Radeon R7 200:  Radeon R7 240 |  R7 250 |  R7 250X |  R7 260 |  R7 265 |  Radeon R9 200:  Radeon R9 270 |  R9 270X |  R9 280 |  R9 280X |  R9 290 |  R9 290X |  R9 295X2
GeForce 700:  GeForce GT 730  |  GT 740  |  GTX 750  |  GTX 750 Ti  |  GTX 760  |  GTX 770  |  GTX 780 |  GTX 780 TI |  GeForce TITAN:  GeForce GTX TITAN |  GTX TITAN Black |  GTX TITAN Z

Feb 12th, 2014

Exclusive: NVIDIA Maxwell GM107 architecture unveiled

This is going to be a very interesting post. Forget everything you heard about the GM107, it was simply not accurate.

The facts about Maxwell

NVIDIA Maxwell GM107 specifications

As it turns out, the 768/960 CUDA count as read by GPU-z was wrong. The application could not read the real number because the architecture has been changed compared to Kepler. GPU-Z assumed that each SMX holds 192 CUDA cores, which was not correct. The real number is 128. Full GM107 has 5 SMs with 128 CUDAs each, which gives 640 in total. This means that GTX 750 Ti has 640 CUDAs (5 SMX) and GTX 750 has 512 (4 SMX).

GM107 has a TDP of 60W

The GM107 will not even utilize the full power delivered by PCI-E connector (75W). While operating at default frequencies it won’t need any additional power source. Although manufacturers will still add the power connector, for the sake of stability or increasing the overclocking headroom.

Maxwell is said to deliver twice the performance per watt of Kepler.

There are two generations of Maxwell GPUs

It has been revealed by newest CUDA driver that NVIDIA will make two generations of Maxwell GPUs:

  • 28nm: GM108, GM107
  • 20nm: GM206, GM204, GM200

Maxwell will kick off with GM107, but this is not the only Maxwell based on 28nm process. The GM108 will follow it later in GeForce GT 840M. Looking even further we shall see GM200 parts, most likely based on 20nm process, that would be real high-end segment replacement. At this point we don’t know when 20nm process will arrive.

GM107 will replace GK107 with a performance of GeForce GTX 480

You should find this particularly interesting. While GM107 utilizes 4 times less power than Fermi GF100, it will offer the same performance (actually even slightly better).

Meet the SMM (Maxwell Streaming Multiprocessor)

Maxwell introduces new architecture, which is not exactly revolutionary, but definitely new. The streaming multiprocessor known from Fermi and Kepler received the biggest change. Those who thought that Maxwell architecture will introduce more CUDAs in each multiprocessor will be disappointed. In fact, each SMM is split into four blocks of operating units. Each of those blocks holds 32 CUDA cores, so the full SMM has 4 × 32 = 128 CUDAs. You can find here references to both Fermi and Kepler.

The GM107 diagram

This is the first ever diagram of Maxwell GPU. There is one Graphics Processing Cluster. Full GM107 has 5 SMMs with 640 cores in total. Each SMM has 8 TMUs, so the total count is 40. The GPU has 16 ROPs and two memory controllers (both 64b).

Maxwell GM107

What is new in Maxwell

Okay, so the SM has been redesigned, the layout has changed, but what exactly makes the Maxwell better? Well here’s the list:

  • Larger L2 cache.
    This is the main difference between Kepler and Maxwell. Larger L2 cache will limit the queries to the GPU. GM107 L2 cache has 2MB. GK107′s cache has 256KB.
  • Workload balancing and complier-based scheduling has been improved.
  • The number of instructions per clock cycle has been increased.
  • SM has been redesigned into four processing blocks (as explained above).
  • Maxwell introduces even faster H.264 encoding and decoding with improved NVENC (which is used, for instance, in ShadowPlay).
  • New GC5 power state (low sleep state).

GM107 has a die size of 148mm2

As opposed to previous leaks, the die size of GM107 is even smaller, not 156 but 148mm2. Compared to GK107 the density of CUDAs per mm2 has increased roughly by 30%. The density of transistors increased by 15%. Remember, this is all on the same fabrication process.

This is all I have for you, I hope this explains everything you need to know.

First Maxwell GPUs will be released on February 18th.