It’s no secret that AMD’s Radeon RX 5500 XT is currently selling rather sluggishly, as the cards on offer are simply too expensive for what many customers are bidding. On average even slower than an old Radeon RX 590 as the last Polaris iteration, this card is nothing more than a much too expensive side-grade and actually only a replacement for the even older Radeon RX 580. And that’s exactly where AMD has created a problem for itself, because while the RX 590 has already been set EOL, the RX 580 will, according to internal sources correctly, still be available well into 2020. This is pure self-cannibalism and not a good starting position for the Radeon RX 5500 XT, which I found quite acceptable in the launch article, except for the price and the rather tight memory interface.
However, the Radeon 5500 XT is certainly anything but a usable Vega replacement, if it is not enough for Polaris. Since these cards have also been EOL for a while, AMD must urgently offer a replacement to the trade and customers, which exactly meets this performance class including price point, if one doesn’t want to lose this field completely to Nvidia. Rumors about a Radeon RX 5600 XT have been around for quite a while and it would be pointless to rehash or summarize all the rumors.
From my own sources at the regular board partners of AMD and a graphics card producer in China, for whom the board partner status is too expensive and as a corset rather complicates the Asia business from his point of view, now concurring information has condensed, which I would like to summarize because it seems plausible. Because there is not so much more time to completely overthrow everything again. In order to be able to provide the chip faster and also cheaper for the board partners, one sets according to the sources in agreement on Navi 10, whereby one can play here naturally also again the large Salvage flute, which is nothing dishonorable, but economically even very meaningful.
The best way to do this is to cut the chips using a hardware strap, so that they would remain pin-compatible with the existing boards of the board partners, which in turn greatly shortens the release time and minimizes costs. But since you can’t “switch off” anything from the outside, you only have to split the shader engine and its 4 big blocks with the 40 RDNA compute units, respectively the 80 Scalar processors, 2560 stream processors, 160 bilinear filter units and the four 64-bit memory controllers and at least 4 MB L2 cache.
If you want to shorten the whole thing in the simplest way, then this would most likely result in a 3/4 solution, which then also fits perfectly to the 6 GB memory, which is well secured as one of the memory controllers has to be omitted. This results in an aggregated 192-bit bus for the memory, which can then be 6 or 12 GB (although the latter is of course only theoretical). Furthermore, you can deduce 3 MB L2 cache and calculate the rest, which would then amount to 30 Compute Units (CU). This would result in 1920 streaming processors, 120 texture units, 48 ROPs for a RX 5600 XT, which could theoretically, assuming the same clock speed, deliver about 75% of the performance of a RX 5700 XT.
But if AMD would use the possible 70 to 75% to the full, which would require the same clock rate, it would only be a replacement for the Vega56, but not for the Vega64, which would be much better in WQHD. One may be curious, but it won’t be more than a Vega side grade with the trend to weaker Vega, just like the sandwich constellation of the RX 5500 XT between the RX 580 and the RX 590. And AMD’s marketing will have to ask itself how to place a 6GB card between two 8GB cards. Technically absolutely plausible but very difficult to convey to the rather ignorant customer who doesn’t know what bandwidth is.
From this point of view the launch in January, if the information is really correct, will simply have to wait without emotion. The third calendar week (week 3) will be the date of the launch, which is rather sporty considering the current development status of the board partners. It can be assumed that AMD prefers to send the exclusive board partners including Asus into the race, which wouldn’t really be new. So let’s be surprised, something will (have to) come, one way or another.