I managed to snag one of the 3950X on launch day and replaced my previous 2700X. Something I was not expecting to find in upgrading was that my GTX1080 cannot keep up with it on 4K BRAW timeline render to H264 in Resolve with mild grading. During a render it fully loads around 20-24 CPU threads with the GPU bouncing off 100% load throughout. It's still nice to have the system be fully responsive and able to multi task during the render but that seems to be the limiting factor on my render speed now.

I'm running it on an Arctic Liquid Freezer II 240 AIO and it is very nice to have replaced the jet engine that used to sit beside me ramping up and down all day long (Wraith Prism).

Looking at reviews you see very hard hand of AMD, as maximally pushed are benchmarks and apps that do not require much memory performance and instead require maximal load on floating point operations, but preferably without much AVX2 (as CPU drops speed like mad in this mode).

Of course, AMD also recommended that memory to use - preferable DDR4 3600 CL14 or better set (can be from 50 to 250% more than price of usual 3200 CL16 set).

Chiplet under microscope

https://www.flickr.com/photos/130561288@N04/49046159127/

AMD Ryzen developer and author of DRAM Calculator for Ryzen, has revealed a new power plan that nets an average increase of 200-250MHz on AMD Ryzen 3000 series processors.

According to the developer, the mod will work on all Zen 2 based processors including the upcoming Threadripper series.

https://www.techpowerup.com/review/1usmus-custom-power-plan-for-ryzen-3000-zen-2-processors/2.html

AMD started slow marketing preparation to 16 core release

Special leaks show that Ryzen 9 3950X is only 10% faster compared to Ryzen Threadripper 2950X in Geekbench 5.

In real app with good memory load difference will be around 2-3%.

https://www.fudzilla.com/news/pc-hardware/49503-rome-voltage-is-high?cid=dlvr.it

@leonbeas

It first leaked as AMD reduced actual boost frequency with new BIOS releases.
After big public noise they returned most of the initial boost, but reason of their initial behavior can be only the fast degradation of chiplets.

Note that 7nm transistor is close to practical limits for such structure (and is not 7nm :-) ), also as you boost voltage and work on high temperatures such small elements degrade much faster and have much more internal leakage. It is good reason why Intel couldn't match 32nm Sandy Bridge overclocking frequency until the last 14nm generation, as 14nm leaks are much higher. If companies made 32nm chiplets using all the improvements used in last processes we had 5.1-5.2Ghz chips.

In reality 22nm-32nm are ideal for high performance CPUs where you can use water cooling, but you need smaller processes if you want big cache memory. So, ideal is the mixed process, where half of chiplet is being made using 22-32nm and half is made using high density 5-7nm and it is area where SRAM memory and some IO can reside.

Not only AMD can't keep doing 3950X on time. But they also can't supply enough 3900X even.

Also note that it is till rumors of internal talks in AMD about fast degradation of 7nm chiplets. CPU life for current 7nm process can be limited by 2-4 years until significant max frequency drop.

16 core model postponed

Actual 7nm yields are still bad, it seems, with lot of defects and very rare all 8 cores on chiplet can work.

The original limits for Ryzen 3000 SKUs were:

• 3600 = 4100MHz (80-95°C) / 4200MHz (< 80°C)
• 3600X = 4200MHz (80-95°C) / 4400MHz (< 80°C)
• 3700X = 4200MHz (80-95°C) / 4400MHz (< 80°C)
• 3800X = 4300MHz (80-95°C) / 4550MHz (< 80°C)
• 3900X = 4400MHz (80-95°C) / 4650MHz (< 80°C)

Since then, it appears that the HighTemperature limit has been reduced further to 75°C (from 80°C). New SMUs also have introduced "MiddleTemperature" limit, but that gets disabled when PBO is enabled.

There's no doubt that the 'reviewer BIOS' is faster and sustains higher boost clocks for longer than the latest releases. Many chips aren't reaching their full boost potential even with the N11 'reviewer BIOS,' so AMD's fix might consist of returning to the original hard 80C boost temperature limit, which we're told hasn't been officially seen in the wild.

Boost statistics for single core

Ryzen 5 3500 coming soon

• Disabled HT, so 6 cores / 6 threads, purely marketing requirement to segment offers
• 3.6 GHz base clock
• 4.1 GHz max boost
• chiplets with worst binning
• 65W TDP (it is lie, of course)
• cost to manufacture of such complete CPU to AMD is around $20-25 :-)