#threadripper

[blubberquark]

Share Your Anecdotes: Multicore Pessimisation

I took a look at the specs of new 7000 series Threadripper CPUs, and I really don't have any excuse to buy one, even if I had the money to spare. I thought long and hard about different workloads, but nothing came to mind.

Back in university, we had courses about map/reduce clusters, and I experimented with parallel interpreters for Prolog, and distributed computing systems. What I learned is that the potential performance gains from better data structures and algorithms trump the performance gains from fancy hardware, and that there is more to be gained from using the GPU or from re-writing the performance-critical sections in C and making sure your data structures take up less memory than from multi-threaded code. Of course, all this is especially important when you are working in pure Python, because of the GIL.

The performance penalty of parallelisation hits even harder when you try to distribute your computation between different computers over the network, and the overhead of serialisation, communication, and scheduling work can easily exceed the gains of parallel computation, especially for small to medium workloads. If you benchmark your Hadoop cluster on a toy problem, you may well find that it's faster to solve your toy problem on one desktop PC than a whole cluster, because it's a toy problem, and the gains only kick in when your data set is too big to fit on a single computer.

The new Threadripper got me thinking: Has this happened to somebody with just a multicore CPU? Is there software that performs better with 2 cores than with just one, and better with 4 cores than with 2, but substantially worse with 64? It could happen! Deadlocks, livelocks, weird inter-process communication issues where you have one process per core and every one of the 64 processes communicates with the other 63 via pipes? There could be software that has a badly optimised main thread, or a badly optimised work unit scheduler, and the limiting factor is single-thread performance of that scheduler that needs to distribute and integrate work units for 64 threads, to the point where the worker threads are mostly idling and only one core is at 100%.

I am not trying to blame any programmer if this happens. Most likely such software was developed back when quad-core CPUs were a new thing, or even back when there were multi-CPU-socket mainboards, and the developer never imagined that one day there would be Threadrippers on the consumer market. Programs from back then, built for Windows XP, could still run on Windows 10 or 11.

In spite of all this, I suspect that this kind of problem is quite rare in practice. It requires software that spawns one thread or one process per core, but which is deoptimised for more cores, maybe written under the assumption that users have for two to six CPU cores, a user who can afford a Threadripper, and needs a Threadripper, and a workload where the problem is noticeable. You wouldn't get a Threadripper in the first place if it made your workflows slower, so that hypothetical user probably has one main workload that really benefits from the many cores, and another that doesn't.

So, has this happened to you? Dou you have a Threadripper at work? Do you work in bioinformatics or visual effects? Do you encode a lot of video? Do you know a guy who does? Do you own a Threadripper or an Ampere just for the hell of it? Or have you tried to build a Hadoop/Beowulf/OpenMP cluster, only to have your code run slower?

I would love to hear from you.

[phonemantra-blog]

The company reported for the third quarter of fiscal year 2023 A few days after Intel, AMD also published its financial report. The company has completed the third quarter of fiscal year 2023. [caption id="attachment_78911" align="aligncenter" width="780"] AMD's[/caption] If Intel's revenue decreased and net profit fell by more than 70%, then AMD's situation is better. Revenue rose 8% to $5.8 billion, and operating profit amounted to $224 million versus a loss of $64 million a year earlier. Net income rose more than 1,000% to $299 million. AMD's net profit grew by 1000% and equaled Intel's net profit By the way, Intel's net profit was only $1 million higher, which is very interesting. At the same time, Intel's revenue is almost 2.5 times higher than that of its competitor. It’s also worth noting that the data center products segment did not grow year-on-year in the past quarter, although it has been actively driving AMD’s revenue upward in recent years. But sales in the consumer segment increased by 42%, mainly due to the full entry into the market of Ryzen 7000 mobile APUs. AMD's gaming business fell 8% due to lower revenue from sales of semi-custom products.

[gobusto]

i can't take AMD "Threadripper" CPUs seriously because Threadripper the Hedgehog sounds like the name of a Sonic OC whose whole deal is that he's like a more edgy version of Shadow.

[govindhtech]

AMD’s Next-Gen TRX50 Motherboards

AMD Threadripper 7000 central processor units will be supported by the next-generation TRX50 motherboards that are currently being developed by AMD’s board partners. These motherboards are now in the process of entering mass production.

Threadripper 7000 HEDT & Workstation CPU AMD TRX50 Motherboards Have Been Found

This time around, ASRock is working on the TRX50 WS, a motherboard that will employ AMD’s TRX50 socket and WS chipset and will provide compatibility for the most current Threadripper 7000 “Storm Peak” processors.

ASRock always has its full stack prepared, regardless of whether they are working with mainstream or workstation systems. This motherboard has been listed by the EEC (Eurasian Economic Commission), and it is projected that it will be accessible in the fourth quarter of 2023, when the lineup is expected to be on shop shelves.

As suggested by its name, this particular motherboard appears like it will be focusing on the workstation side of things as its primary market. ASRock’s current TRX40 product portfolio included the Taichi and Creator products. These weren’t only geared at providing workstation-class functionality, rather, they were also influenced by gaming aesthetics.

It has been four years after the debut of the TRX40 series. However, at the same time, ASRock was also releasing the WRX80 series as part of the Creator family. This series was built specifically for “WX” PRO chips. This motherboard had support for eight channels of RAM and included a generous number of PCI Express lanes.

It’s possible that the same thing will happen here with the AMD TRX50 motherboards, which will keep the lineup basic and make it more user-friendly for high-end desktop computers, while the WRX series range will concentrate on providing customers with an abundance of functionality to meet the needs of the workstation market. That there will be two platforms for the Threadripper 7000 CPUs is something that we are aware of, and something that we have also very recently verified with a variety of motherboard manufacturers.

One will support 4-channel memory, while the other will support 8-channel memory. While the HEDT platform is expected to have the LGA 4844 “SP6” socket with 4-channel memory and 64 PCIe Gen 5 lanes support, the WS platform is expected to have the LGA 6096 “SP5” socket with support for up to 8 channels of memory and 128 PCIe Gen 5 lanes. Similarly, the majority of motherboard manufacturers specialized on HEDT “TRX” components rather than the Workstation SKUs. With the next-gen lineup, it may be the case as well.

Late in the year 2023 is when it is projected that the next-generation AMD Ryzen Threadripper 7000 “Storm Peak” central processing units would be introduced for the first time, provided that all goes according to plan. You should prepare to get further information sometime during the next several months.

[omarvektrapc12]

Platform

Desktop

Product Family

AMD Ryzen™ PRO Processors

Product Line

AMD Ryzen™ Threadripper™ PRO 5000 WX-Series

AMD PRO Technologies

Yes

Consumer Use

No

Regional Availability

Global, China, NA, EMEA, APJ, LATAM

Former Codename

"Chagall PRO"

Architecture

"Zen 3"

# of CPU Cores

32

# of Threads

64

Max. Boost Clock

Up to 4.5GHz

Base Clock

3.6GHz

[arielmcorg]

#AMD - Nuevos Productos de Estaciones de Trabajo AMD Threadripper

AMD anunció los tan esperados procesadores AMD Ryzen Threadripper PRO Serie 7000 WX y reintrodujo la línea de procesadores Threadripper en el espacio de escritorio de gama alta con los procesadores Ryzen™ Threadripper™ Serie 7000, estableciendo un nuevo estándar para el rendimiento informático y la innovaci��n en la industria (Fuente AMD Latam). AMD presenta los procesadores Ryzen Threadripper…

[babeseungmin]

boy I always feel like giffing takes me forever? How are people so fast? 

[miscellaneousmathematics]

Doing server maintenance tonight. At the end of the night I should have 7 different computers all hooked up together. Only problem is I have no idea what I want to do with all this computing power. I essentially have a cheap AMD Threadripper at my disposal.

Maybe I'll keep banging my head against Fortran to try and get my neural network to actually function. I'm running into problems with how to store the neural network data. Fortran doesn't have dynamically sized arrays so I have to either put in a big high rank array or have a bunch of smaller arrays with less manageable code.

[rokkazu]

everyone whose ever designed a piece of clothing where the annoying itchy as fuck tag is just sewn in over with the same stitch that holds the actual fabric together so you can't even get it off with a threadripper without putting a hole in the back of your fucking shirt should die and go to hell forever

[phonemantra-blog]

Already next week The Starfield game, as you know, was created with the active support of AMD and works much better on Radeon video cards. However, owners of GeForce adapters will soon have a reason to rejoice, as DLSS will be added to the game. [caption id="attachment_78515" align="aligncenter" width="780"] AMD's[/caption] AMD's Starfield Game Will Get DLSS 3 Before FSR 3 Next week the developers will release a beta version of the update, which will bring many changes. True, it’s not very clear why the beta version, but we just have to wait for the release. Among other things, the update will bring support for DLSS, and judging by the mention of frame generation, it will be DLSS 3. Bethesda also promised to add FSR 3 to the game, that is, frame generation technology will become available for Radeon adapters. Considering that Starfield turned out to be very demanding, such improvements will clearly not be superfluous.

[amongussexgif]

The post about computer-building resources was SUPER helpful!!!! thank you so much!!!

I know that the way to go for desktops tends to be to build your own- but I've always been a bit preferential to laptops!

AFAIK, there aren't really a ton of resources for how to build your own laptop... would you happen to know of any? Is building a custom laptop something you would recommend, or not really?

I dont know much about vuilding a custom laptop, sorry! I do know that you'll need entirely different parts, like a mobile gpu and ram, but beyond that, Im about as lost as you. If you ask me, though, laptops are only good for business if you need important work to be portable, as they tend to overheat (especially ones from Apple, Dell, and HP). If you want a gaming rig or other and you want it to be small or convenient, I'd consider also looking at Mini-ITX from PCs! They're just ubder a cubic foot and can be just about as powerful as you want (unless you're rich enough to want a Threadripper build)

TLDR, sorry, I got bothing about laptops.

I did hear that there are a few cool nodular gaming laptops, though. Might want to check those out.

[heavenpierceher]

youtube to me is the speedrun analysis+vintage hardware decompilation videos machine. every video i watch is titled some shit like Racing the Beam Explained - Atari 2600 CPU vs. CRT Television or SSL Pyramid from Lower Entrance A Presses (Inside 1xA, Puzzle 1xA) or 100 Copies of Doom vs Threadripper 7995WX PRO 96-core 192-thread CPU or some shit. I think my version of the website is better

[demigirlboss2girlrotboogaloo]

Honestly, this can't last. It'll collapse eventually, and when it does a *lot* of things are going to end up in landfills. Cloud computing was supposed to be cheaper, easier, more scalable, but as time's gone on the price gouging by both hardware vendors and the hyperscalers that provide the services has made it less and less viable for most people and organizations to opt for cloud storage and compute. I can't find the exact quote right now, but Wendell from Level1Techs once made a really good point about how it used to be that you would just buy a thinclient, and hook it up to a cloud compute server, and you wouldn't have to worry about stocking and maintaining expensive hardware. But nowadays, the cost of cloud storage and compute and honestly even the thinclients is so ridiculously astronomical that in a lot of cases organizations are better off just getting high power NUCs instead, since while yes a fleet of those might cost you fifty-odd thousand dollars, you will have already made that money back in a year from all the cloud subscriptions you aren't paying anymore.

And again, as much as hyperscalers like IBM and Google are to blame for this, the hardware vendors are equally to blame. Nvidia's getting as high as 1000% margins on some of their products, and AMD isn't far behind. Intel is thankfully still willing to price themselves into categories that their competitors sneer at, but that's only gonna last until they get established in the dGPU business again (shoutout Knight's Landing my beloved), which would also propel their CPU offerings higher (and frankly at the moment I trust Sapphire Rapids more than I do Threadripper, don't think we forgot about what you did to TR5000 you bastards), which are significantly higher margin products due to being produced directly by Intel Foundry Services rather than an external fab (namely TSMC) like their GPUs are.

So yeah, give it a bit, the bubble will burst soon enough and if it doesn't hopefully USB4 PCIe passthrough catches on on phones and other mobile devices. Fingers fucking crossed. And if we're feeling brave let's maybe pray that USB-IF writes a USB standard that actually includes all the features instead of me having to buy Thunderbolt cables.

smartphone storage plateauing in favor of just storing everything in the cloud is such dogshit. i should be able to have like a fucking terabyte of data on my phone at this point. i hate the fucking cloud

[onedirectdeals]

Carte mere GIGABYTE *TRX40 DESIGNARE* sTRx4/D4*8/M.2/USB3.1/WIFI6 ATX* 7702

Price: Buy Now Last Updated: AMD TRX40 DESIGNARE Motherboard with Direct 16+3 Phases Infineon Digital VRM, Fins-Array Heatsink, Dual Intel LAN, 4 PCIe 4.0 M.2 with Thermal Guards, Bundled GC-Titan Ridge Add-In Card and AORUS Gen4 AIC Adaptor, Intel® WiFi 6 802.11ax,120dB SNR AMP-UP Audio Supports 3rd Gen AMD Ryzen Threadripper ProcessorsSupports 8 x DDR4 Quad Channel Non-ECC Unbuffered DIMMs16+3…

[granth]

Demystifying AI-Driven PC Configurations for Professional Use

Introduction

AI-driven workloads, such as machine learning, data analysis, and neural network training, require specialized PC configurations to handle massive computational demands efficiently. This case study explores Kryptronix's approach to building PCs optimized for AI professionals, ensuring cutting-edge performance and scalability.

1. The Challenges of AI Workloads

AI professionals face unique challenges when working with resource-intensive tasks:

High Computational Power Needs: AI algorithms require GPUs with thousands of cores.

Memory Management Issues: AI processes often need large-capacity RAM for smooth execution.

Data Storage: Managing large datasets demands both speed and storage capacity.

Thermal Regulation: Sustained heavy loads generate significant heat, risking hardware failure.

These requirements highlight the need for customized, AI-optimized PC builds.

2. The Ideal AI-Driven PC Configuration

Custom PCs for AI are built with components designed to handle intense workloads:

Multi-Core CPUs: Essential for managing complex computational processes.

High-End GPUs: NVIDIA RTX and Tesla series for GPU-accelerated tasks.

Ample RAM: Minimum of 64GB for efficient processing of large datasets.

Fast Storage: NVMe SSDs for quick data retrieval and transfer speeds.

Advanced Cooling Systems: Ensures stability during prolonged operations.

3. Kryptronix’s Expertise in AI-Optimized PCs

Kryptronix specializes in crafting PCs tailored to AI and machine learning needs:

Custom Component Selection: Builds that prioritize AI workloads, featuring top-tier processors and GPUs.

Scalability: Systems designed to grow with evolving AI tools and frameworks.

Rigorous Testing: Ensures reliable performance under the stress of real-world applications.

Power Efficiency: Optimized builds for sustainable energy consumption.

4. Real-World Application: Accelerating Machine Learning Research

Case Example: An AI researcher leveraged a Kryptronix-built PC featuring:

AMD Ryzen Threadripper CPU: Accelerated training time by 40%.

NVIDIA RTX A6000 GPU: Enabled large-scale model training with precision.

128GB RAM and 2TB NVMe SSD: Managed complex datasets effortlessly.

This setup dramatically reduced project timelines and improved model accuracy, showcasing the transformative impact of tailored configurations.

5. Why Choose Kryptronix for AI PCs?

Kryptronix delivers unmatched value with:

Expert Consultation: Helping AI professionals select ideal configurations.

Future-Proof Systems: Easily upgradeable components to match advancing AI technologies.

Lifetime Technical Support: Ensuring uninterrupted workflows for mission-critical tasks.

Conclusion

AI professionals need high-performance, reliable systems to meet the demands of cutting-edge innovation. Kryptronix’s expertise in building AI-driven PCs ensures creators and researchers can unlock the full potential of their work.

Discover AI-optimized PCs by visiting Kryptronix.