Old Huangjia NVIDIA new generation graphics card has kept everyone waiting, this time the spy photos are PCB, but the graphics card PCB without soldering components is still very rare.
First of all, we compare with the PCB of the previous generation GTX 1080 Founders Edition (the public version) graphics card, through its PCI-E gold finger length. It can be found that the length of the new generation graphics card is consistent, it should be 269mm. However, the core position is slightly Moved a little on the right.
First of all, we can confirm that this PCB will use GDDR6 memory. The identification method is very simple and rude. It is the number of solder joints in the memory. One line of 18 points, a total of 10 lines, is 180 points, which is just the 180-ball BGA used by GDDR6. The GDDR5X is a 190-ball BGA, and the GDDR5 is a 170-ball BGA. It is easy to distinguish. There are a total of 8 empty soldering positions on the PCB. All used is 32bit×8=256bit width, although GDDR6 will have Single particle 2GB capacity, but currently received the message, still using a single particle 1GB, which is still 8GB of memory.
It is easy to know that GDDR6 memory is available. Next, you can estimate the GPU core area and package area. The GDDR6 memory package size is 14×12mm. Ignore the distortion of the photo shooting angle. It can be calculated that the GPU package area is 37.5mm× 37.5mm=1406.25mm2, and the core area is 20mm×20mm=400mm2. What is the concept of 400mm2? In GTX 10 series graphics card, GP100 has a large core area of 610mm2, GP102 core area of 471mm2, GP104 core is 314mm2, GP106 core area is 200mm2, which is between GTX 1080 and GTX 1080 Ti, 27% larger than the core of GTX 1080. This should not be the flagship model of the new generation of graphics cards. The last leaked engineering card is. Plus will be used Built by 12nm FN process, the full-blood version of this graphics card has 3584 CUDAs. It is flat and GTX 1080 Ti.
In the lower left corner of the graphics card, I found a very interesting thing. In the past, it was HDMI 2.0b×1, DisplayPort 1.4×3, but the video output interface of the new generation graphics card has changed. The bottom right corner of the video output interface pad surface, it In addition to being a DisplayPort interface, you can also use the USB Type-C interface, because USB Type-C is also a DisplayPort channel. Compare the USB-C terminal, Pin pin, and 24 pins to match. The next-generation graphics card can be HDMI 2.0b×1, DisplayPort 1.4×3, or HDMI 2.0b×1, DisplayPort 1.4×2, USB Type-C×1. If it is a non-public version, plus DVI-D No.
When you return to the top of the PCB, you will find that the SLI gold finger is different, and it becomes wider, even like the pre-exposure engineering test card! It means that NVIDIA will give up the SLI that has been used for many years on the new generation of graphics cards. Bridge, switch to a newly designed bridge. Hmmm? This golden finger looks a bit familiar, yes, compared to the fake NVLink bridge on TITAN V, you will think this is a left-right version, very similar, but both should Not universal.
Finally, let's take a look at the power supply. This is too scary. The previous generation of GTX 1080 was only 10 5+1, and the GTX 1080 Ti was only 7+1. This time, there are up to 10 phases of power on the PCB, but I don't know how to allocate it. Because NVIDAI can not be used to fill all the pad positions, it is possible to leave the empty solder. The external power supply part is 8Pin+6Pin, and it is unlikely that the remaining 6Pin will be soldered. This can provide up to 75+75. +150=300W power, the amount of redundancy will be quite a lot.
So the above is some personal analysis of the leaked NVIDIA new generation graphics card spy photos, you can also carefully scrutinize, there are new ideas can be raised in the comment area below.