AMD RTG Polaris and NVIDIA Pascal specs compared

We’re very close to the launch date of first Pascal GP104 Graphics Card, that is Geforce GTX 1080 which will be launched on May 27, 2016. It appears this sudden move from NVIDIA scared AMD RTG which has been making a lot of noise with their Polaris GPU, especially in terms of efficiency. However, Pascal GP104 and Polaris 10 will occupied different price range. This means they won’t be competing each other directly.

Geforce GTX 1080 will reign the price of $600, making Geforce GTX 980 Ti looks no longer attractive, Radeon R9 Fury X looks dumb, and Radeon Pro Duo looks like an overpriced junk. The second GP104-based Graphics Card, Geforce GTX 1070 will reign the price of $380, making Geforce GTX 980 looks pathetic and Radeon R9 390X looks idiot. On AMD side, Polaris 10 will try to gain control of sub $300 price to compete with Geforce GTX 970, not sure if it could really beat the nearly two-year-old Geforce GTX 970 when it comes to performance. Polaris 11 will be on the sub $200 market for low cost gamers.

Here is the table of the specs comparison.

Manufacturer NVIDIA AMD AMD
GPU GP104 Polaris 10 Polaris 11
Unified Shaders 2560 2048 896
Texture Mapping Units 160 128(?) 56(?)
Render Output Units 64 32(?) 14(?)
Processing Power ~9TFLOPS ~5.5FLOPS ~2.5TFLOPS
Memory Interface 256 bit 256 bit 128 bit
Memory Type GDDR5X
TDP 180 Watts 150 Watts 50 Watts

Let’s visit Tahiti and Tonga again. The first Tahiti-based Graphics Card, Radeon HD 7970 has processor power of 4 TFLOPS. And comes the full fledged Tonga with the same amount of stream processor and it has 3.9 TFLOPS compute power which actually a bit slower. However, the report from WCCFTech which is based on VideoCardz which is also based on TechPowerUp report mentioned that Polaris 10 will have 5.5 TFLOPS compute power. How the hell is AMD going to achieve that? One way to boost that is via Clock Speed. This means the GPU will be clocked 35-40% faster than Tahiti or Tonga which is already clocked around 1 GHz. So yeah, guess that the GPU will be clocked around 1.35 GHz or perhaps higher. The TDP is mentioned to be around 150 Watts. So what’s the news? Polaris 10 will feature up to 2048 steam processor. The same amount of stream processor found in more than 4-year-old Tahiti GPU and 2-year-old Tonga GPU. Not sure why AMD really loves to built GPU with 2048 stream processors but Tonga is no faster (actually slightly slower) than Tahiti despite using a 3rd generation GCN architecture while Tahiti is using 1st Generation GCN. I also doubt 4th Generation GCN will offer faster performance per core compared to the older GCN.

Now move into Polaris 11 GPU. This small GPU is more exciting than Polaris 10. It features only 896 Stream Procssors and the memory interface is only 128 bit. This will be the successor of Bonaire GPU. The so not exciting thing with this GPU is 128 bit memory interface. Bonaire has 104 GB/s memory bandwidth with its 6500 MHz GDDR5 memory and that is probably too low. Let’s just hope the compression can save the day and prevent bottleneck on the memory bandwidth. On the exciting side, this GPU will feature 2.5 TFLOPS performance which is faster than PS4 GPU while keeping the power consumption really low! I think the graphics card with this GPU will be priced around S150 which is a great value for low cost gaming system.

Anyway, back to the not so exciting Polaris 10 again. While 37% boost in compute power (compared to Tahiti and Tonga) might sounds great, it’s not that actually exciting. Hawaii already achieve higher computer than that at the cost of high power consumption and terrible heat (for those using reference design). So, Polaris 10 might only ended up as fast as Radeon R9 290X. And guess what? Geforce GTX 970 is about as fast as Radeon R9 290X. The TDP of Geforce GTX 970 is 145 watts only and that is close to Polaris 10 TDP. So, there is real efficiency gain here compared to NVIDIA 28nm GPU. On the other side, Geforce GTX 1070 that might cost less than $100 more, will be significantly faster and feature the same TDP. This just means Polaris 10 is not actually that efficient (for a 14nm GPU) and still behind GP104 efficiency. But it’s still early to take that as a conclusion. Let’s just wait until some real numbers appear. For now, let’s just don’t get too excited over Polaris 10.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s