On September 19th, NVIDIA will finally reveal its new GeForce 900 series.
Two cards will be announced during a special event for gamers (Game24): GTX 980 and GTX 970.
While the GTX 970 is just a gap-filler, GeForce GTX 980 is much more – a proper new flagship card.
Introducing Maxwell GM204
First, let’s have a look at the new processor, the GM204. This is the first GPU based on ‘Second Generation Maxwell‘ architecture. When GM107 was announced, I think everyone was expecting GM2xx architecture to use smaller node, but unfortunately, we are still on 28nm fabrication node.
GeForce GTX 980 is equipped with full fat GM204, the 400 variant. This GPU has 16 Streaming Multiprocessors “Maxwell” (SMM). What you didn’t know, is that GM204 has 64 Raster Operating Units. Yes, GM204 has 64 ROPs, that’s more than Kepler GK110.
The GM204 has in fact, the same number of Texture Mapping Units (TMUs) — 128, but thanks to higher core clock, texture fillrate is actually higher.
There are 5.2 billion transistors packed into GM204, and the GPU has a die size of 398 mm2.
And for those wondering how large L2 Cache is, then the answer is: the same as GM107 — 2MB.
NVIDIA Maxwell GM204 has some cool features like Dynamic Super Resolution, which is basically a new upscaling technology advertised as ‘4k Quality on a 1080p display’.
Other interesting technology is Third Generation Delta Color Compression. This technology will help increase memory efficiency, which is definitely required with 256b interface.
The final buzzword for you is Multi-Pixel Programmable Sampling, technology improving sample randomization, and reducing quantization artifacts.
GM204 die shot and block diagram
GM204 is split into four Graphics Processing Clusters (GPCs). Each GPC has 4 SMMs. There are four 64b memory controllers. Large L2 Cache is accompanied by 64 ROPs.
NVIDIA GeForce GTX 980 Final Specifications
GeForce GTX 980 has 2048 CUDA cores, 128 TMUs and 64 ROPs. Card is equipped with 4GB GDDR5 memory and 256-bit interface. It has a bandwidth of GTX 770, which is 224 GB/s. By comparing it to Kepler parts, we notice that Maxwell-based graphics cards arrive with relatively higher clock speeds, GTX 980 has a base clock of 1126 MHz and boost clock of 1216 MHz.
The biggest news here is that GTX 980 has only a TDP of 165W. That’s amazing power reduction compared to 250W GK110.
Last but not least, I can now confirm that GeForce GTX 980 has HDMI 2.0 support. And if you somehow missed my previous news, GTX 980 has 5 display outputs: DVI-I, HDMI2.0 and three DisplayPorts 1.2 (not 1.3).
NVIDIA GeForce GTX 980 will be released on September 19th. Card is expected to cost around $599 USD (unconfirmed).
| NVIDIA GeForce GTX 980 Specifications | |||
|---|---|---|---|
| GeForce GTX 980 | GeForce GTX 770 | GeForce GTX 780 Ti | |
| Architecture | 28nm Maxwell | 28nm Kepler | 28nm Kepler |
| GPU Codename | GM204-400 | GK104-425 | GK110-425 |
| Die Size | 398 mm2 | 294 mm2 | 561 mm2 |
| L2 Cache | 2 MB | 512 kB | 1.5 MB |
| Transistors | 5.2b | 3.54b | 7.08b |
| CUDA Cores | 2048 | 1536 | 2880 |
| TMUs | 128 | 128 | 240 |
| ROPs | 64 | 32 | 48 |
| Base Clock | 1126 MHz | 1046 MHz | 875 MHz |
| Boost Clock | 1216 MHz | 1085 MHz | 928 MHz |
| Memory Clock | 1750 MHz | 1750 MHz | 1750 MHz |
| Memory | 4GB GDDR5 | 2GB GDDR5 | 3GB GDDR5 |
| Memory Bus | 256-bit | 256-bit | 384-bit |
| Bandwidth | 224 GB/s | 224 GB/s | 336 GB/s |
| FP Performance (SP) | 4.6 TFLOPs | 3.2 TFLOPs | 5.1 TFLOPs |
| Pixel Fillrate | 72.1 GP/s | 33.5 GP/s | 53.3 GP/s |
| Texture Fillrate | 144 GT/s | 134 GT/s | 213 GT/s |
| Power Connectors | 6pin + 6pin | 6pin + 8pin | 6pin + 8pin |
| Thermal Design Power | 165W | 230W | 250W |
| Launch Date | Sep 19th, 2014 | May 30th, 2013 | Nov 7th, 2013 |
NVIDIA GeForce GTX 980 Pictures
