JUN 2022MAXIMU MPC 11
Jarred Walton
TECH TALK
Jarred Walton has been a
PC and gaming enthusiast
for over 30 years.© AMD
AMD 3D V-Cache Boosts
Gaming Performance
L2 caches have grown from around 128KB per core
in the 90s to today’s 512KB to 1MB, sometimes even
2MB, per core. But it’s the L3 cache sizes that have
really ballooned. Intel’s various mainstream Core
processors have included 8MB of L3 cache going
back to the Bloomfield and Lynnfield chips in the
late 2000s. In contrast, AMD chose to put 16MB of
L3 cache (per eight cores) on its first- and second-
generation Ryzen processors, then doubled that to
32MB of L3 (still per eight cores) on the third- and
fourth-generation CPUs, Zen 2 and Zen 3. One big
difference between them was that Zen 3 made it
a unified L3 cache, instead of a split 16MB+16MB
cache, which reduced latencies and improved
performance. But AMD had bigger plans.
From the start, Zen 3’s Core Complex Die (CCD)
has had Through Silicon Vias (TSVs) that were
intended to link up with a larger stacked cache. It
was likely intended for the data center EPYC chips,
but AMD has also released a consumer Ryzen 7
5800X3D with 64MB of stacked 3D V-Cache. AMD is
becoming the cache company, using large caches to
boost performance. The current RDNA 2 graphics
architecture took the previous design and added
up to 128MB Infinity Cache, which alleviated a lot
of GDDR6 memory traffic and allowed a boost in
performance without needing much more memory
bandwidth—unlike Nvidia’s use of GDDR6X.
The cache chiplet sits atop the existing L3 cache
structures of the Zen 3 CCD, with a silicon shim
above the hotter running CPU cores. Stacking aWE’VE KNOWN FOR AGES that large caches can help improve performance,
though larger caches can also increase latency and counteract those
gains. It’s a balancing act, and L1 caches have been sitting in the 32KB to
96KB range for quite some time now.
large cache on top of
an existing design
involves tradeoffs.
The 5800X3D has
lower base and
boost clocks than the
5800X, and the extra layer
of silicon on the main die reduces
the ability to dissipate heat. AMD
has disabled overclocking on the
5800X3D as a result, and the CPU
cores get hotter while using less
power and running lower clocks.
Large CPU caches mostly help
in applications that use a lot of
memory bandwidth—if most of the
code and data can fit inside the L
and L2 caches, a massive L3 cache
won’t do much. While the 5800X3D
shows improved performance in
multi-threaded workloads like
y-cruncher and 7-zip compression,
the biggest beneficiary is games.
AMD claims an average boost
of around 15 percent, but in our
test suite, we measured a 28
percent improvement in average
frame rates at 1080p compared
to the Ryzen 7 5800X. Not every
game will show such huge gains,
and the difference between the
fastest CPUs when running higher
resolutions becomes negligible.
However, faster GPUs are coming,
which should spread things out.
Even with those potential
concerns, tripling the L3 cache
clearly pays dividends. Technically,
the Ryzen 7 5800X3D now reigns asthe overall fastest CPU for gaming.
It’s an incredible comeback when
you consider Intel’s Core i9-
12900K has more cores and uses
up to 85 percent more power. It
held a 17 percent lead in gaming
performance over the 5800X and
was 11 percent faster than AMD’s
previous best gaming chip, the
Ryzen 9 5900X. AMD has turned
the tables, delivering 10 percent
higher gaming performance than
the 12900K, thanks to the big cache.
Intel still has better single-
threaded and multi-threaded
application performance, which is
something AMD will look to combat
with its Zen 4 architecture. Given
what we’ve seen with the 5800X3D,
it’s a safe bet we’ll see more Ryzen
processors with 3D V-Cache. Bring
on the Ryzen 9 6900X3D.
Technically, the Ryzen 7
5800X3D now reigns as the
overall fastest CPU for gaming.AMD stacks
an extra 64MB of L3 cache on top
of a Zen 3 chiplet for a massive
boost in gaming performance.