News [Tom's Hardware] Silicon Lottery Reveals AMD Ryzen 3000 AVX2 Binning Stats

[ao_ika_red]

I know this is probably more suitable if posted in CPU sub-forum but I believe this news will gather more interest here.

- directly pasted from Tom's Hardware -

Silicon Lottery not only sells binned processors, but it's also a great source of information for binning statistics. The company has recently added the stats for AMD's Ryzen 3000-series processors to its database.

The AVX2 frequency is the highest stable all-core speed the processor is able to achieve under AVX2 workloads. Silicon Lottery emphasizes that the AVX2 workloads are comparable to ones from Intel LINPACK and Prime95 version 28.9 and later. Unfortunately, the company didn't mention the sample size for each Matisse chip.



Starting with the AMD Ryzen 9 3900X, Silicon Lottery's data show that only 6% of its samples were able to hit 4.20 GHz on all 12 cores with a voltage of 1.250V. Needless to say, this is a pretty disappointing figure for performance seekers who love overclocking their processors for extra performance. On the bright side, 35% of the tested Ryzen 9 3900X chips could do 4.15 GHz on 1.237V.

The numbers start to look more encouraging as we drop down to the Ryzen 7 models. Silicon Lottery notes that 20% of the AMD Ryzen 7 3800X samples were able to achieve an all-core boost of 4.30 GHz with a 1.300V Vcore. More than half hit the 4.25 mark GHz at 1.287V. Surprisingly, the Ryzen 7 3700X shows lower overclocking potential. Only 21% of Ryzen 7 3700X parts got to 4.15 GHz on 1.262V.

In conclusion, Silicon Lottery's statistics practically confirm what AMD has previously insinuated: that Matisse's performance has been optimized to the teeth, and there really isn't much manual overclocking headroom left. If you think it was hard to win the Silicon Lottery before, it's even harder now with Matisse. Barring you can get your hands on a ton of Matisse parts, like Silicon Lottery or Caseking, the chances of you finding a very superior chip is extremely slim.

- end of article -

 

[crashtech]

Overclocking has become much less attractive to me since transitioning to 24/7 max loading. The latest AMD offerings are attractive because they maintain good clock frequencies at moderate power levels, making them the current throughput per kWh champ in most instances, imo.

 

[StefanR5R]

Interesting. If Silicon Lottery's sample size was large enough to be significant, it shows that there is real binning on AMD's part behind the different SKUs. I.e. they don't simply set some different jumper switches on the PCB, put the different stickers on the boxes, and call it a day.

I do wonder how extensive their binning process for the various server SKUs is.

 

[ao_ika_red]

Overclocking has become much less attractive to me since transitioning to 24/7 max loading.

Actually, I just want to show how high Ryzen 3000 clocks on AVX2 because prior this we used to have special clock discussion on AVX2 intensive project.

I do wonder how extensive their binning process for the various server SKUs is.

I imagine you're getting interested in Epyc Rome platform then. Unfortunately, we won't see any cheap, used Epyc Rome chips in near future.

 

[Assimilator1]

Overclocking has become much less attractive to me since transitioning to 24/7 max loading. The latest AMD offerings are attractive because they maintain good clock frequencies at moderate power levels, making them the current throughput per kWh champ in most instances, imo.

I have (so far) still been overclocking my 24/7 rigs, but I long ago decided to restrict increasing vcore to 5% or less to limit extra power usage, even that modest level did allow for a decent overclock. My current rig's vcore of 1.13v is just a 3% increase allowing my old IVB-E to o/c to 4.1 GHz from 3.9 GHz (+5%), (but 4.2 GHz wasn't stable even at 1.22v), probably the poorest compared to earlier CPUs, but still worth it.

That said, with the miniscule o/c that Ryzen 3000s seem to achieve, I think when I get a Ryzen 3600X I'm simply not going to bother! And that'll be the 1st time in over 20yrs that my main rig won't be overclocked!
In a way that's sad, but in another way it saves the all the hassle of fine tuning! (unless I try to undervolt it I suppose, if that's even do-able for these??).
Is it the end of overclocking??

1st overclocked CPU (main rig) - Pentium 200 MMX @225 MHz
Last(?) overclocked CPU - i7 4930k @4.1 GHz (stable, 4.2 GHz not quite).

And you guys?

 

[StefanR5R]

In a way that's sad, but in another way it saves the all the hassle of fine tuning! (unless I try to undervolt it I suppose, if that's even do-able for these??).

According to The Stilt's "Strictly technical: Matisse" post, undervolting is only feasible at ≤ 3.4 GHz. (See last paragraph and last graph of the opening post.)