Nintendo Fusion Terminal & Fusion DS Rumor Analysis

Sony and Microsoft’s next generation consoles (or should I say, current generation consoles) the Playstation 4 and Xbox One are pretty impressive pieces of kit. The Wii U by technical comparison is lacking, in CPU, RAM and GPU terms it comes up short against either of its rivals. So it’s fairly obvious that another console is being designed in Nintendo’s headquarters, and the it’s highly likely to be more powerful than either the Xbox One or PS4. If you’re arriving to the party late, you better have a damn good reason.

Recently, there have been rumors floating about the internet of anonymous leaks of Nintendo’s Next Generation systems. They’re supposedly called Fusion Terminal (for the home console, which will be the Wii U’s successor) and the Fusion DS (you guessed it, will replace the DS). The supposed leaks come from an anonymous origin and therefore aren’t being pushed as reliable. So let’s take a look at the rumored specs of both systems.

Nintendo Fusion Terminal Rumored Leaked Specs

• GPGPU: Custom Radeon HD RX 200 GPU CODENAME LADY (2816 shaders @ 960 MHz, 4.60 TFLOP/s, Fillrates: 60.6 Gpixel/s, 170 Gtexel/s)
• CPU: IBM 64-Bit Custom POWER 8-Based IBM 8-Core Processor CODENAME JUMPMAN (2.2 GHz, Shared 6 MB L4 cache)
• Co-CPU: IBM PowerPC 750-based 1.24 GHz Tri-Core Co-Processor CODENAME HAMMER
• MEMORY: 4 Gigabytes of Unified DDR4 SDRAM CODENAME KONG, 2 GB DDR3 RAM @ 1600 MHz (12.8 GB/s) On Die CODENAMED BARREL
• 802.11 b/g/n Wireless
• Bluetooth v4.0 BLE
• 2 USB 3.0
• 1 Coaxial Cable Input
• 1 CableCARD Slot
• 4 Custom Stream-Interface Nodes up to 4 Wii U GamePads or 4 DSc
• Versions with Disk Drive play Wii U Optical Disk (4 Layers Maximum), FUSION Holographic Versatile Disc (HVD) and Nintendo 3DS Card Slot.
• 1 HDMI 2.0 1080p/4K Port
• Dolby TrueHD 5.1 or 7.1 Surround Sound
• Inductive Charging Surface for up to 4 FUSION DS or IC-Wii Remote Plus Controllers
• Two versions: Disk Slot Version with 60 Gigs of Internal Flash Storage and Diskless Version with 300 Gigs of Internal Flash Storage.

Nintendo Fusion DS Rumored Leaked Specs

• CPU: ARMv8-A Cortex-A53 GPU: Custom Adreno 420-based AMD GPU
• COM MEMORY: 3 GB LPDDR3 (2 GB Games, 1 GB OS)
• 2 130 mm DVGA (960 x 640) Capacitive Touchscreen
• Slide Out Design with Custom Swivel Tilt Hinge
• Upper Screen made of Gorilla Glass, Comes with Magnetic Cover>Low End Vibration for Gameplay and App Alerts
• 2 Motorized Circle Pads for Haptic Feedback
• Thumbprint Security Scanner with Pulse Sensing Feedback
• 2 1mp Stereoptic Cameras
• Multi-Array Microphone
• A, B, X, Y, D-Pad, L, R, 1, 2 Buttons
• 3 Axis Tuning Fork Gyroscope, 3 Axis Accelerometer, Magnetometer
• NFC Reader
• 3G Chip with GPS Location
• Bluetooth v4.0 BLE Command Node used to Interface with Bluetooth Devices such as Cell Phones, Tablets
• 16 Gigabytes of Internal Flash Storage (Possible Future Unit With 32 Gigbytes)
• Nintendo 3DS Cart Slot
• SDHC “Holographic Enhanced” Card Slot up to 128 Gigabyte Limit
• Mini USB I/O
• 3300 mAh Li-Ion battery

On first glance, the specs of both the Terminal Fusion and the Fusion DS are impressive. Taking the Terminal Fusion’s GPU for instance, it sports 4.6TFLOPS of computing power vs the Playstation 4’s 1.84TFLOPS. That’s around 2.5x the grunt in pure graphics processing speed. Indeed, these figures put it more akin to a current high end PC than a games console. But, there are a few things which worry me about just how true these leaks are, but I’m going to focus on the ‘main’ specs of the systems. Let’s start with the Nintendo Terminal Fusion.

The IBM Power 8 Processor – in all its glory

Terminal Fusion ‘leak’ analysis

The supposed leak of the Terminal Fusion GPU reads as follows “GPGPU: Custom Radeon HD RX 200 GPU CODENAME LADY (2816 shaders @ 960 MHz, 4.60 TFLOP/s, Fillrates: 60.6 Gpixel/s, 170 Gtexel/s)”. We don’t have the ROP count, or the number of TMU’s listed. But it’s likely that the ROP count is around 64. We do have the number of shaders, 2816. Notice the clock speed of the GPU is listed at a rather nippy 960MHZ.

Let’s use the PS4’s GPU as an example, which is also using the GCN (Graphic Core Next) architecture. It has 1152 Shaders, running at 800MHZ and puts out 1.84TFLOPS. Knowing how the GCN architecture works, we can easily express this. 1152 shaders x 2 x 800MHZ. This gives us a number of 1843200. Thus providing us with the 1.84TFLOPS. How this works is simple. You take the number of shaders, and multiply this by 2 (as each shader in simple terms does two operations per clock). Then after you’ve done this, you multiply by the clock frequency and you’re given the result.

Let’s try that with the Terminal Fusion’s GPU. The GPU lists it’s using the R200 based architecture, so that’s likely to be GCN 1.1, this performance is calculated in the exact same way. The Terminal Fusion says it has 2816 Shaders x 2 x 960. This provides us with a number that’s actually significantly higher than that of the advertised specs, 5406720. There are three possibilities. The first is that the clock speed / number of shaders proposed is wrong. The second is that a portion of the GPU’s performance is blocked off to games developers for system operations (much like how the Xbox One’s GPU works). The third is that there’s something seriously suspicious with these numbers.

Some of the current GCN architecture – click for larger image

As for the Fill Rate, GPixels and GTexels, it appears they are right around where they should be for an R9 290 or the R9 290X. The R9 290X for instance has 176 Texture Units, providing 176 GT/s. And there is a similar Fill rate of 64 Giga Pixels /s. Thus we can guess that we’re looking at a similar amount of TMU’s and ROPs.

Now let’s take a look at the memory. “MEMORY: 4 Gigabytes of Unified DDR4 SDRAM CODENAME KONG, 2 GB DDR3 RAM @ 1600 MHz (12.8 GB/s) On Die CODENAMED BARREL”. I’m unsure why a next generation console with so much GPU power would have only 4GB of memory, and stick with DDR4 to feed the GPU. DDR4 RAM (although it’s clock speed or bandwidth aren’t listed here) would provide less throughput than GDDR5. It just doesn’t have enough RAM to make all that shading power worth it. Creators of the middleware, the Havok Engine have already started mentioning that the PS4’s 8GB GDDR5 RAM is looking to be filled up quickly by texture artists. I don’t see what good only 4GB of unified memory would do.

I confess, there is 2GB on die (at a rather strange memory bandwidth might I add), but that could be written off as purely for Operating System functions for the Terminal Fusion. 2GB would be around the number expected, and would fit in with the PS4’s and Xbox One’s RAM.

Now we shall turn our attentions to the two CPU’s. The first we’ll discuss of the Terminal Fusion is the Co-CPU, the IBM PowerPC 750-based 1.24 GHz Tri-Core Co-Processor CODENAME HAMMER. Let’s just cut this down right now and say that’s the same CPU that’s inside Nintendo’s Wii U (wiki link). If these specs are accurate, it’ll be there primarily for the sake of backwards compatibility with the previous generation Nintendo consoles. The main CPU is listed as: IBM 64-Bit Custom POWER 8-Based IBM 8-Core Processor CODENAME JUMPMAN (2.2 GHz, Shared 6 MB L4 cache). This is based on a new architecture which has recently been unveiled by IBM – Wiki Link here.

The Power8’s raw form is designed to handle 96 hardware threads simultaneously, and for most workloads IBM have said that it’s two to three times faster than the previous generation Power7 architecture. Each of the 12 cores of the ‘normal’ Power8 is eight-way hardware threaded (giving the total of the 96). In terms of per core, the Power8 estimated to be about 1.6 times faster than the previous generation from IBM. Obviously because the Terminal Fusion is a listed as a custom 8 core chip, how much of this remains true is up in the air.

The Power8 CPU has an external chip, known as Centaur, which acts as a memory controller, Level 4 Cache and memory buffer. Centaur acts as 16MB level 4 cache. 2x Centaurs would give 32MB of level 4 Cache and so on. There’s a max of 128MB from 8 Centaur chips. With 8 centaurs, memory bandwidth is 230GB/s (sustained) or 410GB/s peak. It’s possible that Nintendo have customized this chip heavily, explaining the strange level 4 cache. It’s unknown of course how many Centaurs are used in this supposed configuration of the Terminal Fusion. Remember though that this type of spec would be if the Power8 was operating at the propsed 4GHZ (not the 2.2 of the Terminal Fusion). In addition, we’d be seeing this in a super high performance rigs. This setup wouldn’t come cheaply.

The Power8 CPU features CAPI, which is layered atop the PCIE 3. With this, you’re able to share devices such as GPU’s and have them address the same memory address space as the CPU. This would allow the usage of GPGPU in very interesting ways. But ultimately, the total amount of memory bandwidth concerns me. DDR4 puts out roughly half the bandwidth as GDDR5 (this depends on things such as Bus width and clock speed). Because we don’t know about the bus-width it’s going to be difficult to make a judgement, but when you consider that the R9 290 uses 320GB/s of memory bandwidth, the GPU inside the Terminal Fusion GPU would need to be similarly fed to avoid being bandwidth starved.

PCIE on a 16x lane device puts out 16GB/s of memory bandwidth, which is enough for GPU communication to the CPU and to handle GPGPU loads. But the Terminal Fusion would need a setup similar to the PS4, with multiple buses.

There are multiple factors which tells me there is something ‘off’ with these specs. Strange memory configuration, and likely the whole console wold be extremely expensive to produce. If Nintendo targeted the Terminal Fusion for say a 2016 release, parts would be cheaper, but even so. I’d take this rumor with a massive grain of salt. There’s a lot more to this tech analysis, but so much felt strange about these specs I’ll wait and see what develops first.

Sources:
Original article – gaminrealm
IBM Technical presentation
AMD GCN White Papers

Comments are closed.