AMD Threadripper 9980X 64-Core CPU Evaluation & Benchmarks
The short model up entrance is that the development on the 7980X is wherever from round 2% to that 58% quantity, however the overwhelming majority of exams are nearer to the vary of 5% to 18%. It actually closely is determined by the workload with these CPUs, significantly AVX occupancy.
These are workstation CPUs supposed for analysis, manufacturing work, or usually duties that both make cash or advance science or one thing related. We’re nonetheless benchmarking gaming, however the focus might be totally on manufacturing workloads immediately with gaming solely there to make sure there aren’t any main damaged outcomes.
These CPUs are costly. That limits the viewers. For that cause, we added all the brand new exams as a possibility to additionally re-run all the pieces else, so all of this information is model new. Even in case you’re not out there for Threadripper, it will have fascinating information from extra “regular” CPUs on the market.
Overview
Right here’s the short recap of the CPU lineup.
AMD is launching two variations of the Threadripper 9000 CPUs. These embody the non-Professional and Professional CPUs. The Professional CPUs had an earlier embargo; immediately, we’re overlaying the HEDT (or “high-end desktop”) non-Professional Threadripper CPUs.
The 9980X is a 64-core, 128-thread half that enhances as much as 5.4 GHz and has 256MB of L3 Cache. The CPU sockets into STR5 platforms and, like the opposite two fashionable Threadripper elements launching with it, has a 350W TDP. It’s $5,000 for the 9980X.
The 9970X is a 32-core, 64-thread half. All of them share the identical enhance of 5.4GHz, however the base clock is increased on the decrease core-count CPUs because of increased energy availability per core primarily based on energy finances. Cache is right down to 128MB L3 on the 9970X and 9960X, the latter of which is a 24-core, 48-thread half. Pricing is $2,500 for the 9970X and $1,500 for the 9960X.
In comparison with final time, costs are acquainted: The 7980X was $5,000, the 7970X was $2,500, and the 7960X was $1,500. Threadripper 9000 costs are elevated from a number of generations again to the 3000 sequence, however the identical because the 7000 sequence.
The architectural enhancements are the identical as in Zen 5 with the 9000 sequence launch. The most important one you’ll see immediately is from the AVX adjustments. After we had been digging into the biggest enchancment in our testing, AMD despatched this rationalization as a refresher on Zen 5:
“At first look [the test] does seem to learn from AVX-512 because the code has a number of references to ZMM. The primary distinction in 7K and 9K is the double vs single pump (256-bit native vs 512 native).”
“Excessive stage, it took two cycles to do a 512 bit operation on Zen4, however one cycle on Zen5 as a result of we doubled the FP datapath width, and elevated the bandwidth from the L2 to the core to have the ability to match that speedup. You possibly can’t at all times reap the benefits of it, however theoretically there’s a 2X enchancment.”
That “theoretically” is essential, as a result of it doesn’t present up in most of our exams; nonetheless, in some, there’s a very massive uplift. This isn’t particular to Threadripper and is as a substitute particular to Zen 5 vs. Zen 4. We’ll spend a while on that in our overview.
As for the take a look at bench: Our chart subtitles apply to the desktop platforms, however the Threadripper CPUs are examined with completely different reminiscence and motherboards. It deviates from among the different benches.
Let’s get into the benchmarks. We’ll begin with manufacturing testing.
9980X Manufacturing Benchmarks
7-Zip Compression
7-Zip compression is up now.
This take a look at is measured in MIPS, or tens of millions of directions per second. Increased is best throughput. The 9980X is the brand new chart-topper right here, at 520K MIPS versus the 7980X’s 508K MIPS. The development is simply 2.3% right here. This is without doubt one of the decrease enhancements.
There’s clear profit to increased core counts on this take a look at. The 7980X over the 7970X beforehand held a lead of 34%, or 67% over the 7960X.
Towards the highest-end desktop half, the 9950X3D, the 9980X sees an enchancment of 151%. Notably, further cache appears to assist right here: Each the 7950X3D and 9950X3D are benefitted over the 9950X.
Threadripper is in a completely completely different class from desktop parts, though the uplift during the last era’s 7980X is proscribed and comparatively boring right here.
7-Zip Decompression
In decompression, the 9980X scaled as much as 926K MIPS, gaining 8% on the 7980X’s 857K MIPS. This can be a higher exhibiting than what we noticed in compression and posts higher scaling. The remainder of the lineup is acquainted to the prior examples. Further cache doesn’t assist as a lot on this take a look at as in comparison with the compression take a look at, because the 7950X3D is now under the 9950X. This could align with the pace profit offering extra of an uplift on Threadripper as effectively.
SpecWS: OptionsPricing
SpecWorkstation 4.0 is subsequent. This runs numerous simulations and mathematical fashions that we aren’t consultants in, however we are able to no less than run the benchmarks. We’ll depend on the Spec web site to elucidate what the exams are.
The primary is Choices Pricing. The web site says that this runs Monte Carlo probabilistic simulations for monetary uncertainty, Black-Scholes pricing fashions for theoretical worth, and binomial choices pricing. We aren’t consultants in how these are used, however these of you in our viewers who’re consultants have repeatedly expressed an appreciation for this take a look at.
The outcomes have the 9980X on the prime, with a 7.8 rating. That has it improved on the 7980X by a powerful 18%, aligning with the findings of the 9000 sequence desktop elements in opposition to the 7000 sequence on this identical take a look at. You possibly can see that within the 9800X3D in opposition to the 7800X3D or the 9900X3D vs. 7900X3D. That is an architectural profit. It’s one of many uncommon break-outs for Zen 5 that we talked about after the preliminary spherical of evaluations once we had validated it as a authentic deviation.
The 9980X improves on the highest-end desktop 9950X3D by 200%, principally being the distinction between being runnable and never. Threadripper actually makes an impression right here, and the generational achieve is actual however uncommon on this one take a look at.
SpecWS: OpenFOAM
OpenFOAM is subsequent. Per the Spec web site, OpenFOAM is a CFD take a look at that “performs 2D Reynolds averaged simulation for fluid dynamics evaluation.” It additionally runs a solver take a look at.
The 9980X scores 23.3 right here, main the 7980X prior era by about 9%. Not as spectacular as Choices Pricing, however higher than another exams we have now immediately.
The lead over desktop elements just like the 9950X and X3D is sufficient that it’s actually not in the identical class, with multiples of uplift.
SpecWS: Convolution
Convolution testing is subsequent. Spec says that this take a look at applies “a convolution filter to a picture, a crucial operation in sign processing and picture evaluation” for edge detection and have extraction. That is closely multi-threaded.
On this take a look at, the 9980X scored 9.3 to the 7980X’s 5.9, an enormous uplift of 58%. That is utterly out of alignment with every other exams. We contacted Wendell of Level1 Techs and AMD’s engineers to get each a first-party and a impartial third-party examine of those numbers. Wendell famous that that is anticipated conduct and had seen related leads to different exams, believing it to be associated to AVX enhancements along with being a take a look at that runs lengthy sufficient to learn from them. AMD famous that it has additionally seen outcomes round 60% for sure related exams.
Though we have now our 9970X overview developing individually, we’ll observe that it additionally noticed a 56% enchancment right here over the 7970X predecessor. The outcomes align. final gen, the 9950X improved on the 7950X by 50%, which is constant generationally. The outcomes look like constant and are current throughout quite a few CPUs cross-generation, which implies the one factor left to contemplate is whether or not this benchmark represents real-world situations. We’re undecided: This explicit take a look at exits our experience, as we don’t do sign processing or picture evaluation work. We’d love to listen to from these in our viewers as as to if the Spec Convolution take a look at aligns with real-world purposes that you just use in your work life.
Regardless, the result’s repeatable — it’s solely a query of whether or not it’s successfully a microbenchmark or is one thing that scales to real-world use.
SpecWS: LAMMPS
LAMMPS testing is subsequent. In keeping with Spec, this take a look at is for large-scale atomic and molecular massively parallel simulation, which Spec says “makes use of MPI to scale to a number of cores to carry out advanced simulations.” This contains life science subtests, comparable to for polymer chain calculations and protein calculations for, what we assume are, medical makes use of.
LAMMPS has the 9980X at a 5.5 rating in Spec, main the 7980X by 17%. That is near the Choices Pricing outcomes. Though Spec has a number of exams the place there’s no actual impression, like Handbrake, it additionally has the next saturation of exams with bigger generational enhancements.
SpecWS: Information Science
The Spec Information Science take a look at is subsequent. Spec says that this take a look at “entails working a sequence of artificial benchmarks that simulate real-world AI and ML workflows utilizing the Python-based information science libraries Pandas, Scikit-learn, and XGBoost.”
Generationally, the 9980X improves on the 7980X by 10%, going from 2.0 to 2.2 within the scoring. For reference, the 9950X improved on the 7950X by 13% at 1.7 from 1.5.
SpecWS: NAMD
NAMD is subsequent, one other take a look at our viewers appreciated earlier than. This can be a molecular dynamics simulator for modeling “conduct of biomolecular methods,” with Spec saying that it’s “used for analysis in biochemistry, pharmacology, and molecular biology.”
The 9980X scored 6.0, with the 7980X at 5.1. That is an enchancment of just about 18%, which aligns once more with LAMMPS and choices pricing benchmarks.
SpecWS: RodiniaCFD
We now have much more Spec outcomes, however we’ll shut on one which’s much less fascinating to deliver it again right down to earth and stability the upper outcomes. That is for RodiniaCFD, which the positioning explains has a pre-euler subtest that “executes an unstructured grid finite quantity CFD solver for 3D euler equations for compressible movement.”
The 9980X scored 3.7 right here, with the 7980X at 3.5. That is one the place the development is much less thrilling, again right down to a couple of 5-6% enchancment generationally.
Chromium
Up subsequent is Chromium code compile. There are plenty of varieties of code compile, similar to there are plenty of varieties of video encoding or photograph modifying or 3D rendering workloads, so we’re simply representing one right here as we do with every other manufacturing take a look at.
The 9980X completes the compile the quickest, at 48 minutes complete. The compile time is 7.5% diminished from the 7980X. Elsewhere, with the newer model of this take a look at, we’re seeing minimal impact in comparison with prior take a look at iterations. The 7980X and 7970X are comparatively shut to one another, with the 9950X3D and 9950X additionally close to one another. A few of these CPUs are working with diminished reminiscence pace as they required increased reminiscence capability and had extra bother holding our standard settings, so these are famous the place true.
The 14900K is Intel’s greatest performer, requiring 100 minutes to finish the compile. The 9950X outperforms this.
Blender
Blender tile-based rendering is up subsequent.
For this course of, the 9980X is the brand new chart-topper at 2.1 minutes to finish a render of a single body from the GN intro animation emblem. The following quickest is the 7980X at 2.5 minutes, which is sensible: This take a look at is closely thread-dependent, so threads will outrank frequency the place massive variations emerge.
Intel’s newer era does OK on this take a look at, with the 285K between the 9950X and 7950X. It’s additionally one of many exams the place Intel doesn’t see regressive efficiency, although we nonetheless wouldn’t advocate the 285K. That complete platform is lifeless, anyway.
Photoshop
Adobe Photoshop is subsequent, examined with the Puget Suite. This makes use of an mixture rating of filters, resizes, and different capabilities inside Photoshop.
The very best performer right here is the 9950X, adopted by different desktop CPUs. The Threadripper elements fall additional down the chart, with the 9980X particularly underperforming. We’ve seen this previously, comparable to with the 7980X: They simply have too many threads for their very own good and Photoshop isn’t capable of make the most of them. The 7980X is outperformed by the 7970X and 7960X, so the 9980X touchdown within the center isn’t a shock. Its primary enchancment comes from the frequency.
It does OK right here if this is only one of many purposes you employ, but when Photoshop is the one factor you do, there are higher and cheaper CPUs for the job.
Premiere
Adobe Premiere is subsequent, our greatest frenemy that we see daily.
On this one, the 9980X improves upon the efficiency from the 7980X and 7970X, however solely barely. It’s positively not value a single generational improve, although we usually don’t advocate that anyway.
The development is nearly 5% from the 7980X to the 9980X within the prolonged rating for Premiere. Intel’s 285K trails most Threadripper CPUs within the chart and roughly ties the 7960X. It additionally technically outperforms the 9950X. Intel’s greatest benefit stays in its QuickSync resolution with the IGP, which might present up disproportionately in additional focused exams quite than the total sweep performed right here.
Threadripper does effectively total right here and is no less than higher than its Photoshop exhibiting, however with out different purposes leaning on threads, a desktop CPU would get a lot of the identical efficiency. The very best use for Threadripper right here can be assigning its extra threads to different duties, like separate audio compositing, whereas working Premiere duties.
DaVinci Resolve
DaVinci Resolve is subsequent. We haven’t examined this in years, so that is an experimental chart.
On this take a look at, the 9980X’s frequency enhance permits it to realize on prior Threadripper CPUs. The earlier lineup had the 7960X and 7980X tied, with the 7970X marginally forward. We noticed this with the 7000-series Threadripper evaluations the place 64-core elements typically prompted slight losses in comparison with 32-core counterparts.
The 285K falls behind the 9950X on this one, although not by so much, with the 14900K roughly inside re-run distance of the 285K.
Threadripper is no less than the perfect performer right here as effectively, regardless of not offering clear worth over desktop elements. It’s simply not constructed for these sorts of workloads, however nonetheless being the perfect is essential.
Gaming Benchmarks
We’ll maintain gaming brief. The one goal of gaming exams on Threadripper, usually, is to verify there isn’t some main downside, however cheaper desktop elements are going to be higher for pure gaming builds. Threadripper is definitely detrimental if all you do is gaming.
You are going to discover that the Threadripper chips do not at all times comply with a strictly logical order, particularly once they’re inside a couple of frames per second of one another. That is due to the elevated run-to-run variance inflicting much less significant averaged outcomes.
Stellaris Simulation Time
Stellaris simulation time is first. The 9800X3D is the present chief at 37.1 seconds on common for simulation time. The 9700X follows this at 43.2 seconds, benefitting from Zen 5 architectural enhancements over the prior era. The 285K isn’t what we’d name “good” right here, however no less than matches its predecessor.
Threadripper is approach on the backside, with the 9980X no less than bettering on the 7980X notably, but it surely’s nonetheless far worse than every other fashionable gaming CPU. The 5700X is almost on the stage of the 9980X. It’s simply detrimental to have this CPU configuration for this recreation, which isn’t actually information to anybody. It will probably no less than play video games, however the distinction in simulation time between the 9800X3D and the 9980X can be noticeable in actual recreation play.
FFXIV: Dawntrail – 1080p
Ultimate Fantasy 14: Dawntrail is subsequent. The 9980X and 7980X excessive core-count CPUs had some main inconsistency points on this recreation, so there’s an asterisk subsequent to their entries.
We’re nonetheless exhibiting the averages of 4 runs, it’s simply that the runs are extra variable than sometimes. The 9980X, for example, had a variety from 274 to 248 FPS AVG, which is a big swing. Our typical customary deviation on this recreation is a pair frames per second for the common. The sport was playable, however all over for the common framerate. Lows had been worse total as effectively, however once more, technically playable.
For this reason we nonetheless take a look at video games with Threadripper, although.
Dragon’s Dogma 2 – 1080p
In Dragon’s Dogma 2 at 1080p, the Threadripper CPUs find yourself within the decrease half of the chart. The 9980X is worse than the 7950X right here and about the identical because the 7970X. The 7980X was under the 7960X and 7970X, indicating that these extraordinarily excessive core counts are detrimental to gaming efficiency. The upper frequency issues greater than the additional cores, so the trade-offs are dangerous to gaming efficiency right here. This has been true without end with Threadripper, so none of that is information.
What issues extra is that they’re nonetheless playable. Threadripper does undergo from worse lows, significantly 0.1%, however not so dangerous that they render the sport stuttery. You might nonetheless play video games on a Threadripper system, however once more, gaming-only customers can buy one thing else.
Cyberpunk 2077: Phantom Liberty – 1080p
Cyberpunk is subsequent. On this one, the 9980X outperforms the 7980X on a technicality, however each are nearer to the R7 2700 than they’re to the 9800X3D. The run-to-run consistency was nice, with the 9980X solely exhibiting a 1.9 AVG FPS customary deviation, so it’s not like what we noticed in Ultimate Fantasy — it’s simply not significantly good. That is nonetheless a satisfactory framerate, so within the context of a pc constructed for particular work duties that may nonetheless play video games when wanted, it passes that want.
Baldur’s Gate 3 – 1080p
In Baldur’s Gate 3, the 9980X finally ends up across the stage of the 265K — which is de facto simply extra embarrassing for Intel than anything. We already know Threadripper isn’t significantly good at gaming, regardless of being OK at it, however the 265K must be doing higher right here. That is additionally a recreation the place the 285K was massively regressive, although, at 109 FPS AVG in opposition to the 14700K’s 117 FPS AVG and 14900K’s 123 FPS AVG.
We did take a look at different video games, however they’re simply not related past proving the purpose we have already got: Threadripper is satisfactory if gaming isn’t your major want. It’s simply not your best option for it.
Energy Consumption
9980X Energy Consumption – Blender
This subsequent chart reveals the ability consumption as measured on the EPS12V cables with an influence interposer between the motherboard and energy provide. Examined with an all-core workload in Blender, we noticed secure and constant energy draw throughout each the 7980X and 9980X CPUs, with the 9980X measuring at 371W from 354W on the last-gen Threadripper flagship. That’s plenty of energy to chill.
This doesn’t present transients.
9980X Thermals
Thermals – Blender
This chart reveals the thermal conduct of the 9980X. We’re utilizing a SilverStone 360mm liquid cooler with a Threadripper-sized coldplate for this, so the outcomes aren’t similar to our 7980X overview thermal outcomes from its launch. Additionally, as standard, thermal testing can be greatest performed with plenty of coolers in contrast and noise-normalized, however that is stored easy for the overview and to assist get an concept (as that’d be a cooler overview).
The 9980X beneath an all-core workload ran at round 58 levels Celsius when working the comparatively loud 100% fan speeds on the 360mm liquid cooler.
Thermals – Equilibrium
This chart reveals thermals at regular state throughout the assorted CCDs. The CCD-to-CCD delta right here is 6.7 levels Celsius, at 49.7 to 56.1 levels in an ambient of 21 levels Celsius, +/- 1C. Usually talking, this may be cooled at decrease noise ranges with 360mm coolers with out main points if assuming a well-ventilated case. The floor space is so large on this CPU that the ability is distributed throughout a big space, making it simpler to chill than a 200W warmth load is perhaps on a standard desktop-sized Ryzen CPU with 2 CCDs.
9980X Frequency
Frequency – Blender All-Core
With all cores working, the CPU sometimes ran at round 3950 MHz, with occasional spikes to 4200 MHz whereas some threads had been bringing in information for brand spanking new tiles to render. The bottom clock is 3.2 GHz, so it’s not dropping that low throughout this explicit workload, but it surely’s additionally far-off from 5.4 GHz. That is typical conduct for all-core workloads.
Frequency – Blender vs 7980X
This chart reveals the 7980X for a similar take a look at. The frequency bottoms-out at round 3.7 GHz, however is all over throughout this workload. The 9980X has improved upon the frequency in all-core situations total.
Frequency – CB Single-Thread
In a single-threaded take a look at with Cinebench, the 9980X ran at 5425 MHz, exceeding the spec itemizing of 5.4 GHz for enhance. That is additionally anticipated conduct, with the extra restricted load boosting increased. The 7980X in the identical take a look at ran at about 5340 MHz, so the single-thread frequency has improved generationally — however not a lot, and that’s why some benchmarks solely present a pair share factors of change.
9980X Conclusion
On the gaming facet, there are typically points with consistency. This implies, run-to-run, you don’t get the identical expertise. There’s much more deviation right here in common body fee and the frametime consistency is worse than conventional desktop elements. That is very true with 64-core Threadripper CPUs. This isn’t actually surprising. These excessive core depend CPUs typically simply have so many cores that it begins inflicting scheduling issues. Generally you’ll additionally find yourself in conditions the place, as a result of the clocks aren’t boosting as excessive (due to the upper core depend), you’re shedding that frequency benefit. Once more, this isn’t surprising. The excellent news is that the video games we examined on the 9980X weren’t damaged. They had been capable of run and weren’t terrible experiences, which wasn’t the case with the earliest Threadripper CPUs.
In comparison with the 32-core Threadripper, the 64-core half technically runs cooler. That’s since you’re taking the identical energy and distributing it throughout extra silicon. This implies you’re going to get decrease temperatures per die per scorching spot. That is still true right here the place the 64-core CPU might be cooler than the 32-core processor.
With its costs, these Threadripper CPUs aren’t one thing you’re shopping for for “worth.” The goal demographic for these processors might be getting cash with their computer systems.
Threadripper additionally has the advantage of PCIe lane availability for machines with a number of accelerators or plenty of I/O, that are extra restricted with conventional desktop CPUs.
64 cores don’t at all times scale effectively. Generally a 32-core CPU is usually a higher match. Be sure that to do analysis on-line to your wants. For instance, we discovered {that a} 32-core CPU would work higher for us whereas utilizing Adobe Premiere for video manufacturing. So it’s greatest to analysis these CPUs along with your wants.