@LabRatKnatz
Account based inUnited States
About this account
- Account based in
- United States
- Connected via
- Web
Account-level information from X, not a live location or the device used for a specific post.
Personal Technologies and Entertainment
null, PNW, USA
Joined August 2023
- Tweets6.6K
- Following376
- Followers312
- Likes87.4K
Yes. Yes, you do.
GoG version + Knossos(dot)NET = Many hours of Space Combat Sim fun (with modern graphics [MediaVPS])
hard-light.net/forums/index.…
If only I could force Win98/ME to run entirely in my 5500X3D's L3 cache. Looks like it's already possible to boot it off 3DXpoint(Optane) NVMe-
Amazing.
Those still strangleheld to AM4, see AMD Radeon punching well above their weight vs. GeForce.
Given the CPU and PCIe bottleneck of the Cezanne 5500, Navi48 cards are meeting and beating GB203 and even GB202 cards. -especially, at 1% lows.
Great review, @hardwarecanucks
I tested the cheapest Ryzen CPU on EVERY modern GPU & NVIDIA cards have a problem...
youtu.be/h4Z3wmVsX6o
LabRat Knatz retweeted
Well here we go, I've been collecting hardware for YEARS. Decided to make my very first real dedicated video about my collection, and this is what happens! @msiUSA , @308Greenfield
youtu.be/BV_Zi2grI88
AMD really is salvaging every single piece of silicon off the wafer. Apparently, 5500F are Vermeer MCM CCDs with bad L3. Threw me for a loop today, too. 😵💫
Practically speaking, it's a 5500 with Gen4 PCIe and PBO+200 out of the box.
What in the world?! Ryzen 5 5500F is a Cezanne-derivative same as the Ryzen 5 5500.
The 5500F lists 3.0Ghz base, 4.4Ghz boost, 65W TDP
The 5500 lists 3.6Ghz base, 4.2Ghz boost, 65W TDP
They are different SKUs with different clock profiles but otherwise nearly-identical silicon.
Introducing the AMD Ryzen™ 5 5500F and AMD Ryzen™ 5 7500, built to deliver practical value for 1080p gaming. From affordable AM4 upgrades to entry-level AM5 builds, both processors offer a solid foundation for your next system.
Upgrade today : ow.ly/r43P50ZLUi3
Correction:
According to WCCFtech's research, 5500F is indeed a Vermeer MCM CPU with bad cache. @SaltMineRanch is apparently correct, I am incorrect.
wccftech.com/amd-silently-la…
LabRat Knatz retweeted
Yeah, my 111 tok/s dual-R9700 post may have underestimated these GPUs.
I found another builder with 2× Radeon AI PRO R9700s documented this progression running Qwen3.8-27B 👇
llama.cpp default:
🐌 19.2 tok/s
llama.cpp tuned for gfx1201 + tensor split:
⚡ 29.7 tok/s
vLLM Radiance + MXFP4:
🚀 184.9 tok/s
Same pair of GPUs.
And the latest Radiance benchmarks are now reporting ...
🧠 183.1 tok/s weighted single-stream
💻 188.8 tok/s code
🛠️ 218.8 tok/s file editing
📐 243.7 tok/s math
📄 249.4 tok/s JSON
And up to 523.5 aggregate tok/s at concurrency 8.
AMD might have a low-key story with the software, finally starting to use hardware to its fullest. 👀
Alex's Link in ALT
ALT https://alexkmiller.com/blog/2026/09/07/dual-r9700-llama-cpp-and-vllm.html?utm_source=chatgpt.com
🤯 This entire 64GB AMD AI workstation cost its builder LESS than a single RTX 5090 and Qwen3.8-27B is hitting 111 tok/s.
This is the kind of rig that makes me want to build it because RTX 5090s are so expensive.
Setup 👇
🟠 2× Radeon AI PRO R9700 32GB
💾 64GB total dedicated VRAM
🧠 Ryzen 7500F
💾 64GB DDR5
🔌 PCIe 5.0 x8 per GPU
⚡ vLLM Radiance + TP2 + MTP
Running Qwen3.8-27B Quark AWQ MXFP4 ...
🚀 111.4 tok/s median decode
⚡ 4,224–4,410 tok/s prefill
⏱️ 81ms TTFT
📚 131K server context
Native FP8 still managed 87.6 tps
🎮 RTX 5090 → 32GB VRAM
🟠 2× R9700 → 64GB VRAM
The builder says the entire machine cost ~€4,000, more than €1,000 less than a single RTX 5090 in his market.
So for less money, he ended up with 2× the VRAM + 111 tok/s on Qwen3.8-27B. 🔥
Reddit Link in ALT
ALT https://www.reddit.com/r/LocalLLaMA/comments/1w92x3j/2x_r9700_64_gb_ddr5_is_an_absolute_beast_machine
RX 9070 XT (Navi 48) is rapidly becoming the RX 480 (Ellesmere) of the DirectX 12 Ultimate Ray Tracing era.
TBQH, I wouldn't be surprised to see AMD 'Rebrandeon' or refresh it into the next generation ala-
HD 7970 Ghz>R9 280X
R9 285>R9 380
R9 290X>R9 390X
or,
RX 480->580->590.
That's certainly... a name. 🤣
(mSDex M.2 B/M-key Adapter)
already exists (oct 2025), just distributed within a small circle of people with a funny name
【[全球首测]当PCIE压缩进小体积:SD Express TF 评测报告-哔哩哔哩】 b23.tv/o6TYyPd
not that I think it would be cheaper than an ssd in regular form factor…
Not even close to 'worth it' but, this makes the point: the market demands more single-slot GPUs.
Personally, I'd really like to run an RTX card off my PROM21 Gen4x4 Slot for NVIDIA Broadcast.
Well before my Eras-of-choice (Pentium III->) to collect and play with but, this is a really cool ATI Inc. artifact from the Pre-PCI/Pre-Pentium Era.
Meet the ATI EGA Wonder 800, a premium EGA card from 1987.
It has 256 KB of RAM, a 9-pin TTL connector, an RCA output for a television, and an internal connector for portables like the IBM 5155.
The DE-9 could drive a 15 kHz RGB/CGA monitor, or a multisync display for proper EGA: 640×350 in 16 colors from a 64-color palette, at about 21.8 kHz. On a multisync it went further still, with 640×480×16 and even 800×560×16.
Its party trick was doing 640×350×16 on a plain RGB monitor by interlacing. The screen flickered badly, so you wouldn’t want to stay there long, but at least high-res EGA software still ran.
LabRat Knatz retweeted
Shame we don't get stuff like this anymore.
X399M Taichi - Small board. Big ambition.
⚡ To deliver the full-featured flagship experience expected from the Taichi series, most Taichi motherboards adopt larger ATX or E-ATX form factors, providing the space needed for robust power delivery, extensive expansion, and high-end hardware.
📅 Launched in February 2018, the X399M Taichi pushed those hardware limitations into a much smaller footprint. It was not only the first mATX motherboard in the Taichi series, but also the only mATX X399 motherboard available on the market at the time.
🔧 By fitting the massive AMD TR4 socket, quad-channel memory support, and a high-end X399 platform into an mATX design, ASRock delivered powerful specifications and performance within a much more compact layout. The X399M Taichi was more than just a smaller Taichi—it further demonstrated ASRock’s technical expertise in motherboard design, circuit layout planning, and space optimization.
🔗 Want to discover more stories behind a decade of Taichi design? Visit the ASRock Taichi 10th Anniversary website and relive the journey with us!
asrock.com/microsite/2026_Ta…
#ASRock #Taichi #Taichi10thAnniversary #ADecadeofMastery #FBS #FlashbackSaturday
