First 96-Core AMD Zen 4 Threadripper Tests Show Utter Domination over Intel

https://www.extremetech.com/computing/first-96-core-amd-zen-4-threadripper-tests-show-utter-domination-over-intel

by thelastgallon • 3 years ago

112 102 3 years ago

Racing04616• 3 years ago

Single core performance?

All i'm seeing is the confirmation that independently parallalizable tasks finish twice as fast on a cpu with twice the cores. cool, amdahl's law is proven.

The handbrake one, all the times seems similar indicating similar single core perf.

This matters for me since single core perf is pretty important for visual studio/other software with alot going on feeling "snappy" even tho the actual compiling/rendering might be faster and scale with moree cores.

wmf• 3 years ago

Single core should be not much worse than Ryzen.

Racing04616• 3 years ago

Yep seems about the same single thread wise which is a bit weird since 7950x3d base clock is 4.2 whereas the 7995wx is 2.5.

https://www.cpubenchmark.net/compare/5726vs5234vs3862/AMD-Ry...

FullyFunctional• 3 years ago

Threadripper is as the name implies, purely MT focused.

I run simulations that are inherently completely single-threaded and I lean on Geekbench which tend to be a good predictor for me. As far as I can tell, Intel still holds the crown in absolute ST perf, granted at horrific efficiencies: https://browser.geekbench.com/processor-benchmarks/ (be sure to ignore the "Top Single-Core Results" which AFAICT is full of BS numbers). Intel's latest 14900K isn't there yet, but should be a smidgen faster: https://www.techspot.com/news/100034-intel-core-i9-14900k-to...

elromulous• 3 years ago

All i'm seeing is the confirmation that independently parallalizable tasks finish twice as fast on a cpu with twice the cores. cool, amdahl's law is proven.

Yes, but if you can squeeze more cores in the same space (and/or power) envelope, and not suffer a performance per core penalty, you're winning.

wingmanjd• 3 years ago

"One big disparity between the three machines was the Zen 4 machine was outfitted with 512GB of memory, while its competition had just 128GB each"

This feels like cheating to me.

RecycledEle• 3 years ago

They had to give the Eypc 4 more RAM to populate all its RAM channels.

I've read it has 12 channels of DDR5.

bhouston• 3 years ago

I believe it is just 8

pyuser583• 3 years ago

Why? Extra RAM has no effect on most types of performance.

adgjlsfhk1• 3 years ago

These chips have 8 channel ram, so you need at least 8 sticks to get full ram bandwidth.

_joel• 3 years ago

They're testing CPU performance, not memory capacity. They need that much for the channels.

ahahahahah• 3 years ago

It just seems so incredibly sloppy to not have that be consistent. Do they expect all of their readers to then audit the code of the benchmarking that they are doing to ensure that it cannot possibly be bound by this?

mumblemumble• 3 years ago

If the benchmark has any cache misses at all then you also want to make sure everything about the memory configuration is the same, too, for a fair CPU benchmark, right? Number of channels, latency, etc.

No, the actual amount of RAM isn't directly relevant, but it seems like a possible brown m&m to me.

(https://www.snopes.com/fact-check/brown-out/)

vfclists• 3 years ago

I guess that depends on how memory bound the test is.

Dork1234• 3 years ago

Larger sticks (or more sticks) of RAM are generally slower than smaller sticks. It might not be possible to duplicate same speed ram and capacity on the Intel board. I would hope they duplicated the test with the same speed of RAM before matching capacity. The results are so one sided, I doubt either capacity or speed matters.

ksec6• 3 years ago

You are not testing for Memory capacity, you could have a test where the whole data fits in 16GB of memory.

But the memory is required there to populate the memory channel which push the CPU performance to max.

sva_• 3 years ago

Playing around with 96x avx512 sounds fun

MaxikCZ2• 3 years ago

Pardon my ignorance, I hear about "fun with avx512" so often yet still have 0 intuition about why its fun? What do people do with that instruction that's so different to all the other stuff?

fancy_pantser• 3 years ago

Professor Lemire has some fun ideas: https://lemire.me/blog/?s=Avx

112233• 3 years ago

https://gamozolabs.github.io/fuzzing/2018/10/14/vectorized_e...

mrichman2• 3 years ago

What use cases is this CPU ideal for? It's not a GPU, so gaming, ML model training, and crypto mining are out.

KiranRao0• 3 years ago

Shoving somewhere in a datacenter and virtualizing so that 96 different clients can use it at once.

cmrdporcupine• 3 years ago

But ThreadRipper is a workstation-targeted product, not server, isn't it?

stn8188• 3 years ago

I use my Threadripper for massive electromagnetic simulations for printed circuit boards and packages - I love it!

Philpax• 3 years ago

Compiling.

serf• 3 years ago

I like to use my lots-of-cores Xeons for media conversion and 3d rendering, it's nice to see that many cores spin up for SMP workloads.

pshirshov• 3 years ago

It's ideal for building big things, like Chromium or AOSP.

not a GPU

Well, 3970x is able to soft-render DotA in 4k with 8-11 fps...

stingraycharles• 3 years ago

I’m using it to still feel productive while working on our massive, TMP-heavy C++ codebase.

xcv1233• 3 years ago

Production 3D rendering is still done on CPU at VFX studios.

nineteen999• 3 years ago

Exactly, no guarantees that your scene will even fit in GPU VRAM given the complexity of some VFX shots.

bitwize• 3 years ago

These things can chew through compiles so fast I/O to disk will be the bottleneck, rather than actual compilation time. Great for Rust hackers.

mrichman2• 3 years ago

This is the most practical answer I've read thus far. However, I actually code via SSH into AWS EC2, so my local CPU is irrelevant for compiles.

adzm• 3 years ago

After Effects and Ableton will love this one

hackerlight2• 3 years ago

Simulations.

musicale• 3 years ago

Makes rust development more practical.

faeriechangling• 3 years ago

Allowing you to plug more GPUs into your motherboard.

coeneedell• 3 years ago

I’m looking at one for research purposes. Agent based simulation, non-GPU-implementable Ml modeling, large scale data processing. There are a lot of tasks in my work that are greatly sped up by having tons of threads.

Lockal• 3 years ago

Developers will love it. With limited resources nobody tests their code for compatibility: good testing requires an order of magnitude more cores. Test against multiple competing libraries? Compiled with multiple competitive alternative dependencies? Maybe it is also important to test dependencies with various flags? Maybe test in multiple Linux distributions? And then cross-product multiply all possible combinations and get the best result.

DrReachAround• 3 years ago

This will be perfect for my Plex server.

preisschild• 3 years ago

Why do you need so much compute in your Plex server?

Wouldn't a cheap GPU be better for transcoding tasks?

fortyseven• 3 years ago

Woosh.

mdhen• 3 years ago

I'm pretty sure that was a joke

userbinator• 3 years ago

IMHO, increasingly inefficient software notwithstanding, CPU performance has been well beyond acceptable since a long time ago.

Thus, the real question is how buggy these new CPUs will be. https://news.ycombinator.com/item?id=37812556

hetman• 3 years ago

I don't think someone doing video editing or 3D rendering is likely to agree with you here...

lmm2• 3 years ago

I don't think someone doing video editing or 3D rendering is likely to agree with you here...

They're hopefully not using a CPU for that - that's all been on the GPU for years.

CobaltFire2• 3 years ago

Tell that to my flight simulators. DCS has gotten better with the recent MT release, but the determinative requirements of that coupled with multiplayer servers in the 50+ population and all the AI brings most systems to their knees. Especially in VR…

hackerlight2• 3 years ago

CPU performance has been well beyond acceptable

For simulation applications there is no such concept as "acceptable", the utility of a CPU scales linearly with how fast it can do the calculations.

kristopolous• 3 years ago

AI systems are the new hungry monster. They're still pretty slow on expensive hardware and not everything is GPU based for reasons I don't fully understand. (I've got GTX/4090s and TR/3990x with 256GB of pretty fast ram and an NVME array for storage and have done a bunch of things related to AI/ML where I walk away for a few hours or even a day - both CPU and GPU get pretty taxed.)

CPUbenchmark is giving the new chip about a 100% improvement over the 3990x which is fairly tempting even at the 6-10k price tag I've seen floating around.

snakeyjake• 3 years ago

CPU performance has been well beyond acceptable since a long time ago.

Our antenna design folks sends simulation runs to a group of R7525s with 2x EPYC 75F3 CPUs and 2x A100 80GB GPUs.

A small problem might have tens of millions of elements and billions of unknowns to solve and run on one server. Larger problems might be spread across all of the servers, after coordination with everyone else.

Simulations take 8-12 hours to run.

This is not acceptable.

Anything slower than "an amount of time imperceptible to even the most responsive human" is unacceptable.

Everything is too slow.

Networks are too slow, memory is too slow, storage is too slow, GPUs are too slow, and CPUs are definitely too slow.

hedora• 3 years ago

Has anyone done a perf / watt scatterplot of the current AMD and Apple offerings?

I'd guess this wins in absolute performance (due to the TDP), but I've heard AMD is pretty competitive on energy efficiency too these days. I'm typing this on an M2, which is simultaneously cooler and faster than my old Intel laptop.

Wondering if it's time to upgrade my decade-old Intel Linux desktop to an AMD yet.

maccard• 3 years ago

I have an older threadripper - a 3970x. It's the most power hungry device in my house after my electric oven. In raw numbers it's hard to beat. But it is _not_ efficient

hamandcheese• 3 years ago

That is now a two generation old CPU at this point - I don't think we started seeing AMD get competitive on power efficiency until Zen 3.

jrockway• 3 years ago

I have the same chip and it's amazing to see it idle at 75W after growing up on Intel chips that use like 5W at idle.

Of note is that Zen 2 (the 3970x's microarchitecture) is 7nm. The new Threadrippers are on a 5nm process. I would expect slightly lower power consumption, but note that the 7000 series of Threadrippers quote a TDP of 350W while the 3000 series quote 280W. So I'm guessing they're using the efficiency for more speed instead of more power savings.

To me, it's fine. I'd always prefer paying more for electricity than to wait longer for something to complete. (But I do respect the desktop chips that can blink my cursor with an efficiency core that uses a few tens of milliwatts.)

dathinab5• 3 years ago

Zen 4 is known to be noticeable more efficient then Zen 3 and even much much more so then Zen 2 so I don't think this comparison is helpful.

pleoxy• 3 years ago

All of the AMD chiplet based designs do poorly at idle power.

These monsters will do better than the 39xx because of the 6mm IO die. But it's still going to have high idle power compared to a monolithic Intel design.

The monolithic AMD dies (laptop/mobile) are competitive in the idle power space.

protastus• 3 years ago

I have a 16-core Ryzen 7950X that I run in "eco mode 105W". This delivers performance close to an Intel 13900K but at less than half the power. It's also much easier to cool and get a quiet PC without water cooling and enthusiast shenanigans (de-lidding, etc.)

I don't know if the Zen 4 Threadrippers can be similarly run with a power limit. If they can, I'd expect much better efficiency than whatever Intel offers.

This has eco mode benchmarks: https://youtu.be/W6aKQ-eBFk0?si=o8l1qGUhSQKxmY_c

kcb• 3 years ago

Surprisingly haven't found great comparisons. A 7840U vs M2/Pro test would be pretty interesting.

https://www.notebookcheck.net/R7-7840U-vs-M2_15023_14521.247...

The last benchmark is interesting but unfair given that's an ancient application with Rosetta.

JohnTHaller• 3 years ago

I'd guess this wins in absolute performance

No need to guess: An M2 Ultra will get around 27,000 in Cinebench R23 while this Threadripper will get around 100,000.

The M2 Ultra has a TDP around 90 watts (~ish according to a couple reviews) and this Threadripper has a TDP of 350 watts.

smolder• 3 years ago

There's some more nuance to efficiency than just max power at max load, of course. The Apple chips will likely have good efficiency compared to the lower core count (say, 32) TR because the watts per core gets high on those smaller TR chips. Yet if you compared their performance at a fixed 90w, I suspect the threadripper would come out ahead, since chips are much more efficient when at fractional loads that don't lead to boosted core clocks. Then again, at idle, the TR is probably guzzling power relative to the M2 Ultra. I'm only speculating to illustrate; a thorough comparison of their perf/watt characteristics would involve sampling many types of workloads of varying thread counts and levels of per-core saturation.

arp242• 3 years ago

If you underclock the Threadripper to get a score of ~27k, then how much power will it use? I wouldn't be surprised if it was roughly equal to 90W, or even lower.

The stock configurations for a lot of desktop chips is to get the maximum performance at the expense of power, especially for things like these Threadripper series which are explicitly aimed at maximum performance. Often even drastically reducing power usage only gives modest performance penalties.

ksec6• 3 years ago

Has anyone done a perf / watt scatterplot of the current AMD and Apple offerings?

Tl;dr, if you want the absolute performance on Desktop, get the Threadripper Pro.

AMD Zen 4 7800X3D, 5nm 5.04Ghz, Single Core GB6 Score @ 2833

Apple A17 Pro, 3nm 3.80Ghz, Single Core GB6 Score @ 2914

Well you can find that with Geekbench but it is very likely worst to publish it as it leads to a wrong conclusion without the basic understanding of Hardware or CPU.

The node and design whether it was tuned for Clockspeed ( Desktop ) or SoC ( Apple ). There are measurable percentage difference in the same core for Risen on desktop and Ryzen SoC. And then you have to factor in Wattage difference in Clockspeed they are operating at. You are expected to know all these basics before the scatterplot would even make sense.

The Desktop 7800X3D will likely use 20W+ in the test, while the Smartphone A17 uses at best 5W.

You will then also have to factor it a lot of these performance test are not sustainable without adequate cooling. i.e You can sustain the above performance for a very long time on 7800X3D, dont expect that on A17 inside a phone.

Many of these are pointing out the obvious. But it seems on HN we are increasingly required to do so otherwise it leads to some sort of flamewar.

bitsandboots• 3 years ago

Anecdotally I find my m1 laptop and ryzen 5000-series laptop to have similar energy efficiency, but I don't have much hard data. When it comes to laptops idle efficiency is king and this is where it gets hard to test the chips because the rest of the motherboard also has a role.

For my m1, opening a vm or just plugging in a wireless mouse dongle is all it takes to shave several hours off the lifespan, like from 20 hours to 10.

For the amd laptop, anything waking up the nvidia gpu will do the same, like 10 hours to 3.

When you're talking such low amounts of power, the networking card, sound card, display efficiency, ssd efficiency, and whether the system chooses to run the fan or let the cpu stay warm starts to matter a lot since just like the cpu these are things that are running at the same time. So I guess a fair test would have to use the same ssd and have the monitor off, using an external monitor. Then you're just stuck with the efficiency of the parts of the motherboard you can't choose.

spandextwins• 3 years ago

Can't wait for Linus Tech Tips to do some great benchmarks on this thing!

Tostino• 3 years ago

Is this a joke? Isn't their whole thing pretty much lacking rigor in everything they do? Why would you want benchmarks from them?

extragood2• 3 years ago

Actually, no - their thing for the past few years has been to build out the most comprehensive test bench across all PC hardware components. "LTT Labs." They are aiming to become the authority on hardware performance and reliability, and have invested heavily in the tools and industry experts required to do that. Linus doesn't do any testing himself; he is just bankrolling it with the proceeds from his main channel and other media properties.

I find this whole thing super fascinating, coming from the perspective of Silicon Valley startups.

dathinab5• 3 years ago

Honestly that thing is a monster.

It's basically a AMD EPYC™ 9654 but with more clock boost speed and some differences wrt. IO (e.g. less but faster memory channels) and I think some differences wrt. enterprise features.

galaxyLogic• 3 years ago

For me as a developer it does not bother so much that some tasks like compilation or rendering or AI-training can take a long time. What bothers me is that I am doing multiple things at the same time like say committing to git or rebasing and trying to read my code at the same time, taking backups simultaneously. So this would seem like an ideal, if not overkill processor for me.

ksec6• 3 years ago

I am just wondering if it would be better and cheaper to buy an Threadripper 7980X 64 Core for Server than EPYC. You get higher single core performance. Specifically for those workload where bandwidth isn't an issue.

Something like StackExchange [1] could fit all their 9 server into 1 ( or 2 with one for redundancy ) without any degraded performance.

[1] https://stackexchange.com/performance

jiggawatts• 3 years ago

I keep trying to make the point that one or two servers could now put out the traffic of many top 100 websites, assuming they apps are written reasonably efficiently (e.g.: with C# or Rust, not Ruby).

Nobody believes me, which says a lot about both the hardware and software industries.

ksec6• 3 years ago

Yes. And we will have 128Core / 256 Thread Zen 4c Server soon.

I would be happy if Ruby Rails could do it in 2x the rendering time i.e slower with 2x Resources / Server. That is combined 4x difference. Unfortunately even with JIT we are not there yet.

throwaway81523• 3 years ago

It had 192 threads, drew 57 amps of current, had a heat sink the size of a Cadillac... they called it a macroprocessor.

That never gets old.

First 96-Core AMD Zen 4 Threadripper Tests Show Utter Domination over Intel

Related Stories