Overclocked Ryzen 9 3950X outperforms 24-core 2970WX in Cinebench

mongeese

Posts: 643   +123
Staff
Why it matters: From six cores through twelve, Ryzen’s third iteration kills it in multicore benchmarks, and we expected nothing less from the upcoming 16-core 3950X. But according to an accidentally published overclocking manual from Gigabyte, performance is even better than anticipated and an overclocked 3950X can surpass all of Intel’s HEDT processors and is beaten only by AMD’s 32-core Threadripper.

Gigabyte’s engineers were able to consistently overclock the Ryzen 9 3950X to 4.3 GHz on their X570 Aorus Master motherboard (pictured above) using EKWB’s custom waterloop kit containing a 360mm radiator. The chip they used for benchmarking was running at 1.416 V with an x43 multiplier applied to a 99.98 MHz bus. Under a Prime95 workload, Die 1 had an average temperature of 97 C and a max of 101 C, and Die 2 averaged 93 C and spiked to 97 C. Not for the faint of heart, but not unmanageable.

At 1.452 V they were able to get one specific chip to run at 4.4 GHz, but they suggest that 4.3 GHz is the general limit for stability. In TechSpot’s own testing, both the 3900X and 3700X maxed out at 4.3 GHz on a 360mm AIO.

And finally, to the results. The 3950X simply obliterates AMD’s current 16-core monster, the Threadripper 2950X, scoring 24% higher. It falls a mere 10% behind the 24-core 2970WX. Overclocked to 4.3 GHz results are even more impressive as it can improve its score by 11%, pushing 1% past the 2790WX.

According to online results for the 18-core i9-9980XE, that chip can only achieve a ~3700 score at stock and ~4200 overclocked, both of which are about 200 points behind the 3950X when similarly configured.

Surely, the soon to be released i9-10980XE might outdo the 3950X, but at $230 more than the 3950X’s $749, Ryzen will probably be the value winner. We’ll know soon when both the 3950X and Intel’s Cascade Lake-X launch in November.

Permalink to story.

 
Even 3900X has already soared up at $749 due to short of supply, the first batch of 3950X is definitely sweet to buy.
 
Good luck getting a 3950 for $749. Retailers have already put the prices of the 3900 to above that.
 
Why does go gaga over these multi-core synthetic benchmarks when most applications can't utilize more than 4 cores?

Intel still wins in IPC (~17%) and clocks (~20%), which matter SO MUCH more in real-world use-case scenarios. Why does no one make an noise about the performance bugs, boosting to wrong cores, memory issues, and the fact that Ryzen 3 can't attain advertised boost speeds... until they create a stable chip that competes with i7/i9 for content creation, I see no reason to switch to AMD besides keeping myself busy with patches and work-arounds.
 
Intel has about 17% better IPC, and about 20% higher clock speeds. Most applications can only utilize 4 to 6 cores, so why is everyone going gaga over synthetic multi-core performance? I'll take real-world performance, reliability, and stability over theoretical gains that can't be realized at the cost of bugs, patches, memory issues, and chips that can't even hit advertised clock speeds under water.

I'll wait for Ryzen 4 or 5, but with Intel's IPC gains for 10th Gen. (12%) and price drops, it's likely a moot point.
 
Most applications can only utilize 4 to 6 cores, so why is everyone going gaga over synthetic multi-core performance?

I run engineering/scientific software that will take every resource you can throw at them cores, RAM, storage space/speed.

You are right that most software packages are limited to a handful of cores, but most peoples workflow/playflow consist of running multiple applications simultaneously which would benefit from additional cores/threads. With more computing resources available people can do more task simultaneously without seeing performance degradation.
 
What chips are you referring to? Zen 1 vs. Kaby Lake circa 2017? Zen 2 has equivalent or higher IPC than Intel. Intel's advantage continues to be clock speed.


Come on. Everyone knows that AMD is still behind in terms of IPC. They have made up ground, but early benchmarks show Intel's 10th Gen. enjoying a ~12% IPC boost, once again putting it out of reach of AMD.

Not to mention the still-buggy Ryzen platform is constantly experiencing its share of growing pains.
 
I run engineering/scientific software that will take every resource you can throw at them cores, RAM, storage space/speed.

You are right that most software packages are limited to a handful of cores, but most peoples workflow/playflow consist of running multiple applications simultaneously which would benefit from additional cores/threads. With more computing resources available people can do more task simultaneously without seeing performance degradation.

I agree, in certain use-case scenarios like yours it helps to have the extra cores. But in 99% of scenarios IPC and clock speed help out much more than extra cores. If the Ryzen platform was more mature it COULD be suitable for professional use, but as of now it's hard to seriously consider it.
 
Come on. Everyone knows that AMD is still behind in terms of IPC. They have made up ground, but early benchmarks show Intel's 10th Gen. enjoying a ~12% IPC boost, once again putting it out of reach of AMD.

Not to mention the still-buggy Ryzen platform is constantly experiencing its share of growing pains.

"Everyone knows" does not count as evidence. Benchmarks on this website show Zen 2 having equal or higher IPC compared to current Intel CPUs in most tasks. Intel still has an overall performance lead in some tasks (e.g., gaming) thanks to higher frequency and lower latency.
 
Come on. Everyone knows that AMD is still behind in terms of IPC. They have made up ground, but early benchmarks show Intel's 10th Gen. enjoying a ~12% IPC boost, once again putting it out of reach of AMD.

Not to mention the still-buggy Ryzen platform is constantly experiencing its share of growing pains.
Are you trolling? Because that's the only way your could write something like that.

First off, Intel's 10th gen, even with a 18% increase in IPC, can only barely match it's 9th gen in terms of single core performance because they can't reach the same high clocks (based on what they released so far). 10nm is clearly broken and can't scale well with clock speeds and potentially even cores (very low yeilds most likely). If intel could have released desktop parts on 10nm they would have done it already.

Second, tests show that at the same clock speeds AMD has the advantage in many single core tests (you can look at techspot's 4GHz tests they did with the 3900x/3700x vs 9900k). It depends on how sensitive the workload is to cache latency (which is why some games do slightly better on Intel).

And there are no game-braking bugs that affect Zen 2 right now. A new BIOS update is also coming soon.
 
Are you trolling? Because that's the only way your could write something like that.

First off, Intel's 10th gen, even with a 18% increase in IPC, can only barely match it's 9th gen in terms of single core performance because they can't reach the same high clocks (based on what they released so far). 10nm is clearly broken and can't scale well with clock speeds and potentially even cores (very low yeilds most likely). If intel could have released desktop parts on 10nm they would have done it already.

Second, tests show that at the same clock speeds AMD has the advantage in many single core tests (you can look at techspot's 4GHz tests they did with the 3900x/3700x vs 9900k). It depends on how sensitive the workload is to cache latency (which is why some games do slightly better on Intel).

And there are no game-braking bugs that affect Zen 2 right now. A new BIOS update is also coming soon.
I’d wait to see what Intel’s 10nm actual performance looks like before condemning it as DOA. Looking at techspots tests, any IPC increase whatsoever would give Intel the edge in this area. If they get the advertised 18% then the clock speeds could even be 10% lower than AMDs and still deliver more performance.

The node has been fraught with problems, its been heavily delayed. It’s a blessing for AMD really as their 10nm equivalent - 7nm isn’t that much faster than Intel’s 14nm node, implying that it would perform worse than Intel’s 10nm. If Intel can get it’s foundries back on track we could see some large performance gains being delivered over the next few generations from Intel. Which is important for consumers. At the moment Ryzen isn’t really reducing prices, in fact the 3xxx range is more expensive than any of Intel’s consumer grade lineups have been for a long time (although are better value, they still cost more). If Intel can challenge this we might get some cheaper CPUs! We need both companies producing the best they have.
 
I kind of expected this.
Zen was/is a new architecture and history has shown that means there should still be lots of headroom for further tweaking of the process. OTOH Intel is still using a 10 year old architecture, older if you consider the early Core processor (non I) as the first initial design.
Yes it's been heavily tweaked over the years and that's the big question, how much more room for improvements does the Intel design have? And turn around being fairplay how much do the Zen derived processors have? I'd bet money Zen has more room then Core does, and I think we're going to see this being proved as AMD further tweaks the design.
 
Back