AMD Ryzen 9 3950X CPU Breaks Three Overclocking Records

[biffzinker]

The AMD Ryzen 9 3950X 16-core, 32-thread processor isn’t out yet, but that hasn’t stopped AMD from using it to break some world records at E3 2019. AMD used the MSI X570 Godlike motherboard paired with a G.Skill Trident Z Royal DDR4 memory to get the most out of the Ryzen 9 3959X CPU. To help push the 3950X over 5GHz on all cores they pushed the yet to be released processor to the limits with LN2 cooling and some extra voltage in the UEFI. AMD Ryzen Master software was also used by those overclocking at the event.

AMD Ryzen 9 3950X CPU Breaks Three Overclocking Records - Legit Reviews

www.legitreviews.com

 

[phanbuey]

low for ln2 no? 5Ghz? didn't even beat the 1950x in 3dmark.

 

[dirtyferret]

The only thing I got from that was the ViewSonic stand must be cleaned!

 

[HTC]

low for ln2 no? 5Ghz? didn't even beat the 1950x in 3dmark.

Didn't they mention 5.3 GHz all-core in the video from the link? Perhaps i heard wrong.

Still, i believe it was @TheLostSwede that said there was an ES sample a while back that was clocked @ 5.5 all-cores: dude appears to have access to info most of us do not.

Personally, i never expected a high core count chip to be able to have very high all-core clocks. It's to be expected on lower core count CPUs but not high core count CPUs.

 

[biffzinker]

low for ln2 no? 5Ghz?

In the video the one guy says they started the OC at 5.3 GHz in Ryzen Master.

 

[Fluffmeister]

The only thing I got from that was the ViewSonic stand must be cleaned!

Sadly, all the budget went on the MSI X570 Godlike.

 

[Tomgang]

That´s good and all. But what i really want to know is how this thing overclock and performe on normal cooling (Air/custom water/AIO). LN2 performance is useless for every day use and only suited for records/benchmark.

And to that i will say 5.3 GHz is rather low on LN2. I mean in the past we have seen cpu´s go up to around the 7-8 GHz mark. So 5.3 GHz on its own really dosent impress me.

 

[HTC]

That´s good and all. But what i really want to know is how this thing overclock and performe on normal cooling (Air/custom water/AIO). LN2 performance is useless for every day use and only suited for records/benchmark.

And to that i will say 5.3 GHz is rather low on LN2. I mean in the past we have seen cpu´s go up to around the 7-8 GHz mark. So 5.3 GHz on its own really dosent impress me.

And how many cores did those CPUs have?

 

[Vya Domus]

And how many cores did those CPUs have?

The number of cores wasn't exactly the most relevant thing really. When you go for Ghz as your main goal most records are made with everything but one core active anyway.

 

[HTC]

The number of cores wasn't exactly the most relevant thing really. When you go for Ghz as your main goal most records are made with everything but one core active anyway.

For CPU-Z yes, but not for stuff such as CB R15 / R20: such benchmarks require the full chip enabled.

 

[Vya Domus]

For CPU-Z yes

And that's exactly how those 7-8 Ghz records were made which he was referring to.

 

[HTC]

And that's exactly how those 7-8 Ghz records were made which he was referring to.

Still, one can't expect to be able to run 1 core (with 15 disabled) @ ... say ... 6 GHz on these CPUs, right? Even if that were possible, it would most likely end up killing the whole chip.

Out of curiosity, how high is the record frequency of Intel's 16c / 32t CPU(s), and @ what frequencies were their respective R15 / R20 records achieved?

 

[TheoneandonlyMrK]

low for ln2 no? 5Ghz? didn't even beat the 1950x in 3dmark.

Memory Bandwidth?.

 

[biffzinker]

Memory Bandwidth?.

Shouldn't be constrained by memory bandwidth when their running the G.Skill Trident Royal Z overclocked at 4,533 MHz.

 

[TheoneandonlyMrK]

Shouldn't be constrained by memory bandwidth when their running the G.Skill Trident Royal Z overclocked at 4,533 MHz.

That is'nt quite it though, lower speed ,lower latency.
Plus with quad channel memory it probably nudges it over the line.

 

[Tomgang]

Talking about benchmark. Look at this.

3950X properly an early es model takes on intels flagship i9 9980X in geekbench

If this is true, 3950X gonna be a beast and even at 749 usd a barging deal compared to intels price tag.

AMD Ryzen 9 3950X 16 Core CPU Benchmark Leaks Out, Obliterates The Intel Core i9-9980XE 18 Core CPU in Single and Multi-Threaded Workloads

The latest benchmark of the AMD Ryzen 9 3950X 16 core CPU has leaked out which shows it performing twice as fast as Intel's i9-9980XE.

www.google.com

 

[dylricho]

And to that i will say 5.3 GHz is rather low on LN2. I mean in the past we have seen cpu´s go up to around the 7-8 GHz mark. So 5.3 GHz on its own really dosent impress me.

It's not about what frequency you can get to, but what you can do with the frequency you achieve. First-generation Intel Core i7 and AMD FX processors are good examples here. Modern Intel Core i7s and AMD Ryzens will wipe the floor with both of them, with significant frequency deficits. (The world record for CPU frequency goes to an FX-8370 @ 8,723 MHz, but due to its significant instruction throughput deficit in its design, a stock Ryzen 7 1700X beats it with ease.)

There are other factors too; node specifications, architecture design and heat density, to name a few.

It's much easier to achieve high-frequency overclocks on architectures designed around using them to offset their weaker instruction throughput. The problem is, physics doesn't care about your goals. So while it might be easier to bump up the frequencies, you'll run into the same walls at some point. To gain higher frequencies, you need to increase voltages, increase current, or both. Thermal output increases disproportionately to the voltage/current increase. That applies to any type of architecture. You don't need to increase these variables to improve instruction throughput. That's why it's the better design.

For first- and second-generation Ryzen, the node itself severely limits the frequencies achievable because the voltage curve is at its most efficient between 2.00–3.00 GHz. From there, it scales very badly because Samsung never intended for its node to be used for 4.00 GHz processors.

As others have pointed out here, I would rather see all-core overclocks at lower frequencies, than a world record frequency on a single core. It might be unrealistic to overclock on liquid nitrogen for most people (and it is), but it's even more unrealistic to buy a high-core-count processor to only use one core at a super-high frequency. I'll take the 5.00 GHz over 7.00 GHz easily.

Out of curiosity, how high is the record frequency of Intel's 16c / 32t CPU(s), and @ what frequencies were their respective R15 / R20 records achieved?

Core i9-7960X @ 5,737 MHz | 5,149 cb
Core i9-9960X @ 5,928 MHz | 5,320 cb
Ryzen Threadripper 1950X @ 5,370 MHz | 4,514 cb
Ryzen Threadripper 2950X @ 5,487 MHz | 4,758 cb

 

[Vayra86]

It's not about what frequency you can get to, but what you can do with the frequency you achieve. First-generation Intel Core i7 and AMD FX processors are good examples here. Modern Intel Core i7s and AMD Ryzens will wipe the floor with both of them, with significant frequency deficits. (The world record for CPU frequency goes to an FX-8370 @ 8,723 MHz, but due to its significant instruction throughput deficit in its design, a stock Ryzen 7 1700X beats it with ease.)

There are other factors too; node specifications, architecture design and heat density, to name a few.

It's much easier to achieve high-frequency overclocks on architectures designed around using them to offset their weaker instruction throughput. The problem is, physics doesn't care about your goals. So while it might be easier to bump up the frequencies, you'll run into the same walls at some point. To gain higher frequencies, you need to increase voltages, increase current, or both. Thermal output increases disproportionately to the voltage/current increase. That applies to any type of architecture. You don't need to increase these variables to improve instruction throughput. That's why it's the better design.

For first- and second-generation Ryzen, the node itself severely limits the frequencies achievable because the voltage curve is at its most efficient between 2.00–3.00 GHz. From there, it scales very badly because Samsung never intended for its node to be used for 4.00 GHz processors.

As others have pointed out here, I would rather see all-core overclocks at lower frequencies, than a world record frequency on a single core. It might be unrealistic to overclock on liquid nitrogen for most people (and it is), but it's even more unrealistic to buy a high-core-count processor to only use one core at a super-high frequency. I'll take the 5.00 GHz over 7.00 GHz easily.




Core i9-7960X @ 5,737 MHz | 5,149 cb
Core i9-9960X @ 5,928 MHz | 5,320 cb
Ryzen Threadripper 1950X @ 5,370 MHz | 4,514 cb
Ryzen Threadripper 2950X @ 5,487 MHz | 4,758 cb

This.

But let's just get things straight, these OC records never have anything to do with real world performance. Its a number epeen show, entertainment and branding is the purpose.

 

[dylricho]

Its a number epeen show, entertainment and branding is the purpose.

Agreed.

 

[advanced3]

On air or water I would be surprised if they overclock better than 2nd Gen Ryzen, which isn't much. They're great CPUs, just not all that fun for overclocking.