⋮  

NVIDIA’s Full Fat Pascal GP104 GPU Pictured With GDDR5X Memory – Features 8 GB Micron Chips To Drive GeForce GTX 1080 Cards

Hassan Mujtaba
Posted Apr 23, 2016
204Shares
Share Tweet Submit

NVIDIA’s next-gen Pascal GP104 has once again been leaked but this time we are looking at the full fat version which will be featured on the fastest GeForce graphics card that will be available to consumers in the coming months. The latest Pascal GP104 GPUs will be built on TSMC’s industry leading 16nm FinFET process which will allow NVIDIA to drastically increase the transistor count of their GPUs while retaining a small and efficient die package.

NVIDIA’s Fully Enabled Pascal GP104-400-A1 GPU is finally leaked.

NVIDIA’s Fully Enabled Pascal GP104 GPU Features Micron’s Next-Gen GDDR5X Memory

Yesterday, we got our first detailed look at the Pascal GP104 GPU itself. The Pascal GP104 GPU which was seen yesterday was a qualification sample (QS) which indicates that the chip is no longer in engineering state. Same is true for the latest picture which is also a QS chip and means that graphics cards based on both GPUs will be unveiled by NVIDIA and their AIBs relatively soon.

A block diagram of NVIDIA’s Pascal SM (Streaming Multiprocessor) Design.

The first thing we note is that the GPU pictured yesterday had the codename GP104-200-A1 while the GPU that has been leaked today is codenamed GP104-400-A1. Comparing these codenames to the list of GP104 SKU that leaked earlier, we can note that the GP104-400-A1 variant is the fully enabled variant which doesn’t comes with any disabled SM (Streaming Multiprocessors) units and features the full specifications available on the GP104 GPU.

NVIDIA Pascal GP104 Features Support For Both GDDR5X and GDDR5 Memory – First GDDR5X Chips Pictured

The other major thing about this leak is that we are getting our first sight at GDDR5X memory. Yes, you heard me right. This is the first graphics board featuring the Micron GDDR5X memory that have the codename listed as 6GA77 Z9TXT. The card features 8 such chips on the board which means that GPU features the same 256-bit bus interface as its GDDR5 based sibling, the GP104-200-A1. This indicates that NVIDIA’s Pascal GPUs will come with support for GDDR5, GDDR5X and even HBM2 memory (GP100).

Micron’s GDDR5X memory is featured on the GP104-400-A1 graphics board.

You can see in the pictures below that the Micron memory, even though it has a larger ball grid compared to GDDR5, has a compact design. Each GDDR5X DRAM measures at 14x10mm compared to 14x12mm that is the size of a normal GDDR5 DRAM. The DRAMs feature the same Micron logos and labels at the  positions where the official markings should be. The codename however is not listed under the product catalog which could be due to the fact that the memory used on the board are still samples but ready to use. Micron commenced the sampling for their GDDR5X memory last month.

Rise Of The Tomb Raider DX12 ASync Compute AMD Performance Improved; NVIDIA Users Better Off Using DX11

We do note that the board is of a very different design compared to the previous PCB which was used for the GP104-200 SKU hence it confirms that GDDR5X and GDDR5 aren’t pin-to-pin compatible and require different boards to be manufactured to support them even if the GPU support both GDDR5 and GDDR5X memory type.

NVIDIA’s Pascal GP104 GPUs Will Feature Faster and Denser GDDR5/GDDR5X Memory Modules!

The GDDR5 DRAM uses a 170 pin ball grid while GDDR5X DRAM uses a 190 pin ball grid layout. Both cards feature the same amount of VRAM of 8 GB but the GTX 1080 will have higher bandwidth available of up to 384 GB/s while retaining a 256-bit memory bus.

The NVIDIA Pascal GP104-200-A1 chip was pictured with GDDR5 memory.

The GeForce GTX 1070 on the other hand can go as high as 256 GB /s due to the faster GDDR5 chips being used on the graphics board (SK Hynix 8 GHz). Following is a list of SKUs which are allegedly reported to be based on GP104 design:

NVIDIA Pascal GP104 SKUs:

Maxwell Graphics CardMaxwell GPU SKUPascal GPU SKUBoardsAvailability
NVIDIA GeForce GTX 980 TiGM200-310-A1GP104-400-A1Reference and AIBEarly June
NVIDIA GeForce GTX 980GM204-400-A1GP104-200-A1Reference and AIBEarly June
NVIDIA GeForce GTX 970GM204-200-A1GP104-150-A1AIB OnlyMid June

NVIDIA’s GP104 Measures at 317mm2 and Houses Around 8 Billion Transistors – Powers The GeForce GTX 1080 Graphics Card

Coming to the talk on the die, the GP104 measures at approximately 317mm2 which is smaller than both the GM200 and GM204 GPUs. Just to put things in perspective, NVIDIA’s current Titan X has 8 Billion transistors on a 601mm2 die. The GeForce GTX 980 with GM204 on the other hand is a 398mm2 die with a total of 5.2 billion transistors.

FinFET really proves to be a miracle when it comes to increasing the overall transistor numbers and at the same time, decreasing the chip size. This allows more chips to be manufactured on a 16nm Wafer that saves cost for NVIDIA developing such GPUs.

Gigabyte Reveals Battlefield 1 Benchmarks - Nvidia Takes The Lead, No Single Graphics Card Can Hit The 4K 60 Standard

NVIDIA GP104 400 and GP104 200 SKU Dimensions:

NVIDIA’s Pascal GPUs are gearing to be the same revolution which was their Maxwell GPUs at the time of launch. The Maxwell GPUs launched back in 2014 and allowed NVIDIA to capture up to 80% of the discrete graphics market share. Competitive pricing on Pascal cards will allow many current and last generation GPU owners to upgrade to the Pascal GeForce cards which will deliver performance and efficiency increases along with better tech support for next-generation AAA titles.

NVIDIA Pascal GP102 GTX Titan X Specifications:

Graphics Card NameNVIDIA GeForce GTX 980NVIDIA GeForce GTX 980 TiNVIDIA GeForce GTX Titan XNVIDIA GeForce GTX 1070NVIDIA GeForce GTX 1080NVIDIA GeForce GTX Titan (Pascal)
Graphics CoreGM204GM200GM200GP104GP104GP102
Process Node28nm28nm28nm16nm FinFET16nm FinFET16nm FinFET
Transistors5.2 Billion8 Billion8 Billion7.2 Billion7.2 Billion12.0 Billion
CUDA Cores2048 CUDA Cores2816 CUDA Cores3072 CUDA Cores1920 CUDA Cores2560 CUDA Cores3584 CUDA Cores
Base Clock1126 MHz1000 MHz1000 MHz1506 MHz1607 MHz1417 MHz
Boost Clock1216 MHz1075 MHz1075 MHz1683 MHz1733 MHz1531 MHz
FP32 Compute5.6 TFLOPs6.5 TFLOPs7.0 TFLOPs6.5 TFLOPs9.0 TFLOPs10.1 TFLOPs
VRAM4 GB GDDR56 GB GDDR512 GB GDDR58 GB GDDR58 GB GDDR5X12 GB GDDR5X
Bus Interface256-bit bus384-bit bus384-bit bus256-bit bus256-bit bus384-bit bus
Power Connector6+6 Pin Power8+6 Pin Power8+6 Pin PowerSingle 8-Pin PowerSingle 8-Pin Power8+6 Pin Power
TDP165W250W250W150W180W250W
Display Outputs3x Display Port
1x HDMI 2.0
1x DVI
3x Display Port
1x HDMI 2.0
1x DVI
3x Display Port
1x HDMI 2.0
1x DVI
3x Display Port 1.4
1x HDMI 2.0b
1x DVI
3x Display Port 1.4
1x HDMI 2.0b
1x DVI
3x Display Port 1.4
1x HDMI 2.0b
1x DVI
Launch DateSeptember 2014May 2015March 201510th June 201627th May 2016August 2016
Launch Price$549 US$649 US$999 US$379 US$599 US$1200 US

[socialpoll id=”2353542″]

Share Tweet Submit