AMD Polaris 11 in shows CompuBench has 1024 Shader processors
This is a pretty interesting find, in the CompuBench database some device entries of what seems to be AMD Polaris 11 (the mainstream SKU) has appeared. It's all a bit dodgy to dig out, but each graphics hard has an identifying device number, a Device_ID. These Device_ID have appeared in the CompuBench databases and as such you can extrapolate products from that.
The most recent find it AMD Polaris 11, Device ID 67FF:C8 codenamed “Goose”. This would be the base GPU for a several entry-level products. Now, the CompuBench database reports back that this device has 16 CUs with a maximum clock frequency of 1000 MHz. Multiple your CUs (compute units) by the number of shader processors per cluster (assuming that AMD keeps 64 per cluster) and you'll notice that Polaris 11 in this configuration has 1024 Shader processors tied to a 128-bit bus and 2048 MB of memory.
Polaris 10, codenamed "Ellesmere," would then feature over 2304 stream processors (36 CUs); and Vega 10 featuring 4096 stream processors, with 64 CUs. Things could end up looking like this:
AMD Polaris / Vega GPU Specs (rumored) | |||
---|---|---|---|
April 8th, 2016 | AMD Vega 10 | AMD Polaris 10 | AMD Polaris 11 (Dev_ID 67FF) |
GPU | Vega 10 / Greenland | Polaris 10 / Ellesmere | Polaris 11 / Baffin |
Positioning | Enthusiast | High-end | Mainstream |
Fabrication Process | 14nm FinFET | 14nm FinFET | 14nm FinFET |
Compute Units | 64 | 36 | 16 |
Stream Processors | 4096 | 2304 | 1024 |
Computing Power | ~8.2 TFLOPs | ~3.7 TFLOPs | ~ 2.0 TFLOPs |
Core clock | ~1000 MHz | ~800 MHz | ~1000 MHz |
Effective Memory Clock | ~2000 MHz | ~6000 MHz | ~7000 MHz |
Memory Bus | 4096-bit | 256-bit | 128-bit |
Memory | 16GB HBM2 | 8GB GDDR5(x) | 2GB GDDR5 |
Bandwidth | 1024 GB/s | 192 / 384 GB/s | 112 GB/s |
Launch Date | Late 2016 – Q1 2017 | Q2 2016 | Q2 2016 |
Mind you that the specs shown in the CompuBench database might not be the full unlocked GPUs, so the CU numbers might even be higher. Which makes this news-item very speculative. We are inclined to say that Polaris 10 really would get 2560 shader processors.
Senior Member
Posts: 2887
Joined: 2013-03-10
Aren't those clock speeds too low? Nvidia has long been able to beat or compete with AMD's GPUs despite using less transistors (that is, cheaper chips) by having higher clocks. It seems strange to me AMD wouldn't use this opportunity to get some power for free doing the same. Unless I remember completely wrong, I seem to recall reading that shifting to 14/16 nm would allow increasing the speed.
Senior Member
Posts: 7191
Joined: 2006-09-24
Well they might be going for lower clocks. Who knows how high these might overclock if true tho.
Senior Member
Posts: 12965
Joined: 2014-07-21
I wonder if 16GB for a single GPU are true (I know they can do it, but do they really need it?)
Senior Member
Posts: 11809
Joined: 2012-07-20
It has been benchmarked with some very specially old driver... As it states CL_Driver_Version: 1956.4 (well, it at least matches time of test)
We have 2004.6
Then it has 16 compute units and only 1 tenth of Fiji performance (64CU).
It has basically parameters of HD 7850, but is still 3 times slower? Why would someone test it with old driver? Likely it is modded driver/vBIOS on HD 7850 and downclocked.
If I did same thing with mine, I can call it Vega Engineering Sample. I can set default vBIOS max clock as 600MHz, but OC it to standard 1050 or above. And world will believe that 600MHz Vega performs as well as 1050MHz Fiji.
Internet will flop over it at least twice.
Senior Member
Posts: 4804
Joined: 2005-07-08
Interesting!
So close to unveiling of next generation cards!