AMD Ryzen 9950X

Many reputable websites and channels “measure” the power consumption of Mac computers using apps like Max Power Gadget or asitop. These apps are built on top of the powermetrics MacOS facility. However, in the man pages of the facility it is clearly explained that the readings are estimated and not measured values and that they shouldn’t be used for making power consumption comparisons. In my performance investigation studies I found that the reported CPU and GPU power consumption is vastly underestimated: in case of the GPU the reported value is just over a third of the actual consumption, as is explained in the video. It is unfortunate that said reviewers do not bother to sanity check the numbers with a power meter and thereby perpetuate misleading information on Apple silicon performance/Watt. In a prior investigation, where I did extensive Linear Algebra performance investigations on the M4 Max, I found that it’s CPU performance/Watt in dense matrix multiplication is well below that of Ryzen 9950X, when the computations are carried out on its Neon vector units. Similarly for Apple’s M4 Max GPU vs nVidia last gen RTX 4090.

This sucks! I know more about software than hardware. Crucial is the only ram I have bought for decades now. As a practical matter does this mean one needs to buy DDR5 ram now from MicroCenter for a build planned for next year? I just put 128gb into my latest Linux workstation. I had been planning to build a new NAS to replace my aging TrueNas (nee FreeNAS). I was just thinking about possibly building another dev box after being very happy with this AMD 9950x performance.

I recently upped to a 9950X with a gen5 nvme.. TBH, even installing a few programs from cargo (which does compiles) is pretty quick now. Even coming from a 5950X with a gen4 drive.

Happy 9950X user here. Super happy with it, everything is crazy fast. Not a gamer, according to internet and benchmarks the extra cost is only worth it for gaming workloads. I use Arch, btw ;)

I think it was a Radeon 3600x, state of the art 6-7 years ago. Replaced with 9950x. I was surprised by how big of a difference the CPU update had on frame rates. (GPU: 4080) I also do a lot of rust compiling (Which you hinted at), and molecular dynamics sims leveraging a mix of CUDA/GPU, and thread pools + SIMD.

There are peaks in long-term CPU value. That is, CPUs that are 1) performant enough to handle general purpose computing for a decade and 2) outperform later chips for a long time. The i7-4770 was one. It reliably outperformed later Intel CPUs until near 10th gen or so. I know shops that are still plugging away on them. The first comparable replacements for it is the i7-12700 (but the i5-12400 is a good buy). At 13th gen, Intel swaps E for P cores. They have their place but I still prefer 12th gen for new desktops. Past all that, the author is right about the AMD Ryzen 9950x. It's a phenomenal chip. I used one in a friend's custom build (biz, local llm) and it'll be in use in 2035.

I bought one too. It's bananas how powerful the R9 9950X is. And it runs great with a €40 Thermalright cooler too - no need for AIO liquid cooling! It's amazing how great AMD CPUs got.

wat... my 9950x cpu that i just bought is way faster than the similarly priced cpu i bought 6 years ago. The difference is night and day. What are you talking about.

I just got a Crucial T700 last month which is a PCIE Gen5 x4 NVMe M.2 drive. I put it in an ASUS PRIME Z890M-PLUS motherboard with an Intel Core Ultra 7 265K Started to install Fedora Linux version 41. The drive would just completely disappear from the OS and the kernel would report I/O errors on a missing device. Sometimes this happened during the initial install. Sometimes 5 minutes after the install when starting a terminal. I couldn't even type "ls" because the "ls" command is on the drive that went away. Saw reports of PCIE Gen5 incompatibilities so I moved it to a Gen4 slot and then it worked. But the machine had so many other random crashes and errors reported in system logs saying "This is a hardware error not software" and stuff like that. Returned it all. Just got an AMD Ryzen 9 9950X and Gigabyte X870E AORUS PRO The Gen5 drive seems to be working at Gen5 speeds. lspci -vv shows 02:00.0 Non-Volatile memory controller: Micron/Crucial Technology T700 NVMe PCIe SSD (prog-if 02 [NVM Express]) LnkSta: Speed 32GT/s, Width x4

I've found compile times on large C++ code bases to be the only thing I really notice improving. I recently upgraded my work machine from a 2017 i7 to a shiny new Ryzen 9 9950x and my clean compile times went from 3.5 minutes to 15 seconds haha. When I compile with an M2 Max, it's about 30s, so decent for a laptop, but also it was 2x the price of my new desktop workstation.

I purposely picked a CPU with the same thread geometry as your 9900K to avoid calls of "apples & oranges" or whatever. If you want more threads, the 9950X is right there in the same socket. Or Core Ultra 9 285k. Either of which will run circles around a 9900K in code compilation. You can research microarchitecture differences if you want, it's a fascinating world, or you can just skip to looking at benchmarks/reviews. Little hard to compare against quite that large of a generation gap, but eg https://gamersnexus.net/cpus/rip-intel-amd-ryzen-7-9800x3d-c... or https://www.phoronix.com/review/amd-ryzen-7-9800x3d-linux/2

What is easy to do is to emulate FP128 with FP64 (double-double) or even FP256 with FP64. The reason is that the exponent range of FP64 is typically sufficient to avoid overflows and underflows in most applications. On the other hand, the exponent range of FP32 is insufficient for most scientific-technical computing. For an adequate exponent range, you must use either three FP32 per FP64, or two FP32 and an integer. In this case the emulation becomes significantly slower than the simplistic double-single emulation. With the simpler double-single emulation, you cannot expect to just plug it in most engineering applications, e.g. SPICE for electronic circuit simulation, and see that the application works. Some applications could be painstakingly modified to work with such an implementation, but that is not normally an option. So to be interchangeable with the use of standard FP64 you really must also emulate the exponent range, at the price of much slower emulation. I did this at some point in the past, but today it makes no sense in comparison with the available alternatives. Today, the best FP64 performance per dollar by far, is achieved with Ryzen 9950X or Ryzen 9900X, in combination with Inter Battlemage B580 GPUs. When money does not matter, you can use AMD Epyc in combination with AMD "datacenter" GPUs, which would achieve much better performance per watt, but the performance per dollar would be abysmally low.

While implementing double-precision by double-single may be a solution in some cases, the article fails to mention the overflow/underflow problem, which is critical in scientific/technical computing (a.k.a. HPC). With the method from the article, the exponent range remains the same as in single precision, instead of being increased to that of double precision. There are a lot of applications for which such an exponent range would cause far too frequent overflows and underflows. This could be avoided by introducing a lot of carefully-chosen scaling factors in all formulae, but this tedious work would remove the main advantage of floating-point arithmetic, i.e. the reason why computations are not done in fixed-point. The general solution of this problem is to emulate double-precision with 3 numbers, 2 FP32 for the significand and a third number for the exponent, either a FP number or an integer number, depending on which format is more convenient for a given GPU. This is possible, but it lowers considerably the achievable ratio between emulated FP64 throughput and hardware FP32 throughput, but the ratio is still better than the vendor-enforced 1:64 ratio. Nevertheless, for now any small business or individual user can achieve a much better performance per dollar for FP64 throughput by buying Intel Battlemage GPUs, which have a 1:8 FP64/FP32 throughput ratio. This is much better than you can achieve by emulating FP64 on NVIDIA or AMD GPUs. Intel B580 is a small GPU, so it has on…

Please try running this on a "desktop" CPU as well, like a Ryzen 9950X. (edit: sorry, didn't realise you shared the script at the end. i will test this myself :)) In my experience CI/CD tasks are more "single thread" bound that people expect. The Epyc CPU cores are _slow_ per core, so trading less cores for fewer faster ones actually works out well. So if you are wanting fast CI/CD builds you are much better off using desktop CPU cores vs enterprise server cpus like this. Wrote some thoughts up on this a while back https://martinalderson.com/posts/how-i-make-cicd-much-faster... - more focussed on the change from github to selfhosted runners, but I'd be interested to see a comparison on desktop class CPUs.

Beelink GTR9 Pro. It has dual 10G Ethernet interfaces. And get the 128GB RAM version, the RAM is not upgradeable. It isn't quite shipping yet, though. The absolute best would be a 9005 series Threadripper, but you will easily be pushing $10K+. The mainstream champ is the 9950X but despite being technically a mobile SOC the 395 gets you 90% of the real world performance of a 9950X in a much smaller and power efficient computer: https://www.phoronix.com/review/amd-ryzen-ai-max-arrow-lake/...

Well there's much more to memory performance than bandwidth. Generally applications are relatively cache friendly, thus the X3D helps a fair bit, especially with more intensive games (ones that barely hit 60 fps, not the silly game benchmarks that hit 500 fps). Generally CPUs have relatively small reorder windows, so a cache miss hurts bad, 80ns latency @ 5 GHz is 400 clock cycles, and something north of 1600 instructions that could have been executed. If one in 20 operations is a cache miss that's a serious impediment to getting any decent fraction of peak performance. The pain of those cache misses is part of why the X3D does so well, even a few less cache misses can increase performance a fair bit. With 8c/16 threads having only 2 (DDR4) or 4 (DDR5) cache misses pending with a 128 bit wide system means that in any given 80-100ns window only 2 or 4 cores can continue resume after a cache miss. DDR-6000 vs DDR-7800 doesn't change that much, you still wait the 80-100ns, you just get the cache line in 8 (16 for ddr5) cycles @ 7800MT/sec instead of 8 (16 for DDR5) cycles @ 6000MT/sec. So the faster DDR5 means more bandwidth (good for GPUs), but not more cache transactions in flight (good for CPUs). With better memory systems (like the Apple m3 max) you could have 32 cache misses per 80-100ns. I believe about half of those are reserved for the GPU, but even 16 would mean that all of the 9800X3Ds 16 threads could resolve a cache miss per 80-100ns instead of just 2 or 4. Tha…

While I mostly agree with what you say, I am quite happy that AMD uses their worse server chiplets for Ryzens, because this makes possible the assembling of very cheap but high-performance servers using e.g. 9950X, which has a much better performance per dollar than any Epyc or Xeon CPU or than any datacenter-oriented GPU. Such cheap servers are the right choice for small businesses and for individuals, for whom the great increases in the prices of the server CPUs that have happened starting in 2017, since the launch of the Skylake Server CPUs, have made them either unaffordable or not worthwhile. Because Intel cripples all their consumer CPUs, they cannot really compete with AMD for this purpose, except for some very restricted applications, e.g. for a server which does nothing else but code compilation, or for a server which does nothing else but hosting a Web server. Even for such applications, buying one of the very few motherboards for Intel that support ECC memory is likely to increase the price well over the AMD solution.

Important to me, faster on compilation than a 7950x (if you use a Linux desktop). Much faster, if you jump one generation, than the 5950x. While using less power (less fan noise while coding). It doesn't seem to be a good gaming chip (get an X3D), but it's a good chip for coding. https://www.phoronix.com/review/amd-ryzen-9950x-9900x/2 I do think memory bandwith might hold it back, Halo Strix is rumored to have double the memory bandwidth, we'll see how that works for coding.