Ryzen 9 3950X vs Threadripper 2950X: Best Value? (2020)

| | , ,

Performance Winner
AMD Ryzen 9 3950X 16-Core, 32-Thread Unlocked Desktop Processor, Without Cooler
Second Place
AMD Ryzen Threadripper 2950X Processor (YD295XA8AFWOF)
Model
AMD Ryzen 9 3950X
AMD Ryzen Threadripper 2950X
Test Result
Test Result 9.8/10 Excellent June 2020
Test Result 9.6/10 Very Good June 2020
Manufacturer
AMD
AMD
Performance category
High-End CPU segment
High-End CPU segment
Cores
16 cores (3.5 GHz up to 4.7 GHz Max Turbo )
6 cores (3.8 GHz up to 4.4 GHz Max Turbo )
Performance for Gaming
Workstation (Applications) Performance
Overclocking possible?
Hyperthreading possible?
First release date
Q3/2019
Q2/2018
Max. RAM
DDR4-3200 up to 128GB
DDR4-2666 up to 128GB
Pros
  • Fastest prosumer CPU on the market
  • Compatible with most Socket AM4 motherboards
  • Completely unparalleled power efficiency
  • Native PCIe Gen 4.0
  • Offers more performance, uses less power and runs cooler than its Core i9 competition
  • Lots of horsepower packed into 16C/32T configuration
  • Works on existing Threadripper motherboards
  • Easy-to-use Ryzen Master software utility
Cons
  • Doesn’t always catch coffee lake in ST workloads
  • Requires quad-channel memory for optimum performance
Performance Winner
AMD Ryzen 9 3950X 16-Core, 32-Thread Unlocked Desktop Processor, Without Cooler
Model
AMD Ryzen 9 3950X
Test Result
Test Result 9.8/10 Excellent June 2020
Manufacturer
AMD
Performance category
High-End CPU segment
Cores
16 cores (3.5 GHz up to 4.7 GHz Max Turbo )
Performance for Gaming
Workstation (Applications) Performance
Overclocking possible?
Hyperthreading possible?
First release date
Q3/2019
Max. RAM
DDR4-3200 up to 128GB
Pros
  • Fastest prosumer CPU on the market
  • Compatible with most Socket AM4 motherboards
  • Completely unparalleled power efficiency
  • Native PCIe Gen 4.0
Cons
  • Doesn’t always catch coffee lake in ST workloads
Second Place
AMD Ryzen Threadripper 2950X Processor (YD295XA8AFWOF)
Model
AMD Ryzen Threadripper 2950X
Test Result
Test Result 9.6/10 Very Good June 2020
Manufacturer
AMD
Performance category
High-End CPU segment
Cores
6 cores (3.8 GHz up to 4.4 GHz Max Turbo )
Performance for Gaming
Workstation (Applications) Performance
Overclocking possible?
Hyperthreading possible?
First release date
Q2/2018
Max. RAM
DDR4-2666 up to 128GB
Pros
  • Offers more performance, uses less power and runs cooler than its Core i9 competition
  • Lots of horsepower packed into 16C/32T configuration
  • Works on existing Threadripper motherboards
  • Easy-to-use Ryzen Master software utility
Cons
  • Requires quad-channel memory for optimum performance

When it comes to top performance with high end CPUs every percent more in performance counts.

In this article we take a closer look at AMDs Ryzen 9 3950X vs Threadripper 2950X.

We answer the questions of which of them has more performance and which has the better price-performance ratio.

AMD Ryzen 3000 is officially available since July 2019 and has turned the rankings upside down – and in November 2019 AMD added the Ryzen 9 3950X with 16 cores.

Almost all models can be recommended without restriction to anyone who is thirsty for multi-threading performance in applications; no Intel CPU at a price level that is only rudimentarily similar to the newcomers.

In games, Intel remains minimally in the lead with the fastest CPUs, but only the top models are still in the lead with a 20 percent increase in CPU limit for Ryzen.

The total package of performance, power consumption and price, on the other hand, is decided almost everywhere by AMD.

Test Results: Ryzen 9 3950X vs Threadripper 2950X

Ranking First: AMD Ryzen 9 3950X

AMD Ryzen 9 3950X

Pros

  • Fastest prosumer CPU on the market
  • Compatible with most Socket AM4 motherboards
  • Completely unparalleled power efficiency
  • Native PCIe Gen 4.0

Cons

  • Doesn’t always catch coffee lake in ST workloads

AMD Ryzen 9 3950X – Best performing CPU

Since the first announcement of the Ryzen 3000 in January 2019, it was clear that AMD was planning the first 16-core processor for a midrange platform.

At the start of the CPU series, the 12-core Ryzen 9 3900X was still the top model, but now the Ryzen 9 3950X is entering the market with four additional cores and higher clock speed.

And this much can be said in advance: never has a mid-range chip been faster.

With the Ryzen 9 3950X, AMD consistently demonstrates its chip design consisting of several dies for one processor, so that the CPU often beats even more expensive Intel models with more cores and also requires less energy, since AMD has the dies manufactured in the 7-nm process.

Also important is the Zen-2 microarchitecture, which has a significant speed increase in performance per clock (IPC, Instructions per Cycle) compared to Zen(+) of the first two Ryzen generations.

A quick reminder: In Zen 2, AMD has introduced a day jump prediction in the front end of the CPU cores in addition to the Perceptron-based jump prediction; this is slower but gives better results.

In addition, there were significantly larger branch buffers, while the L1 instruction cache was halved in favor of the doubled micro-op cache.

A third address unit (Store-AGU) helps in the backend, the bandwidth of the load/store units and the L1 data cache was doubled.

The four floating-point units now work with 256 bits instead of 128 bits, the pipelines can therefore process AVX2 instructions in one clock; important in the HPC segment.

At chip level, the L3 cache has also been doubled from 8 MByte to 16 MByte and the memory controller can now officially handle DDR4-3200 instead of DDR4-2933.

In terms of CPU cores, AMD speaks of a 15 percent higher IPC, while the improvements in L3 cache, memory and clock speed increase performance even further.

The Ryzen 3000 alias Matisse have cores manufactured in 7 nm, which is why AMD was able to increase the frequency compared to Ryzen 1000/2000.

The Ryzen 9 3950X achieves 4.7 GHz at its peak and thus comes close to the 5 GHz of a Core i9-9900KS.

However, to achieve such a clock speed, AMD had to select chips over months.

This so-called binning is necessary, so that not only the highest frequency is applied to a core, but the ratio of clock and voltage on all cores is as efficient as possible.

The Ryzen 9 3950X has the same nominal thermal power dissipation as the Ryzen 9 3900X despite having 16 cores instead of 12.

Only the best 8C cores

Like its predecessor, the Ryzen 9 3950X also consists of three chips internally, which is why this design is known in the industry as chip design: AMD uses a central I/O die (IOD) and two compute cache dies (CCD).

The IOD includes the dual-channel DDR4 memory controller and other I/O functions such as the 24 PCIe Gen4 lanes, plus four USB 3.2 Gen2 ports and up to two Sata 6 Gbps ports.

Of the lanes, four are reserved for connection to the chipset, four are intended for an NVMe SSD and another four can be used for a second SSD.

It is up to the motherboard whether there are four or two lanes for an SSD, in the latter case the two Sata 6 Gbps ports mentioned above are additionally available.

Not all boards use all connectors, sometimes the focus is on the ports of the respective chipset or PCIe lanes are branched off for external controllers.

The compute cache dies (CCD) always comprise eight cores, internally they consist of two 4-core CCXes (Core Complex) with 16 MByte L3 cache each.

The eight cores do not behave identically, as they are subject to minimal manufacturing variations as with all other processors.

For example, the quality of chips on the edge of a wafer drops slightly, while those in the middle have better electrical properties.

All of these are tested and evaluated after the wafer has been separated, i.e. cut into individual chips.

In the industry, this is called binning (pre-selection): Manufacturers test dozens of dies at once to find out which clock and which voltage at which power consumption and which temperatures they reach.

Some cores are better suited for high peak frequencies and others can achieve 4 GHz at very low Vcore.

With the Ryzen 9 3950X, both of these requirements must be met, since the 16 cores fit into a power budget of 105 watts (more precisely: up to 142 watts) and at the same time at least some of them must deliver up to 4.7 GHz.

AMD Ryzen 9 3950X

AMD at least reveals which cores are best for overclocking: In the tool called Ryzen Master, a gold star indicates the best core of a chipset, a gray star indicates the best core of each CCX, and a gray circle indicates the second best.

The operating system’s scheduler, on the other hand, uses confusingly different information to pack single-threaded tasks onto a core, since a high boost clock via CPPC2 is crucial here and not just the electrical quality.

Basically Windows 10 v1903 can do this, but with Windows 10 v1909, which was released on Tuesday, the scheduler behavior was improved according to Microsoft.

The 19H2 update for the operating system is supposed to prioritize favored CPU cores more strongly instead of moving tasks to slower cores, which may process the workload more slowly.

In the case of AMD, there is also the fact that threads are kept within a CCX for latency reasons instead of being moved to the second cluster or even to the second chipset.

Faster Agesa 1004b

The CPU cores and their clock is controlled by the System Management Unit (SMU), which in turn can be updated by Agesa (AMD Generic Encapsulated Software Architecture) as part of the mainboard firmware.

The Agesa contains the necessary microcode for the respective chips and is the basis for each motherboard UEFI. For the Ryzen 9 3950X, AMD lists the Agesa 1004b as necessary for full support.

In general, the Ryzen 3000 alias Matisse has been running since Agesa 0072, but this early version with preliminary support was more for evaluation.

Our sample used before the launch had problems with the clock and the generation of temperature sensor data, among other things, and only the Agesa 1001 provided a remedy.

For the launch, AMD provided the rather performance-optimized Agesa 1002, which we used to test the Ryzen 9 3900X and Ryzen 7 3700X.

AMD Ryzen 9 3950X

With the Agesa 1003, more specifically with versions from 1003a to 1003abb, AMD fixed a bug in the random number generator (RDRAND) of the Ryzen 3000 as well as a problem where PCIe Gen2 was incorrectly activated instead of PCIe Gen4.

In the meantime, some functions were missing on boards with 16 instead of 32 Mbyte firmware, but these are now back.

With the Agesa 1003abba, AMD changed the boost behavior of the Ryzen 3000.

Since then, the Ryzen 3000 clocks faster in the peak range of 25 to 50 MHz and no longer run high under very light background load.

The current status – also for this test – is the Agesa 1004b.

It reunites the code base, since Matisse (Ryzen 3000) was developed differently from the Ryzen 1000 and Ryzen 2000(G) as a fork of the mainline.

Therefore, the firmware base is also implemented by all manufacturers for older boards such as those with B350 chip, although so far primarily updates for X570 boards are available.

In terms of performance, the Agesa 1004b does not differ from the 1002 or 1003 variants:

Depending on the benchmark, we measure minimal advantages or disadvantages, but no significant differences.

Only PCMark10’s app-startup sub-test shows a greater improvement on the Ryzen 9 3900X, where AMD’s mentioned optimizations of the boost clock and the use of the favored CPU cores for the Agesa 1004b app app app app startup seem to have an impact.

According to AMD, there is also a revised stability on the X570, better interoperability with PCIe devices, more stable ACPI power states and shorter post/boost times.

These depend on the motherboard; in our ROG Crosshair VIII Hero (X570) from Asus it was 7 seconds with standard settings.

Last but not least, the Agesa 1004b is still a prerequisite for the new Eco Mode: 105/95 watt chips can thus be set to 65 watts and 65 watt models to 45 watts via Ryzen Master under Windows.

We tried this briefly, but first the regular benchmarks at full 105 watt TDP.

18 Intel cores outdated

AMD Ryzen 9 3950X

We test the AMD processors on a ROG Crosshair VIII Hero (X570), a ROG Crosshair VII Hero (X470) and a ROG Zenith Extreme (X399), the Intel models on a ROG Maximus X Hero (Z370) and a Prime X299-Deluxe.

The memory clock rate meets the manufacturer’s specifications, a Geforce RTX 2080 Ti in the Founder’s Edition delivers the pictures.

The Windows 10 v1903 including applications and games is on a WD Black SN750, the Ubuntu 19.10 on a Samsung 970 Evo Pro.

All mitigations against Microarchitectural Data Sampling, Meltdown and Spectre are active in the firmware versions of the motherboards and in the chips.

Our Core i9-9900KS uses the current R0 stepping, which minimally reduces the performance compared to the P0 revision.

Across all measurements under Windows, the 16-core Ryzen 9 3950X beats the 12-core Ryzen 9 3900X by about 11 percent.

If we leave out the benchmarks, which only burden one or a few cores, the gap increases to 27 percent.

At this point, however, it should be said that the course used was primarily designed for desktop chips with eight cores and not for those with twice as many.

With the release of Threadripper v3 we will show results that scale much more with multithreading, here the distance of the 3950X to the 3900X and other CPUs increases accordingly.

Intel’s eight-core Core i9-9900KS is beaten by 50 percent in the course presented here, in applications like Blender the Ryzen 9 3950X is almost twice as fast.

The comparison to the Core i9-7980XE is exciting: This 18-core chip was Intel’s top HEDT model two years ago, the current Core i9-9980XE doesn’t compute any faster.

The Ryzen 9 3950X overtakes the 7980XE by 14 percent, whereby it is drastically faster when compiling Unreal Engine 4 and somewhat slower when exporting with Adobe’s premiere.

The values of the new 18-core Core i9-10980XE are still under NDA.

The Ryzen 9 3950X beats the Threadripper 2950X, which also has 16 cores, by 21 percent.

Under Linux, the quick test shows that the Core i9-7980XE also has a hard time against the Ryzen 9 3950X:

When encoding some 4K-HDR-resolved frames of Tears of Steel with AV1, the Intel chip is slower, the same applies to Blender with the classroom scene.

Only when compiling Qt 5.12 with the GCC is the Ryzen subject to the dual-channel instead of quad-channel interface.

Again, we will be releasing the results of the 18-core Core i9-10980XE in a few days.

AMD Ryzen 9 3950X

In terms of power consumption, there are no surprises in the Ryzen 9 3950X:

AMD specifies it with 105 watts, but with a case temperature below 62 degrees Celsius, it can take up to 142 watts like other Ryzen 3000 models.

The 3950X is somewhat more economical under load than the 3900X, probably because AMD has selected the better chips.

Both Ryzen 9s require significantly less power than the Core i9-7980XE or a Threadripper.

The eco mode reduces the power consumption, the speed remains high.

Verdict: AMD Ryzen 9 3950X – Best performing CPU

A whopping 16 cores – this has never been seen before in the middle class.

We find it impressive how AMD has quadrupled the number of CPU cores in the mid-range segment in just over two years, starting with quad cores and leaving out the modules of the bulldozer architecture.

The Ryzen 9 3950X delivers extremely high performance, which even in well-scaled workloads such as Blender or Cinebench comes close to that of a 24-core Threadripper 2970WX.

Thanks to Zen 2 technology and the high clock speed provided by the 7nm production, the 16-core AMD chip is also convincing under mixed load and is just as suitable for games as all other Ryzen 3000’s.

Only processors like Intel’s Core i9-9900K(S) have slight advantages – measurable instead of noticeable – but these are simply without a chance with massive multithreading.

Even Intel’s models with more cores, such as the Core i9-7980XE with 18 cores, calculate almost consistently slower than the Ryzen 9 3950X.

In terms of price, the AMD chip competes with the expiring 14-core Core i9-9940X and the soon to be released Core i9-10940X, but both are not worthy opponents for the Ryzen 9 3950X.

Only in very bandwidth-limited scenarios do the two have an advantage; these are rare apart from server applications.

Thus, the Ryzen 9 3950X is currently the best mid-range processor on the market.

Especially for creative people, the Matisse CPU is a good choice.

Only those who need more PCIe lanes should choose Intel’s Core i9 or AMD’s thread rippers, where not only the CPUs but also the platform are even more expensive.

Ranking Second: AMD Threadripper 2950X

Pros

  • Offers more performance, uses less power and runs cooler than its Core i9 competition
  • Lots of horsepower packed into 16C/32T configuration
  • Works on existing Threadripper motherboards
  • Easy-to-use Ryzen Master software utility

Cons

  • Requires quad-channel memory for optimum performance

Great CPU – AMD Threadripper 2950X

Two years ago, 16 cores and 32 threads were absolutely unimaginable in the desktop PC, but then came AMD with the 1st generation Ryzen Threadripper.

Just one year and three days later, AMD Ryzen Threadripper 2000 breaks the new ceiling, which had just been recognized as unbelievable, and once again sets out to offer extreme performance at a relatively low price.

Because 32 cores of the Ryzen Threadripper 2990WX for 1,829 Dollar is a real challenge.

28 cores at Intel currently cost at least 8,700 US dollars as Xeon.

Intel has no alternative at all for desktop PCs on paper, where 18 cores for just under 2,000 Dollar for the Core i9-7980XE is the end of the road.

No less interesting, however, is the AMD Ryzen Threadripper 2950X with 16 cores and 32 threads, also tested today, which takes up the direct legacy of the first flagship Ryzen Threadripper 1950X for the X399 platform.

In some cases significant clock adjustments coupled with internal improvements promise performance gains in all areas – and all this at an entry price of 100 Dollar less than the year before.

AMD Ryzen Threadripper 2000 at a glance

AMD divides the new high-end CPUs into two series, because addressing 32 cores in a classic desktop is not so easy.

Consequently, the new flagship AMD Ryzen Threadripper 2990WX with 32 cores and 64 threads and combined 80 MByte L2 and L3 cache is directly promoted as a product for the workstation market (“WX”).

AMD hopes to meet customers here who know if they can use 64 threads. In the shadow of the flagship follows the Ryzen Threadripper 2970WX with 24 cores and 48 threads – the exact same applies to it.

In the X-Series, however, almost everything remains the same.

The previous 16-core 1950X and 12-core 1920X will be transferred 1:1 to the Pinnacle Ridge architecture and will therefore offer the same key features as their predecessors.

First and foremost, this means a faster cache due to lower latencies and, thanks to 12-nm manufacturing, also somewhat higher clock rates.

Ryzen Threadripper 2950X and 2920X are also the two new entry-level solutions for the socket TR4 platform.

The smallest Threadripper 1900X will not be renewed, it will expire without a successor.

Tabular overview of the new and old Threadripper

Four models replace three old ones, and the smallest one has no successor at all.

Besides the basic clock rate AMD only mentions the maximum possible clock rate, because with XFR, XFR2 and Precision Boost there are always different clock rates.

They are based on a wide range of parameters and are therefore strongly dependent on the system, its environment and the applications.

There are no longer any fixed values for load on four cores, for example.

Traditionally, the maximum possible clock rate is only valid for one core, even with AMD, and the minimum never in the ideal case.

The prices have once again been significantly adjusted compared to the previous year.

In particular, the 12-core processor has been significantly reduced and costs $649 at the start, $150 less than last year.

Exactly double the amount of cores is also twice as much: 1,299 US dollars are charged for 24 cores.

Today only the 32-core CPU starts

The models will not all be available at the same time.

As of today, only the 32-core processor will be available from stock.

The 16-core model will follow on 31 August. Without a fixed date, the 12- and 24-core processors will be available in October.

4 × Pinnacle Ridge – except for memory

The two small Threadripper 2000 continue the tradition of their predecessors:

There are two active and two dummy dies under the soldered heatspreader.

The flagship with 32 cores as well as the 24-core, however, has four dies activated and installed on the large LGA-4096 package for sockets TR4.

At first glance, the CPU looks like an Epyc Server processor, but something essential is missing:

The memory controllers of two CCXs are not routed directly to the outside, but are connected to the controllers of the other two CCXs via Infinity Fabric.

This leaves the familiar quad-channel memory interface. From now on, however, it can handle DDR4-2933 – just like Ryzen 2000.

12-nm production for higher clock speed

The processor cores use exactly the same components from the Zen+ family as those found in Ryzen 2000.

Thanks to the 12 nm production, higher clock rates are offered.

And, as with its predecessor, AMD says that the top five percent in chip manufacturing are sorted out during binning and kept for Threadripper.

In addition to the higher TDP, this ensures that Threadripper 2000 can again clock higher than Ryzen 2000 in certain load scenarios.

Cache and memory latencies have been reduced

AMD Threadripper 2950X

Aside from the process, AMD has been known to put most of the work in Zen+ into reducing cache and memory latencies.

However, since Threadripper 1000 already had some latency optimizations compared to Ryzen 1000, the differences between Ryzen Threadripper 1000 and Ryzen Threadripper 2000 are slightly less than between Ryzen 1000 and Ryzen 2000.

While AMD estimates that the latency of the L2 cache of Ryzen 2000 is up to 34% better than that of Ryzen 1000 and that the latency of the L3 cache is up to 16% better, the progress of Threadripper 2000 compared to its predecessor is only 9% and 15%, respectively.

In DRAM, too, only 2 percent of the 11 percent advantage for Ryzen remains with Threadripper.

There have been no other architecture changes in Zen+.

However, an additional increase in performance is achieved by a modified turbo mechanism.

New Turbo for multi-core loads

Precision Boost 2 is also called Threadripper’s new Turbo mode, which takes many variables into account and thus determines the clock for the CPU.

The second generation of this technology has been in use since Raven Ridge in mid-February and raises and lowers the clock in 25 MHz steps within milliseconds.

Among other things, the limit is defined as in modern graphics cards:

The processor has a power limit and a temperature limit.

What is reached first reduces the clock (below the maximum possible and firmly defined turbo).

Techtestreport had already explained all the details in detail in the test for Ryzen 2000.

XFR 2 and 27 degree temperature offset

Threadripper 2000 also supports XFR 2, and the auxiliary turbo maintains the known limits on temperature and power consumption, but now works even when all cores of the CPU are in use.

In this case, the temperature should be too high in most cases, but there is no general exclusion for XFR with XFR 2, nor is there a limitation of the cores.

AMD does not provide an indication of how much additional clock speed XFR 2 can get out of the CPU.

To make XFR even more effective, AMD has built an offset temperature into the X versions of the first Ryzen, to make the fan controller believe that the temperature is 20 °C higher.

If the CPU was actually 60 degrees warm, with a 20 degree offset like the Ryzen 7 1800X, the fan controller reacted as if the CPU was actually 80 degrees warm, to keep the CPU for XFR below real 60 degrees.

With Ryzen Threadripper 2000 this value is even 27 degrees. Most tools like HWiNFO already know about this peculiarity and display both temperatures.

DDR4-2933 only with up to four modules

AMD Threadripper 2950X

During the PR presentations, AMD embellished the specifications for maximum bandwidth a bit and operated the systems with DDR4-3200, but officially the CPUs are only specified for DDR4-2933 – and even this only in operation with single-rank modules.

When the platform is fully equipped with dual-rank memory, only DDR4-1866 is offered.

Test results for Threadripper 2990WX & 2950X

AMD Ryzen Threadripper 2990WX and 2950X compete together on an Asus ROG Zenith Extreme in the test.

The BIOS has been updated to a beta version with the status at the end of July.

The test system used is otherwise exactly the same as in the big Ryzen 2000 test, which also included the Threadripper predecessors, so that comparability is given.

According to AMD, all motherboards available for the TR4 socket should be able to handle 32 cores.

This requires a BIOS update in each case, but BIOS flashback should not cause the same problems as with Ryzen 2000, where new CPUs were only bootable on mainboards that already had the new BIOS installed.

32 cores are (still) causing problems in everyday life

The restriction of the two large CPUs to a quad channel interface only indirectly serves to distinguish Epyc from the X399 platform, because even the X399 platform with its socket TR4 simply only provides a quad channel memory interface.

The I/O has therefore also been relocated.

In the end, two of the four dies not only have no memory interface, but no connections to the outside.

Instead, the “Compute Dies” are passed internally to the other two “I/O Dies”, which then pass them to the outside.

In the worst-case scenario, this inevitably leads to higher latencies, since requests from the cores of the Compute Dies first pass through switching centers and the Infinity Fabric On Package (IFOP) into the I/O Dies and have to be sent from there to the RAM, only to have to return the same way afterwards.

WikiChip has illustrated this clearly in a diagram.

The memory bandwidth measurement of AIDA64 shows the limitations of the implementation method.

For example, the memory bandwidth in MB/s on the 32 core is less than that of the 2950X with half as many cores, although nothing on the platform has been swapped except for the CPU and both offer a quad channel interface to the outside.

The fact that the editorial staff didn’t make any mistakes here is shown by the latency measurement, because the results presented by AMD are hit here – and the disadvantages for the 2990WX are also visible.

The deviations in DRAM can be explained by the fact that AMD measured with DDR4-3200, but Techtestreport used the official maximum of DDR4-2933.

What AMD didn’t talk about much in the press beforehand is the performance losses in applications that (currently and) potentially still accompany the cuts.

And even with 64 threads many programs have to be able to handle at all.

According to AMD, the Windows scheduler is also involved here.

Together with Microsoft, they are already working on optimizing Windows to handle more than 32 threads on Threadripper 2000.

The software has problems with 64 threads and the memory interface

The fact that the configuration of the RAM controllers is not to the liking of every software can be seen, for example, in the TrueCrypt successor VeraCrypt in the latest version 1.22.

The performance of the 32-core is only half as high as that of the 16-core and corresponds to the level that would be achieved if only the dual-channel memory mode was run on the 32-core.

The test with only two instead of four memory locks on the Ryzen Threadripper 2990WX therefore gives exactly the same result as with four memory locks: around 10.5 GB/s.

At present, this program does not like the compute-die I/O-die arrangement at all.

Threadripper 2950X with 16 cores, on the other hand, always gets the quad channel memory interface running and, like its predecessor, delivers just under 21 GB/s.

Even the legacy mode in the Ryzen Master Tool and the halving of the cores doesn’t bring any improvement in the 2990WX.

Even 7-zip in the current version 18.05 hardly reacts to the difference between two and four memory banks on the 2990WX: 107,335 MIPS are with two modules, 111,624 MIPS in quad channel mode.

This represents an increase of only three percent. The 16-core unit already achieves almost 106,000 MIPS – even here, scaling via the additional memory bars and cores is ultimately not possible.

Other programs such as rendering in Cinebench or POV-ray or applications from Adobe sometimes scale potentially via the threads, but the performance of the 2990WX is in the end still below that of the 2950X in some cases because the bandwidth of memory/fabric is much too low to supply the 32 cores with sufficient data.

AMD had therefore made three additional programs available to editorial offices for testing one working day before the start: all three are rendering programs.

The course used by Techtestreport independent of manufacturer recommendations and these recommendations suggest that rendering is currently the main application area where 32 cores can be unfolded.

And indeed, many workstations are being purchased precisely for this purpose.

But there are exceptions here as well.

Apart from the standard benchmark in Blender, which sees the 2990WX before the 2950X, the picture even reverses in another project file:

While the 16-core completes it in an hour, the 32-core takes an hour and 20 minutes.

It doesn’t matter whether Blender is used in final version 2.79 or experimental version 2.80: In either case, the specific project cannot load all threads on AMD’s Threadripper.

Also an additional test with two projects in Handbrake with the latest version 1.1.1 looks rather bad for the 32-core. It is close to the 16-core, but not before.

It also fails here due to the load of the cores.

Benchmarks in games

For games, AMD also has Threadripper 2000, the game mode introduced last year.

This halves the number of cores and threads on the 16-core, on the 32-core, they are even quartered. With this AMD wants to counter compatibility problems in games.

AMD Threadripper 2950X

In comparison to last year, the use of Game Mode was no longer necessary with the 16-core processor to get games to work together.

A bit more performance was delivered by one or the other title in this mode, though:

With Game Mode, Threadripper 2950X can catch up with Ryzen 7 2700X in some games.

This should ideally be the case, as the clock rates are also at almost the same level.

The 2990WX needs the “Game Mode”

With the 32-core, however, Game Mode had to be used to prevent crashes in Far Cry 5 or extremely low FPS in Total War.

But also in other titles the FPSs with less active cores attracted attention. Also in the games, however, the rating is waived in the end because it distorts the picture too much.

Verdict: Great CPU – AMD Threadripper 2950X

Threadripper 2950X: 16 cores without problems

In contrast, the Ryzen Threadripper 2950X with 16 cores shows a consistently positive image.

Compared to its predecessor, its performance is increasing in all areas and in summer 2018, no more tangible problems have been reported on the test track.

The processor thus presents the much rounder overall picture.

Of course, 16 cores and 32 threads also want to be loaded by the right programs, but they are more diverse due to the predecessor’s preparatory work for a year now.

And because even the 32 threads are not yet used by every software that is dedicated to multi-threading support, there is still room for improvement in this CPU.

Either way, it should be clear at the end of the day before buying, for which purpose the CPU is needed.

The HEDT platform is and will remain associated with a significant surcharge compared to the middle class, the classic gamer is much better off there.

But the more PCIe lanes paired with the right applications make the 16-core especially attractive.

The 32-core, on the other hand, still has too many construction sites in the home desktop, only those who know how to handle it should consider it as an alternative to Epyc.

Final Verdict: Ryzen 9 3950X vs Threadripper 2950X

Performance Winner
AMD Ryzen 9 3950X 16-Core, 32-Thread Unlocked Desktop Processor, Without Cooler
Second Place
AMD Ryzen Threadripper 2950X Processor (YD295XA8AFWOF)
Model
AMD Ryzen 9 3950X
AMD Ryzen Threadripper 2950X
Test Result
Test Result 9.8/10 Excellent June 2020
Test Result 9.6/10 Very Good June 2020
Manufacturer
AMD
AMD
Performance category
High-End CPU segment
High-End CPU segment
Cores
16 cores (3.5 GHz up to 4.7 GHz Max Turbo )
6 cores (3.8 GHz up to 4.4 GHz Max Turbo )
Performance for Gaming
Workstation (Applications) Performance
Overclocking possible?
Hyperthreading possible?
First release date
Q3/2019
Q2/2018
Max. RAM
DDR4-3200 up to 128GB
DDR4-2666 up to 128GB
Pros
  • Fastest prosumer CPU on the market
  • Compatible with most Socket AM4 motherboards
  • Completely unparalleled power efficiency
  • Native PCIe Gen 4.0
  • Offers more performance, uses less power and runs cooler than its Core i9 competition
  • Lots of horsepower packed into 16C/32T configuration
  • Works on existing Threadripper motherboards
  • Easy-to-use Ryzen Master software utility
Cons
  • Doesn’t always catch coffee lake in ST workloads
  • Requires quad-channel memory for optimum performance
Performance Winner
AMD Ryzen 9 3950X 16-Core, 32-Thread Unlocked Desktop Processor, Without Cooler
Model
AMD Ryzen 9 3950X
Test Result
Test Result 9.8/10 Excellent June 2020
Manufacturer
AMD
Performance category
High-End CPU segment
Cores
16 cores (3.5 GHz up to 4.7 GHz Max Turbo )
Performance for Gaming
Workstation (Applications) Performance
Overclocking possible?
Hyperthreading possible?
First release date
Q3/2019
Max. RAM
DDR4-3200 up to 128GB
Pros
  • Fastest prosumer CPU on the market
  • Compatible with most Socket AM4 motherboards
  • Completely unparalleled power efficiency
  • Native PCIe Gen 4.0
Cons
  • Doesn’t always catch coffee lake in ST workloads
Second Place
AMD Ryzen Threadripper 2950X Processor (YD295XA8AFWOF)
Model
AMD Ryzen Threadripper 2950X
Test Result
Test Result 9.6/10 Very Good June 2020
Manufacturer
AMD
Performance category
High-End CPU segment
Cores
6 cores (3.8 GHz up to 4.4 GHz Max Turbo )
Performance for Gaming
Workstation (Applications) Performance
Overclocking possible?
Hyperthreading possible?
First release date
Q2/2018
Max. RAM
DDR4-2666 up to 128GB
Pros
  • Offers more performance, uses less power and runs cooler than its Core i9 competition
  • Lots of horsepower packed into 16C/32T configuration
  • Works on existing Threadripper motherboards
  • Easy-to-use Ryzen Master software utility
Cons
  • Requires quad-channel memory for optimum performance

The AMD Ryzen 9 are the bosses of the upper middle class.

AMD’s best processor for modern software in the upper mid-range is the Ryzen 9, two models with currently 12 and 16 cores offering unprecedented performance on a four-year-old infrastructure, the AM4 socket.

Intel’s new mainstream flagship Core i9-10900K is also very fast, but with only ten cores and 20 threads it doesn’t stand a chance against the AMD Ryzen on average across all applications.

All in all the AMD Ryzen 9 3950X will be the better choice for any user really.

Especially from a performance point of view and the price-performance ratio point of view.

There are rumors AMD will bring out a Ryzen 4000 series at the end of the year.

We will keep you updated!


Are you interestet in more hot tech deals? Subscribe to our Email-list and we let you know when the best deals come up.

Leave a Comment