Today I bring you a two-part AMD news story, the first of which concerns AMD’s Navi 23 GPU, which is the high-end monster graphics architecture AMD have been working on based on the second generation of RDNA.
regular viewers/readers will know that we had first leaked the existence of Navi 23 in August of this year, along with Navi 21. According to my source, AMD internally refers to Navi 23 as the ‘Nvidia Killer‘.
We now, the GPU has been spotted in drivers for the first time, obviously all but confirming that Navi 23 is indeed a thing. The code was spotted by a forum goer on 3dcenter.org full credit to Ronald there who also emailed me to let me know that it exists.
The code added to the AMD graphics drivers is as follows (Navi 23 and Navi 22):
#if ADDR_NAVI22_BUILD
#define AMDGPU_NAVI22_RANGE 0x32, 0x3C
#endif
#if ADDR_NAVI23_BUILD
#define AMDGPU_NAVI23_RANGE 0x3C, 0x46
#endif
We also see a Navi 12 Lite piece of code too, which is interesting given that apparently Navi 12 Lite is rumored to be for console use.
#if CHIP_HDR_NAVI12_LITE || CHIP_HDR_NAVI21_LITE || CHIP_HDR_NAVI_LITE
struct
{
unsigned int count : 22;
unsigned int : 10;
} gfx10Bard;
#endif // CHIP_HDR_NAVI12_LITE || CHIP_HDR_NAVI21_LITE || CHIP_HDR_NAVI_LITE
As you can see, there aren’t any specs listed yet for Navi 23 or Navi 22 either, so things are still up in the air. Given that Nvidia is rumored to be launched Ampere in the 1st half of 2020, we can guess AMD will be facing a lot of competition in the video card arena. The rumor has it that Ampere will also double down on both ray tracing and also traditional rasterization performance.
There’s also been a very curious update from Microsoft, with the company moving to update DX12 ray tracing support to Tier 1.1, which features numerous enhancements over the ‘base’ spec. One of those is execute indirect, which allows the GPU to essentially adjust the number of rays being generated based on what’s going on in the GPU pipeline.
I find this particularly of interest given what we know about AMD’s second-generation Navi/RDNA architecture. I first leaked that it supported ray tracing back in March, but then AMD confirmed this with the ray-tracing vision slide. I find this slide very interesting given that it seems to indicate that the first-gen of Navi will run ray tracing in software, but the second will be a hardware-based solution (matching up to my leak).
Then we saw the ray tracing patents filed by the company, which is a hybrid solution using the traditional graphics pipeline along with some new and custom hardware to calculate ray traversal. This is pure speculation on my part, but given Microsoft has confirmed that Xbox Scarlett will feature hardware ray tracing, this update will likely benefit the next-gen consoles too.
Indeed, another of my recent leaks also said that Microsoft’s ray tracing solution was a little more advanced than Sony’s in the Playstation 5, though in terms of raw CPU performance and many other areas the two systems so far are quite close to each other.
I was originally told that Navi 10 was running quite far behind schedule, and was originally scheduled to launch much earlier this year (2019), versus the July date, it shared with the third-gen ryzen 3000 products. This, of course, was the reason we saw Vega 20 come to gamers in the form of Radeon VII (which we also leaked exclusively might I add).
Navi 12 I have been told is launching next year, and will be a ‘big Navi’ part, ie, more compute units/shaders compared to the RX 5700 XT and the Navi 10 die (which is comprised of just 40 CU). Navi 12 is also seemingly a slightly different architecture compared to Navi 10 – although we don’t know many of the specifics yet.
We can presume that the second-gen Navi parts will launch in the latter part of next year, and it will be fascinating to see how AMD’s GPU roadmap evolves over the next few years and what both Intel and Nvidia do to counter it. Navi 23 (and this is speculation) likely wouldn’t launch sooner than Q3 2020.
The second part of the leak today concerns the 7nm APU known as Renoir, as a source of mine has given a little more insight into what we can expect with the next-generation parts.
I was told that Renoir is being manufactured on a monolithic die (and therefore, isn’t chiplet like say… an AMD Ryzen 9 3950X). This actually makes sense as earlier this year there were reports that APUs will not be using chiplets – yet anyway.
I was also told that Renoir will go up to 8 CPU cores, and of course, use the Zen 2 architecture… albeit with a cut back amount of L3 cache. This is likely for cost-saving and also space if I had to take a guess. I was told it will feature a Vega 12 CU for certain, but we’ve seen that Renoir appears to be Vega-based in terms of the CU, but the display units are more Navi based. It will support LPDDr4 too (although we’ve already had that leaked and confirmed by other folks).
