Today ARM fills out its graphics portfolio with the new Mali-G51 and Mali-V61 blocks. SemiAccurate thinks this GPU and video block will up the mainstream phone and VR ante in year or so when devices hit the market.
The most important of the two is the Mali-G51, the second device built on the Bifrost architecture. If you are thinking about this as the little brother to the recently unveiled G71, you would be right. It isn’t just a G71 chopped in half like the larger desktop GPUs, mobile has a different set of requirements than desktop. More importantly since G71 comes in 4-32 shader units, it is a completely different architecture that uses the same ISA as the larger variant.
You might recall that the G71 processes a 4-wide warp called a quad in one cycle. SemiAccurate’s moles are saying the G51 is more analogous to a 2-wide G71 so think of it as a warp every two clocks, or it would be if the G71 hadn’t moved to scalar processing. Instead of warps think two threads worked on per clock, the warps still remain 4-wide. G51 also adds a few instructions to the Bifrost architecture, essentially it has transcendentals in hardware now.
Now more with less, the new G51
ARM is claiming 1.6x higher efficiency than the older Mali-T830 it supplants, a number that isn’t hard to believe at all. What will be more interesting is to see the efficiency vs the G71 when silicon arrives. Because the scaling down is more than linear for area, the G51 should be less than half the size of a G71 per ‘core’ but only half as fast unit for unit. It also adds a newer version of the framebuffer compression found in G71 for a little better efficiency.
All this translates to a pretty substantial cost savings for silicon vendors looking to attack the mid-range market where cost is a big issue. Most of the performance can be made up by going wide if necessary, there will undoubtedly be performance overlap between the Bifrost devices. Think of G51 as another point on the price/performance curve more than the performance per watt curve,
Mali-V61 does better than older video blocks
Then there is the companion video block called V61 which brings VP9 and improved HEVC to the next generation of devices. The headline numbers are a 50% savings in bit-rate, presumably vs H.264 on the older V550, and up to 4K120 video output. That last number is important, it allows for 4K60 x2 for VR uses at a much lower cost point than before.
Mali-V61 core performance
V61 scales from 1-8 cores which as you can see from the above will allow fairly inexpensive 28nm parts to hit the magic 4K120 mark. When devices hit the market with V61 a little over a year from now, it will change the phone based light VR market substantially. As a bonus it also supports 10-bit/HDR data paths, something we expect all but the bottom of the market to require by the time the V61 debuts in silicon.
So that is what ARM is bringing us today, a new mainstream GPU that will put another sweet spot on the OEM curve, this one aimed at the middle of the market. To back it up we have a new scalable video block that will potentially enable mainstream 4K VR setups. Together ARM calls this duo the Mail-G51 and Mali-V61, something SemiAccurate doesn’t have a snappy name for. Punning and snark is hard work, designing chips may be harder but we still try.S|A