MAR 2022MAXIMU MPC 11
Jarred Walton
TECH TALK
Jarred Walton has been a
PC and gaming enthusiast
for over 30 years.© AMD
AMD’s RDNA2-Powered
Rembrandt APU
Now, with AMD’s upcoming Rembrandt Ryzen
6000-series processors, consoles might get some
competition. Make no mistake, consoles are still a
big step up in overall performance, but compared
with the previous generation Renoir (Ryzen
4000G) and Cezanne (Ryzen 5000G) processors,
Rembrandt will double the level of performance
we’ve come to expect from integrated graphics.
The Ryzen 7 5700G still posts some respectable
performance in games. It’s not sufficient for 1080p
medium at 60fps, but it could handle 720p gaming.
It was also about 50–100 percent faster than Intel’s
Xe-LP found in the Tiger Lake laptops (which is
about twice as fast as desktop UHD 770 Graphics).
But with Rembrandt, we’re at the point where
integrated graphics won’t completely suck.
First, AMD has boosted the maximum number of
compute units (CUs) from eight to 12. That should
provide 50 percent more performance. Clock
speeds on the GPU cores have been increased from
a maximum of 2.0GHz on the 5700G to 2.4GHz with
Rembrandt, another 20 percent increase. The CUs
now use AMD’s latest RDNA2 architecture and, at
least on dedicated graphics cards, that’s about 50
percent faster than the previous Vega architecture.
Architectural improvements include L2 cache
sizes that are now twice as large and double the
number of render backends. With DDR5 memory,
potentially with support for speeds up to DDR5-
4800, memory bandwidth is 50 percent higher.
When comparing the upcoming Ryzen 7 6800U
to the existing Ryzen 7 5800U, AMD showed a
consistent doubling of performance—it wasn’t an
entirely fair comparison, as the 5800U was running
in a 15W TDP configuration while the 6800U used
a 28W TDP. AMD also demonstrated superiorIF YOU WERE TO ASK ME for the best integrated
graphics solution that’s ever been created, until
recently, the answer would have been easy—the
Xbox Series X. Powerful graphics, a fast CPU, and
tons of memory bandwidth to go around. In the
world of PCs, there’s nothing even remotely close.
performance with the 6800U
compared to an i7-1165G7 laptop
and Nvidia’s GeForce MX
dedicated graphics solution.
The Zen 3+ architecture includes
improvements, like clock speeds
of up to 5.0GHz, enhanced power
management features, and deep
sleep states. More importantly, the
Rembrandt chip will be fabricated
on TSMC’s N6 process node. In
many ways, it’s the same as N
but supports more EUV layers and
promises higher logic density and
better performance per watt, while
remaining design-compatible with
the existing N7 node.
The first Rembrandt chips
should arrive in a few months, but
there is one potential concern.
DDR5 memory prices are currently
higher than DDR4 prices, which
could negate many advantages
of using fast integrated graphics
rather than a dedicated GPU. It also
looks as though Rembrandt will
be limited to DDR5 or LPDDR5—
there’s no support for DDR4, unlike
Intel’s Alder Lake CPUs.
There’s another integrated
graphics newcomer that makes
AMD’s 12 CUs and 768 GPU cores
look tame. Apple’s M1 processor
already supported up to 1024 GPU
cores back in late 2020. Apple
doubled down with the M1 Pro thatsupports up to 2048 GPU cores,
and the M1 Max doubles that again
to a maximum of 4096 GPU cores.
Combined with a 512-bit memory
interface and LPDDR5 memory,
the M1 Max has up to 400GB/s of
bandwidth. But the M1 Max has 57
billion transistors as well—over
four times as many as Rembrandt.
It’s a shame no one is willing to
create a console-style integrated
GPU for PCs. The Xbox Series X
uses 15.3bn transistors and has 52
RDNA2 CUs, with a 320-bit GDDR
memory subsystem. It’s clear AMD
could do more than Rembrandt’s
12 CU GPU, but there are other
factors at play. Maybe when AMD
moves to 5nm, we’ll see integrated
graphics with 20 or more CUs.AMD’s Ryzen 6000-series
processors pack up to eight Zen 3+
CPU cores and 12 RDNA2 CUs.