. These have server performance in a power envelope of 5 to 20W.

https://www.eenewsautomotive.com/news/texas-instruments-integrates-imagination-gpu-jacinto-processor

O BXS parece ser apenas uma variante do BXM, específica para a certificação necessária para o mercado ADAS.

 

[Dark Kaeser]

Imagination, update on 2021 progress​

Imagination Technologies today announces initial unaudited results for the first half of 2021 (H1/2021), where overall revenues increased by 55% to $76m (H1/2020: $49m).

In the first half Imagination signed over 35 licensees across segments including automotive, datacentre and desktop, mobile, and DTV. Imagination has a 37% share in smartphone graphics processing unit (GPU) (source: TSR) and is the largest GPU supplier to the automotive industry with approx. 45% share and over a quarter of a billion automotive units shipped (source: company data).

The company expects to launch a further GPU family, C-Series, in Q4/2021, maintaining a rapid pace of technology development and roll-out.

This year Imagination is re-entering the CPU market with designs based around the RISC-V open ISA. Imagination’s heritage in CPU enables it to provide innovative and patent protected technologies for the discrete CPU market as well as addressing demand for heterogeneous solutions that combine GPU, CPU and AI processors. This strategy will enable further growth.

https://www.design-reuse.com/news/50482/imagination-2021-progress.html

 

[Dark Kaeser]

Não é para o mercado PC, mas Server

Chinese Xindong Fenghua GPU with GDDR6X Takes on AMD and Nvidia​

The company recently successfully tested its first Fenghua No. 1 graphics processing unit developed entirely in-house and then designed by Innosilicon, a contract chip designer.

The Fenghua No. 1 seems to be a multi-chiplet GPU with a GDDR6X-based memory subsystem that supports all modern (and legacy) application programming interfaces for graphics and compute, including DirectX, Vulkan, OpenGL, OpenCL, and even OpenGL ES. It can also work with Android, Linux, and Windows.

https://www.tomshardware.com/news/chinese-xindong-fenghua-gpu-announced

 

[muddymind]

Olha, já suportam mais APIs

 

[Dark Kaeser]

This Chinese company made a graphics card that supports Vulkan and DirectX​

Interestingly, the Fenghua 1 is based on PowerVR IP from Imagination Technologies. It also supports APIs like OpenGL, Vulkan, OpenGL ES, OpenCL, and DirectX, which is a first for any Chinese GPU. And since it’s also aimed at the consumer market, it comes with an HDMI 2.1 port, a DisplayPort 1.4, and even a DVI port.

https://www.techspot.com/news/92303-chinese-company-made-graphics-card-supports-vulkan-directx.html


De resto há quem se tenha lembrado de um post com quase um ano, da própria Imagintation aquando do lançamento da série B

One of them, Innosilicon, has announced their plans today alongside the B-Series launch. Innosilicon has integrated the newly launched IMG B-Series BXT multi-core GPU IP in PCI-E form-factor GPU SoCs for desktop and data centre applications. Our companies are also exploring a long-term strategic partnership to bring even more powerful GPU SoCs to the market.

https://www.imaginationtech.com/blog/back-in-the-high-performance-game/

 

[muddymind]

Ui! Era tão bom que isto viesse para fora do mercado chinês nem que fosse pela curiosidade de ter a sucessora de powervr/kyro disponível

Seria também curioso ver o quanto o IP da imagination consegue escalar fora de mobile.

 

[Dark Kaeser]

Também tenho alguma curiosidade em que alguém fizesse um teardown, para saber a quantidade de componentes chineses, e qual a FAB usada. SMIC?

 

[muddymind]

Muito provavelmente será mesmo na SMIC.

Já agora este GPU é de certa forma parente do apple GPU uma vez q também é baseado em IP da imagination.

 

[muddymind]

Afinal não é assim tão claro qual a fab que vai produzir isto:

"Innosilicon is a world class one-stop shop of high-speed mixed signal IPs and ASIC custom solutions for a full range of processes from 130nm to 5nm across the world’s top foundries (TSMC/Samsung/GlobalFoundries/SMIC/UMC/Intel/HLMC/XMC, etc.), especially for FinFET processes below 22nm. So far, Innosilicon is the sole technology partner in China who has access to the 5nm process library provided by Samsung and TSMC combined with the design-to-tapeout capability."

https://www.innosilicon.com/html/about/about-us/

 

[blaster_00]

A AMD e a NVIDIA com a história das criptocoinas estão a descurar a gama de entrada e quando derem por isso podem estar numa má posição.

Uma porcaria de uma GT730 custa 90 euros !! e é se houver stock! e uma 6600 custa praticamente um salário mínimo nacional !!?

 

[ToTTenTranz]

A AMD e a NVIDIA com a história das criptocoinas estão a descurar a gama de entrada e quando derem por isso podem estar numa má posição.

Gamas mais baixas = margens mais baixas e ASP mais baixo, ambos fatores que tendem a desvalorizar o desempenho das empresas face aos accionistas. Os dGPUs sub-$300 parecem estar a caminho da extinção, e a menos que haja um crash das crypto massivo que meta todas as gráficas mineiras de volta no mercado de usados, essas gamas não vão voltar.

Eventualmente, a AMD vai substituir as gamas baixas de dGPUs com APUs / SoCs com DDR5 e VCache, esses sim que podem vender a um preço mais alto mas que no conjunto fica mais barato para o consumidor. A Intel provavelmente vai pelo mesmo caminho.

 

[Dark Kaeser]

Aparentemente apresentadas

Innosilicon graphics cards based on “Fantasy One” GPU feature up to 32GB GDDR6X memory​

As many as four products have been revealed, including dual-GPU and single-GPU solutions. Type A is a consumer/workstation graphics card featuring a single Fantasy One GPU, which is actually a multi-chip (chiplet) design. Unfortunately, none of the news reports from China on this announcement had the exact configuration of the Fantasy One.

According to the data provided by Innoslicon, this GPU offers up to 5 TFLOPS of single-precision compute power and 160 a fillrate of GPixel/s. It is equipped with DisplayPort 1.4, HDMI 2.1, and VGA connectors. This card is to be equipped with up to 16 GB of GDDR6(X) memory across a 128-bit interface. So far, the G6X technology has been exclusive to NVIDIA/Micron Ampere GPUs, but apparently, Innosilicon made a lot of research in PAM4 signaling and was able to unlock up to 19 Gbps memory bandwidth for their GDDR6X implementation.

The Type B (a triple-fan or fanless design pictured below), is a dual-GPU solution featuring two Fantasy One GPUs connected by an interface known as Innolink. The company claims up to 10 TFLOPS of computing power and 320 GPixel/s fillrates. This card can offer 32 simultaneous 1080p/60FPS streams or 64 streams at 720/30FPS. It is to feature up to 32GB of GDDR6(X) memory but is again limited by dual 128-bit interfaces from each GPU. All these cards feature a PCI-Express 4.0 interface at full X16 width.

https://videocardz.com/newz/innosil...tasy-one-gpu-feature-up-to-32gb-gddr6x-memory

Core Motion Technology officially released the "Fenghua No. 1" graphics card: the first domestically produced 4K high-performance GPU
https://news-mydrivers-com.translat...zh-CN&_x_tr_tl=en&_x_tr_hl=pt-PT&_x_tr_pto=sc

Seguem-se as Gen2 e 3...

 

[Dark Kaeser]

Imagination Tech Publishes Open-Source PowerVR Vulkan Driver For Mesa

Imagination Tech has been working on a new Vulkan driver, shader compiler, and Linux kernel driver for their PowerVR graphics processors. An initial merge request for landing the Mesa Vulkan driver has been submitted and undergoing review. There is the preliminary DRM PowerVR kernel driver also available via a Git branch but still undergoing additional changes before trying to upstream that into the Linux kernel.

Imagination's current driver focus is on their newer PowerVR "Rogue" GPUs with initially focusing on the GX6250 and the AXE-1-16M and BXS-4-64 GPU designs. The developers have been using a Chromebook Acer R13 laptop as the target device for development due to its GX6250 GPU found with the Mediatek MT8173 SoC.

PowerVR's Mesa Vulkan driver is inspired by the RADV open-source Radeon Vulkan driver and can work both with their new PowerVR kernel driver and the PVRSRVMKM kernel driver included with their development kit. At the moment the PowerVR Mesa Vulkan driver is very limited in its abilities to like triangle demos with Vulkan, but over the months ahead Imagination Tech engineers expect to get this driver into better shape so it can be of use to end-users. At this stage it also doesn't appear Imaginaiton will be supporting pre-Rogue graphics IP with this new open-source driver stack.

https://www.phoronix.com/scan.php?page=news_item&px=Open-Source-PowerVR-Vulkan

Imagination e drivers OSS na mesma frase

 

[Dark Kaeser]

Isto ainda é para as mais antigas, penso eu de que, mas para não estar a abrir um tópico

Imagination Open Access removes license fee for selected GPU and AI IP​

Imagination Technologies has made it easier for developers to access selected GPU and AI accelerator IP from the company by removing the licensing fees for those cores. Its new Open Access program provides access to four PowerVR Series8XE GPUs and three PowerVR Series3NX NNAs. A software driver is supplied by Imagination, and compatible open-source drivers are now also available in the market.

Imagination said the way in which its program differs from Arm’s standard Flexible Access package is that the latter provides access to a range of Arm IP, support, tools and training for an annual $200K fee. Once the customer tapes out their design they pay an additional license and once they reach production a unit-based royalty.

In contrast, Open Access provides access to a range of Imagination IP, software, and tools with a $0 license fee. Unlike Arm, there is only a $135K/year fee for engineer support for the first two years. Customers also only pay a royalty once they reach production. This decreases with volume and follows the commercial model after 10 million units.

https://www.embedded.com/imagination-open-access-removes-license-fee/

 

[Dark Kaeser]

Andava aí um grande alarido sobre as Fenghua 2, afinal é uma gráfica para ficar abaixo da Fenghua 1...

https://twitter.com/i/web/status/1555064097347084288

 

[Dark Kaeser]

Está quase...

Imagination PowerVR Rogue DRM Linux Kernel Driver Out For Review​

Imagination Tech engineers sent out their initial DRM driver that is a clean sheet driver for PowerVR graphics. Like their Vulkan driver, the Direct Rendering Manager driver is intended for their Rogue architecture and newer. The hardware currently being tested on and their primary focus are the GX6250, AXE-1-16M, and BXS-4-64 GPUs.

Simultaneously they have been working on a proper, upstream-friendly open-source DRM/KMS kernel driver and that code now is far enough along that it's been sent out for initial review.

This Imagination PVR DRM driver has wired up support for DMA-BUF, PRIME, DRM sync objects, per-context virtual address spaces, power management, render job submission, compute job submission, GPU hang recovery, and more. But there still is work left around the areas of supporting their RISC-V firmware processor, GPU hang detection, dynamic voltage and frequency scaling, transfer job submission, and more.

https://www.phoronix.com/news/PowerVR-DRM-Kernel-Driver-RFC