(2/2)
The process is so greedy on PC, we say that it is an announcement effect, a triple headline A can never handle it properly.
Just take a look at Nvidia, where the arrival of Ray Tracing is in pain. The loss of performance (falling fps and / or limited resolution) was close to 45%: Ray Tracing activated on Battlefield V with a RTX2080 , all the same.
Fortunately, things are getting better and performance losses have been reduced (unless they are mistaken they are around 30% today, depending on the games, the level of Raytracing requested, and the optimization of the game).
Paradoxically, one of the most beautiful demo of Ray Tracing comes from Minecraft: and we realize, again, that the resources that the process requires (in "High" level) are important, which impacts quite concretely the resolution and / or fps of the game (config PC close to the One X: 720p / 30Fps - Config PC HDG of the moment: 1080p / 60Fps);
Result, with all these elements, we had quickly ranked this story of Ray Tracing in the same bag as the 8K: arguments straight out of a com plan to create the illusion to the publIc.
Let's get to the point! These handwritten documents are a mine of information.Some areas have been blackened as if to hide some key details. The visible elements, however, are sufficiently numerous and precise to be considered and analyzed.
One of the pages of the leak: Lockhart SoC (left) and Anaconda (right)
As you can see in this picture, it is a diagram of the Lockhart's System on Chip (SoC) and the Anaconda model, usually consisting of a CPU, a GPU, and a memory controller.
And there, housed on the left side, 1 to 2 chips "RCC" are grafted on the SoC , depending on the model.
We wonder for a few seconds, and the little light bulb in our heads lights up (it's rare but it happens!) These are chips " Raytracing Compute Chiplet " deploying 1 Teraflop each according to the document.
The Lockhart would benefit from an RCC chip , while the Anaconda would have two.
The interest will be appreciable in both cases. The developers will be able to use these chips without coming "to type" in the "11,17 Teraflops" of your Anaconda. The latter having 2 Teraflops available for Ray Tracing (among others).
Note that you can not really add the Teraflops together. We will not be able to say that this Xbox has 13.17 GPU power Tflops. That would not be very honest (but marketing voices are impenetrable, is not it, Mr. Cerny?)
In short, yes, our future consoles would have every chance to handle Ray Tracing in hardware! BOOM!
And it's a (small) revolution! At AMD, there was a tendency to state, even recently, that it was "too early" to see Ray Tracing management directly integrated into the US brand's GPUs .
The semiconductor manufacturer's officials then invoked their wish to wait for the Microsoft DirectX12 API to be updated with the DXR (DirectX Raytracing), to finally get really into the race for high-end or new generation GPUs. managing the Ray Tracing.
But then, a conclusion is necessary. If we rely on the presence of a Navi GPU in our future Xboxes, knowing that they will be the last built under the GCN architecture (see
part 1 ) and if we take into account the declarations of "AMD, we did not imagine for a moment that the management of Ray Tracing in hardware would be possible (the engraving in 7nm of Navi helping a lot to this result).
At best, we thought of a software management of the process, which would have been disappointing in a sense (partial Ray Tracing), but never to such a scenario of chips dedicated in hardware. We were the first to calm the enthusiasm of some enthusiasts on this subject, to avoid false hopes and disappointments. The surprise is all the more beautiful.
This leak is definitely surprising, because no one saw coming this AMD acceleration on the subject.
Microsoft x AMD
In fact, there is a great collaboration between Microsoft and AMD that has been put in place, about Ray Tracing . The tandem work would have started 4 years ago, underwater, and very clever one who claims to have been aware at the time.
Thus, AMD would have let Nvidia pull the media coverage on him with his "RTX GPU" between late 2018 and 2019. Letting believe that they were a little behind and almost overwhelmed on the subject of Ray Tracing. It was far from being the case if these dedicated chips are confirmed!
Conclusion: 2020 will be the year of Ray Tracing for AMD , whether on consoles (Xbox Scarlett and PS5) or on PC (Navi 20).
The functionalities of RCC chips
By the way, these RCC chips are not irremediably intended to make Ray Tracing: they can be used to make " 3D sound " or even " physics "!
"Physics" ? A few weeks ago, a rumor about the presence of a chip dedicated to physicsin the next Xbox, you remember? Was it actually an RCC chip? Possible. The thesis takes a little more thickness as to its veracity.
Better still, this corroborates Mark Cerny's statements on the subject of Ray Tracing .He also evoked the management of 3D sound on the "Playstation Next" with the "Ray Tracing technique".
The other elements to take into account.
As the Ray Tracing part is well underway, let us take a quick look at the rest of the information available on this document:
We observe that :
- The Lockhart would have 6 GDDR6 memory chips at 2GB each or 12GB . Not to mention 4GB of DDR4 (not visible on the diagram) dedicated to the OS and applications .
- the Anaconda version would have 12 GDDR6 memory chips at 2GB each, ie 24GB. Again, 4GB DDR4 would be at work for the OS and applications . We will come back in more detail on this subject of the RAM in our 3rd part!
- In both cases, Microsoft would use the same GDDR6, the same Zen 2 CPUs, the same RCC chips and the same type of GPUthe whole being simply divided by two (excluding CPU).
An ingenious production for a foolproof cost control?
When you see what the Xbox Anaconda has in the guts, it's almost hard to imagine that the price of 499 € is possible .According to this leak, apart from the 28GB of RAM (24 + 4), the 2 RCC chips, the 11.17 TF of the GPU, we note the presence of 2 hard drives (one to 512GB SSD, the other 2TB HDD). Hard to imagine the same price as a One X launch in 2017.
But who says difficult, does not necessarily say impossible.
Always relying on documents, we are told a clever system to avoid losses during production.
As the table shows, 4 elements make up the "Scarlett" range:
- A dev kit
- XCloud servers
- An Xbox Anaconda
- An Xbox Lockhart
In
part 2 , we talked a lot about Compute Units (CU) . And this is where the production genius begins. Exactly as expected, the Xbox Anaconda will benefit from 60 CU on its GPU (as well as xCloud according to the document). The dev kit would be 64 CU and the Lockhart 30 CU.
How to know ? The document does not mention it!
Indeed, the person at the origin of this leak does not evoke the number of CU present on each model. Not very important, we can do it ourselves:
We have to repeat the calculation we had done in part 2, namely:
60 (CU) x 64 x 2 x 1455 (Mhz) = 11.17 Teraflops .
And now, we have our CU number. You must add 4 CU to the dev kit or 64 CU to get the 11.92 TF shown on the chart. 64 CU is the maximum possible on the GCN architecture of AMD (on which Navi depends).
Microsoft will order its 64 CU Navi GPUs from AMD. Some chips can leave the production line with less than 64 CU. We had seen it previously, and that is generally why a console manufacturer provides a margin of error (here of 4 CU).
Concrete example: even if the GPU comes out of the factory with 62 CUs active, it will remain usable for an Anaconda at 60.
Several cases then arise to limit the dry losses:
- If the GPU comes out at 64 CU : it will be able to benefit the dev kit in priority (and / or to the rest of the range, by disabling the surplus CU if necessary)
- If the GPU comes out between 60 and 63 CU : Microsoft will be able to use it for its Xbox Anaconda and for its xCloud servers, both being at 60 CU.
- If the GPU comes out below 60 CU : No loss again, since it can be used in a Lockhart at 30 CU (by massively deactivating the CU not useful.Beware, this point does not mean that Microsoft will not order for as many GPUs directly in 30 CU)
- If the Lockhart GPU comes out under 30 CU : well ... no solution: trash.
Initially we had articulated our predictions and calculations for the next Xbox around the sacro-saints CPU, GPU, RAM and other HDD: elements that are found at all times and that can be projected.
Today, we touched the subject of Ray Tracing that we did not imagine at all this way. RCC chips can go further. The "glass ceiling" in terms of raw power reaches its almost current limit on console with more than 11 Teraflops (under GCN architecture obviously). The other solution to climb even more in the tower would have been a dual GPU , but for obvious reasons of cost and production, this scenario is impossible.To overcome this, the introduction of RCC chip is new and makes sense. It leaves the raw power of the GPU (11.2 Tflops) free for its usual work by unloading Raytracing calculations. In this, the next Xbox would be more surprising in its capabilities than we thought!
If all this is confirmed, have no doubt about the exceptional character of your future Xbox. The "Monster" has not finished roaring.