Submitted by Shortly after Windows 11 launch, AMD and Microsoft jointly discovered that Windows 11 is poorly optimized for AMD Ryzen processors, which see significantly increased L3 cache latency, and the UEFI-CPPC2 (preferred cores mechanism) rendered not working. In our own testing, a Ryzen 7 2700X “Pinnacle Ridge” processor, which typically posts an L3 cache latency of 10 ns, was tested to show a latency of 17 ns. This was made much worse with the October 12 “patch Tuesday” update, driving up the latency to 31.9 ns.
That’s one hell of a regression. It seems fixes are incoming soon, though.
AMD putting out the fix so it was an AMD issue. Windows often takes the rap for 3d party drivers.
and rightfully so that Windows takes the rap so often when Microsoft has such a piss poor development process.
Adurbe,
There’s no shame in being beta testers for linux 🙂
j/k
On a serious note it makes me wonder what the updates contained that they couldn’t be rolled back immediately and try again in some future update. If it wasn’t a security critical flaw, couldn’t they just roll back the L3 performance killing patches?
Uhhh no, sorry. Its been reported on several YT channels (Gamers Nexus, LTT, Hardware Unboxed just to name a few) that Windows 11 might as well be called “The Wintel edition” as the devs have been focused on NOTHING but trying to get the mess that is the new Intel Big.Little design to work and its breaking stuff left and right.
For those that haven’t heard instead of ya know, actually competing with AMD by building a new design that isn’t a giant power pig so they can put more cores like AMD without being a Netburst style space heater Intel is going to a Big.Little or as Intel is calling it P.E with P being the “Power”cores and the E being the “efficient” (AKA shite Intel Atom) cores and its gonna require a VERY VERY smart scheduler as its gonna have to understand 1.- Which jobs need to be on which cores, 2.- When it should use P or E cores for a specific task, and 3.- When to switch a task from a P core to E core or vice versa…and its gonna have to do all that in the nanosecond range to keep it from being a laggy mess…..sigh.
So yeah expect Windows 11 to follow the “every other version sucks” rule as I seriously doubt MSFT is gonna get something that complex working anytime soon, hell they never did figure out how to use the AMD FX chips correctly (Just look at the Linux VS Windows scores on AMD FX, its like night and day as Linux coders actually figured out how to use the FX design properly, MSFT? Not so much) and this is a HELL of a lot more complex than the shared FP that FX had. Oh and before someone brings up AMD has their own Big.Little coming out for laptops? Not even slightly similar as the AMD chips are using the cores from the XBone and PS5 for the lower power chips so MSFT has had years to figure them out and they are a hell of a lot more powerful than an Intel Atom core so its not gonna cause a system to lag if it takes a few seconds to hand off from one chip to the other, anyone who has tried doing…well anything on an Intel Atom knows how damn painfully slow those chips are even in 2021.
Since a 1970’s PDP-11 could read core memory in 400ns, that means that the Ryzen 7’s L3 cache latency is roughly 10 times faster than a 50 year old computer’s memory cycle.
I understand there’s an apples to oranges comparison here, with bandwidth being many orders of magnitude faster than the PDP-11. But it’s still interesting to see just how far we haven’t come.
Processors used to be way slower (<1MHz) than memory in that era (SRAM). Then processors improved ten folds but memory lagged behind (DRAM). Despite the impressive speed of DDR6, it's still way slower than L1 and L2 cache speed. But way larger too.
Anyway, I wouldn't return back on these, I cherish the raw number crunching of today's CPU. Be the RAM considered slow or not.
Cache memories operate differently than RAM, since those 10ns include the tagging, associativity, etc.