30. Apr. 2025: Neue PPC-Hardware angekündigt: "Mirari" mit USB3, NVME, Onboard-FPGA (Update 3)

Dave 'Skateman' Koelman stellt im "Hardware"-Channel des Amigans Discord-Servers ein neues, von Harald "Geennaam" Kanning entworfenes PPC-Motherboard vor. Inzwischen hat sich auch Kanning selbst im Kommentarbereich von amiga-news.de zu Wort gemeldet und uns eine ausführliche Beschreibung des Projekts zukommen lassen:

Aktuell weist das "Mirari" getaufte Projekt die folgende Eigenschaften auf:


Background and Mission

The Mirari board is designed with the intention to provide the Amiga community with low-cost PowerPC hardware. Therefore, the design is based on a bare minimum with nice-to-haves based on what makes sense, but always with the mission in mind. Due to the minor cost difference between a dual-core T1022 and the quad-core T1042, we've decided on the T1042, resulting in a low-cost board with mid-range specifications.

The T10x2 is based on the e5500 core, the same core found in the X5000. Initial benchmarks confirm that the single-core performance difference is just the clock difference, with a 1.4 GHz T1042 performing at 70% of the 2GHz X5020. The design also accommodates the pin and footprint-compatible T2081, allowing for a mid-cost board with high-end specifications. To ensure compatibility, we had to make some compromises to the T10x2 design.

Memory Support

The T10x2 memory controller supports both 64-bit DDR3L and DDR4 at 800 MHz (DDR1600). The T2081 memory controller supports both 64-bit DDR3L at 933MHz (DDR1866) and DDR3 at 1066MHz (DDR2133). The common denominator is DDR3L, hence the choice for a 1.35V SODIMM slot.

If we had focused solely on the T1042, DDR4 would have been the better choice from an availability standpoint. However, since the clock speed is the same but latencies are larger for DDR4, the overall memory performance would have been lower.

For the T2081 we can increase the memory controller voltage to 1.5V which would enable the DDR2133 interface speed.

Although we have only one memory controller, initial Linux benchmarks show that the single and SMT memory performance is at least on par with the X5020.

Audio

For longevity reasons, we've decided to use an industrial-grade audio codec from TI (TLV320AIC3204). This codec is routed to three audio jacks at the rear (mic, line-in, line-out) and to an onboard HD audio connector. Additionally, there's a digital optical output (Toslink). The T1042 features an I2S interface for direct connection to external audio codecs, while the T208x does not. Therefore, we've implemented the I2S and S/PDIF controllers in the FPGA. This means there's no audio driver readily available for Linux. For OS4 and MorphOS, I have experience writing AHI drivers, so consider this covered.

Storage

The mainboard features two SATA ports and two NVMe slots. However, we had to make some tough choices again:

For the T1042 variant, there's always one SATA port and one NVMe slot available. The second SATA/NVMe is mutually exclusive, currently jumper-based but eventually u-boot configurable.

The T2081 does not have SATA interfaces. If you need SATA, it must be added with a PCIe card. However, keep in mind that SATA is being phased out, with modern PCs and laptops often using NVMe over SATA SSDs.

Tested Linux distributions are all booting and running from NVMe. I also have experience with NVMe drivers, so consider this covered for OS4 and MorphOS as well.

USB3: We based u-boot on the latest stable version available earlier this year, 2024.10. This has many benefits over using the 2014-2018 versions available in other PowerPC hardware. However, the developers behind u-boot changed the driver model around 2018, with a transition period ending in 2020. This means u-boot is a mixture of old drivers and drivers compliant with the new model. For USB on NXP SoC, there's a catch: NXP didn't rewrite their OHCI and EHCI drivers for the new model. Activating the XHCI driver for USB3 support causes conflicts. Fortunately, this is not an issue under Linux, so USB3 works correctly, but we have work to do for u-boot.

Prototypes

We have ordered 5 prototypes and managed to make them all work (100% yield). This was a great concern to me personally, as the layout requirements for the 780 balls FPGA stretched our supplier's capabilities to its limits. The boards run stably and have undergone hours of burn-in testing with Linux stress testing tools. One observation is that the CPUs run around 30°C idle and around 40°C under full load, which is good for overclocking potential. The core power supply has quite some headroom to support this. You can check out the Hyperion Entertainment X5000 Linux-only forum for posts from Skateman. Whenever he talks about his 4-core e5500, he is actually talking about this board.

Update: (30.04.2025, 23:20, cg)

Einige Anmerkungen des Hardware-Entwicklers aus einem nur für registrierte Discord-Nutzer zugänglichen Kanal (Official Commodore Amiga, #next-gen):

Update 2: (01.05.2025, 13:15, cg):

Board-Designer Harald 'Geennaam' Kanning hat im Kommentarbereich von amiga-news.de eine ausführlichere Beschreibung des Projekts veröffentlicht.

Update 3: (01.05.2025, 15:39, cg):

Meldung komplett überarbeitet um sämtliche bisher verfügbaren Informationen besser strukturiert zu präsentieren. (cg)
< Veranstaltung: Canberra Vintage Computer Exhibition 2025 | Lesestoff: Das kurze Commodore-Intermezzo von Thomas Rattigan >