Some RTX 50 cards do not have 8 ROP units to lose several percent of energy

The release of video cards RTX 50 was definitely not without problems. From the unsatisfactory shift on the price/power scale, thanks to the return of molten connectors or the absence of 32 -bit Physx support, the problems are constantly breaking up, and in the last few days or two. Some owners of new cards found that there are no 8 units of ROP in the chip, the so-called rendering device that takes care of the final processing of pixels, including anti-Alias. Of course, this is not a pleasant find at all, because the absence of several Rops leads to a decrease in performance.

For the first time in the TechPowerup magazine, whose editors noticed that one of their RTX 5090, in particular, is not enough ZOTAC with the ZOTAC, already mentioned 8 ROP units, out of 176, only 168. Add other users with a report the same problem, not only on RTX 5090 (and as soon as it turned out, regardless of the manufacturer, the concrete structure), but also on RTX 5070 Ti and in accordance with the latest reports and RTX 5080.

NVIDIA recognized the error relatively quickly, and the company representative in the reports of technological journals said that only half of the percent of the total number of users should be affected, but all these customers should apply for the manufacturer of cards. For recently made cards, the error was to be fixed.

At the same time, the company claims that the sanction in graphic indicators should be about 4 %. But it is not entirely clear how they came to the quantity, as it depends on the model. While RTX 5090 turns off 8 ROP units, it will lose a little more than 4 % of the total number, RTX 5070 Ti is more than 8 %.

Why 8 units? The entire case was provided for a detailed Nexus video game in which Steve Berke explains that this is due to the structure of the chip. The one that uses RTX 5090 carries GB202 and has a total of 12 GPC (graphic cluster), each with 8 TPC (texture processing cluster), and each TPC has 16 units of ROP (in two blocks 8 8) and two SM (two SM ( Streaming multiprocessor), each SM contains 128 Cuda Jader.

The full chip GB202 has 96 TPC, but RTX 5090 has only 85, at least actively. Thus, nothing is unusual, manufacturers of chips protect against defects, which would mean the inability to use such a product when deploying a full chip. Therefore, the standard procedure is that if the chip has a number of defects within the boundaries, it can be used as a given model, but, as you can use a fully functional chip, only the specified TPC number will be turned off. But it seems that NVIDIA has made a mistake in creating a series of RTX 50.

Burk suggests that if the company deactivated one GPC (8 TPC + 16 ROP units), and then another TPC from another GPC, there may be a situation that 3 deactivated TPCs have been next to one block of 8 units, can also block access For these divisions ROP. Nevertheless, this is actually just an assessment of what can happen in production, it is quite possible that NVIDIA simply sells chips that had more defects than are allowed in accordance with specifications.

If you want to make sure that your card also includes one ROP unit, you can do this in the GPU-Z program, which will give you detailed information about the map. The correct number of ROP units should be 176 for RTX 5090, 112 for RTX 5080 and 96 for RTX 5070 Ti.