Valient retweeted
Our concept for a fanless and fin-less mini PC. This 1kg aluminum pebble can handle a low-power CPU. What do you think?
43
10
212
9,455
But he didn’t forget the hardcore player that wants manual gameplay and something deep..
Flabbergasted at how good this game is with over a weeks play against the cpu AI and against friends on the couch..
1
Well I just can help myself when I find a good deal.
Finally retired the 2600X CPU after finding a great deal on a 3950X.
I've always wanted one of these CPUs. It was flagship in 2019. I was a still in my early days of mining but it was as close to a threadripper you could get at the time.
The parts bin cooperated. Retiring the faithful 2600X today, it served through a lot of builds and light duty without complaint. It has/was my #xmrvsbeast #gupax donor for a long time and if I find the parts it may go back to that!
This isn’t about needing 16 cores or 32 threads for daily gaming, I have those rigs built for those roles, it’s about finally completing the build I’ve had in my head, getting that historical ‘first mainstream 16-core’ chip, and having quiet headroom for whatever local experiments or homelab ideas come next. All while supporting one of my truest loves and passions...#monero #xmr
All local. All mine. Privacy intact. Feels good to close the loop on an old want.
This is how I support #monero #xmr. This is my passion part of it. I love to tinker and build. I love efficient private systems that are mine and mine alone.
@monero gives me that with every fucking hash!
To the GOAT.
Thank You.
5
FRAZE retweeted
Jun 10
Finally got Qwen3 0.6B working on bare metal - no linux (or any OS for that matter). It’s a quad-core CPU with 8 GB ram.
Currently operating at ~4 tok/sec on bf16. Hoping to write some mega ARM kernels to speed it up 💪🏻
3
1
23
682
【兴证策略】全球科技投资映射周跟踪（20260608-20260612）
海外映射对于A股投资机会的指引效果愈发显著，我们基于全球科技龙头，构建了覆盖六大科技板块、52大细分方向的国内与海外股价映射指数，定期进行跟踪比较。
过去一周，国内【光纤光缆、工业软件、企业软件、AI编程、AIGC】等方向表现相较于海外更好，而【半导体材料设备、CPU、芯片封测、存储、光模块、告诉铜缆等上游硬件】相较于海外公司仍有进一步扩散上涨的空间。
14
While we are on this subject, I'll never get deep into CPU testing myself. I know you cant please everyone, but when it comes to CPU testing you cant please anyone 🤣
>Why a 5090 at 1080p
>Why not test with RAM 8000 CL30?
>Why test at 8000 CL30 when no one can run that?
CPU testing is never just about the CPU, and no one agrees on what RAM speeds / timings to use. Do you max it out? Do you use JEDEC? Do you use only XMP?
The issue is also stability as not everyone is even able to run XMP fully stable. And while you might be able to do 8000 CL 30 in 95% of your games, there is always the possibility of 1 being unstable. Or it might he gake stable but the moment ypu open Premiere Pro it crashes 😭
The comments on this Reddit post again highlight that people dont understand CPU testing at lower resolutions.
Yes, nobody is going to run 1080p DLSS Performance on a 4090, but there are tests where the 5700X3D is below 50 fps. No matter how much load you shift to the GPU by increasing resolution, settings, etc, you wont get more than 50 fps in that scene.
Even if you pair it with a 5070 as I've seen some people recommend and the 5070 becomes the bottleneck at 1440p, you are still only going to get 50 fps in the scene tested here.
A GPU cant render more frames than what a CPU can prepare, except when using frame generation, and even then it isnt rendering more frames, but interpolating.
3
20
Cool, but I do LOD selection without a tree, with separate tiny chunks fully on the GPU so this approach is a bit orthogonal to mine.
I cant use CPU tree traversal quickly due to VRChat VM speed.
2
First of all this CPU is more than 1 years old right now, amd compatibility with ai solutions sucks really, really hard.
28
But then I had to do the same for every single cpu tested as they will all get different limits.
3
24
入力する情報が100%再現されるわけではない、というのはアナログも同じだと考えています。画材の性質を100%制御できるわけではありませんからね。
CPUによる補正は画材の性質のようなもの、といささか乱暴な主張をする人も見たことがあります。正誤は置いておいて、文脈としては近いかもしれません。
19
0!=1 retweeted
Jun 13
SPEC CPU 2026 Workload Analysis (INT Rate): jia.je/software/2026/05/22/s…
1
1
2
KJ retweeted
Great test! 9850X3D vs. 5700X3D benchmarks in high CPU-usage Ray-Traced games. 📊
The Zen 5 chip crushes its predecessor: ~30% more IPC × ~30% higher clocks = ~70% more single-thread performance.
reddit.com/r/overclocking/s/…
3
5
51
8,089
10m
La parte menos vistosa está en los fixes:
- Seguridad y confianza de proyecto: se corrige un caso en el que agentes en segundo plano podían leer ajustes de otro directorio en workers precalentados, incluyendo aprobaciones .mcp.json y trust.
- Permisos más predecibles: las reglas wildcard de WebFetch para subdominios vuelven a funcionar y se aceptan patrones intermedios en permisos de fichero como Read(secrets-*/config.json).
- Estado de agentes más limpio: se arreglan sesiones que seguían marcadas como Working, subagentes que quedaban activos tras parar un agente anidado y fallos EAUTH al adjuntar sesiones antiguas tras autoactualización del daemon.
- Mejor experiencia en terminal: menos renderizados innecesarios, menor CPU en reposo, carga batched de herramientas de Claude in Chrome y correcciones específicas en VSCode para PowerShell.
- Memoria y workflows: las sesiones remotas ya encuentran stores de memoria de equipo montados y la validación de workflows deja de rechazar scripts solo por mencionar Date.now() o Math.random() en comentarios o prompts.
No es una versión para enseñar una función vistosa. Es una versión para quitar puntos de rotura cuando Claude Code se usa muchas veces al día dentro de equipos reales.
1
26
10m
Claude Code 2.1.172 ya está disponible.
Claude no es una demo. Es una pieza de infraestructura para dejar que un agente toque trabajo real sin romper confianza. Esta versión va justo a eso: agentes más profundos, menos estados atascados y más consistencia en entornos con permisos, modelos y sesiones largas.
Lo más destacado:
- Los subagentes ya pueden lanzar sus propios subagentes, hasta 5 niveles de profundidad.
- Bedrock lee la región desde la configuración de AWS si no está definida en AWS_REGION.
- Se corrigen bloqueos en sesiones con contexto de 1M sin créditos de uso.
- Se endurecen restricciones de modelos disponibles en subagentes, dispatch y advisor.
- Mejoras de rendimiento en conversaciones largas y menor uso de CPU en reposo.
Te cuento más ↓
1
33