Ruen
LightNews LightNews
ruen.bsky.social
Ruen
@ruen.bsky.social
170 followers 81 following 1.5K posts
I am surprised this username is not taken
Posts Replies Media Videos
ruen.bsky.social
AMD's marketing names are bad and kinda shady (selling old APUs as new ones) so I don't really blame you for getting confused.
Table of the AMD Ryzen Z2 Series Processors. From the best to worst AMD Ryzen AI Z2 Extreme: - 16 GPU cores (RDNA 3.5) - 8 CPU cores / 16 threads (mix of Zen5 + Zen5c) - LPDDR5X-8000 - 15W to 35W TDP - 50 TOPS NPU AMD Ryzen Z2 Extreme: Same as above but without an NPU AMD Ryzen Z2: - 12 GPU cores (RDNA 3) - 8 CPU cores / 16 threads (Zen4) - LPDDR5X-7500 - 15W to 30W TDP Essentially a re-release of the Z1E AMD Ryzen Z2 Go: - 12 GPU cores (RDNA 2) - 4 CPU cores / 8 threads (Zen3+) - LPDDR5-6400 - 15W to 30W TDP AMD Ryzen Z2 A: - 8 GPU cores (RDNA 2) - 4 CPU cores / 8 threads (Zen2) - LPDDR5-6400 - 6W to 20W TDP
October 16, 2025 at 9:00 PM
ruen.bsky.social
The base Xbox Ally has a Z2A: four Zen2 CPU cores, 8CU RDNA2.

Codename is also Aerith Plus. The Steam Deck's APU is also codenamed Aerith.
Screenshot of the Xbox website for the ROG Xbox Ally handhelds scrolled down to the processor information. The Ally X has a AMD Ryzen AI Z2 Extreme Processor while the ROG Xbox Ally has a AMD Ryzen Z2 A processor.
October 16, 2025 at 8:43 PM
ruen.bsky.social
Leading up to the Pixel 10 release we got full specs leaked including the GPU and one interesting thing was that a driver for these specific GPUs has hardware based virtualization which made complete sense given the terminal.

But Google doesn't use it???
From the page for the "IMAGINATION DXTP-48-1536 GPU": Hardware-based Virtualisation Imagination's hardware based virtualisation solution supports up to eight separate operating systems on a single GPU, each fully isolated in memory, for secure multitasking - including mixed criticality workloads.
October 13, 2025 at 3:29 AM
ruen.bsky.social
Screenshot of a tweet from That Wine Guy 🍷 (@thatwineguy98): "Nothing better than a glass of #RedWine... 🍷Except maybe #Men 😄 #Yep #ImGay 🌈" Attached to the post is a selfie of the persona behind the account and a picture of a glass of wine spilling out Photoshopped to have the colors of the rainbow.
October 7, 2025 at 11:06 PM
ruen.bsky.social
I guess this doesn't work. Or maybe it only shows up on the PDS but the AppView doesn't know how to handle it
Screenshot of the OpenVibe draft composer. When pressing the warning icon you can choose a content warning category of Graphic, Nudity, Porn and Custom
October 1, 2025 at 12:39 AM
ruen.bsky.social
Just a kind of boring upgrade. At least there were some efficiency improvements for both sets of cores unlike the D9500.

What Xiaomi did for their FIRST CPU is still super impressive, crazy and competitive. Taking a Cortex A725 and improving its efficiency that much is insane.
Screenshot of the video linked above. It shows two SPEC 2017 performance vs power curves, one for INT performance and the other for FP performance across various cores. The Xiaomi XRING 01's A725L core has a gap lead between it and its Android competition with a curve that is more to the left and top (meaning more efficient). As an example, the XRING's A725L core matches the 8 Elite Gen 5 M core's top INT performance while using half the power and matches it's top FP performance while using 25% less power. Its max performance is higher than the competition as well and uses less power than the Snapdragon. Only the Apple A19 and A19 Pro's E cores, which are plotted as points instead of a curve, are better in efficiency with them being more to the left (less power usage) than the A725L but at about 2/3 of the A725L's max point on the curve.
September 25, 2025 at 6:38 PM
ruen.bsky.social
Also forgot to point out that the peak power draw increased compared to the predecessor.

The previous test was testing each core type individually. But it does seem to be better and more efficient for multicore.

Apparently the core to core latency got reduced significantly and they increased the
Geekerwan test of the Dimensity 9500 with Geekbench 6's multicore. It scores around 11000 using ~18W. Around 9800 at 12.5W and around 8000 at 7.5W. It has a power efficiency curve similar to the 8 Elite. But the 8 Elite is slightly more efficient under 13W where they meet. The D9500 is slightly more efficient at the upper power limit of ~18W
September 23, 2025 at 2:45 AM
ruen.bsky.social
The new ARM cores tiering goes like this, from big to small.

C1-Ultra
C1-Premium
C1-Pro
C1-Nano

The D9500 doesn't have any nano cores so the smallest cores it has are C1-Pro.

And it's just what. The C1-Ultra is more efficient at every power level, compared to even the C1-Pro.
Screenshot of Geekerwan's video analysis on the Dimensity 9500 regarding CPU core power efficiency. Each CPU core was tested with SPEC 2017. The C1 Ultra core is very power efficient at every range. The C1 Premium gets slightly below 8 points in INT using 5.5W. While the C1 Premium gets 9 points in INT at the same power draw. The C1 pro geta around 4 points in INT while using like 1.2W. The C1 Ultra gets around 4.5 points at the same power level. In FP operations it's the same story. The C1 Premium gets 11 points using 6.5W while the C1 Ultra gets nearly 15 points at the same power level. The C1 Pro gets slightly above 6 points while using 1.75W while the C1 Ultra gets 7 points at the same power level.
September 23, 2025 at 2:26 AM
ruen.bsky.social
Geez ARM wasn't kidding.

The Mali G1 Ultra in the D9500 is indeed 2x faster than its predecessor, the D9400, in Solar Bay Extreme.

No power draw measurement or curve for this benchmark though.

It's also the first Android SoC to have BVH traversal capabilities.

youtu.be/tDvr1YOdlWg
Screenshot from Geekerwan showing the performance of the 3DMark Solar Bay Extreme scores of multiple chips and the devices they are in. The Dimensity 9500 (MTK reference device) scored 2570 The A19 Pro (iPhone 17 Pro Max) scored 2407 The A19 (iPhone 17) scored 2115 The A18 Pro (iPhone 16 Pro Max) scored 1539 The A18 (iPhone 16 Plus) scored 1353 The A17 Pro (iPhone 15 Pro Max) scored 1253 The Snapdragon 8 Elite (Red Magic 10 Pro Plus) scored 1192 And finally the Dimensity 9400+ (Vivo X200s) scored 825
September 23, 2025 at 1:22 AM
ruen.bsky.social
Yeah I don't think they'd contribute SVE for Windows code to LLVM for no reason either

www.qualcomm.com/developer/bl...
Screenshot of a section of the blog post linked: AArch64 The focus of our contributions to LLVM for AArch64 this release cycle has been enabling SVE (the Scalable Vector Extension) support for Windows. This work has been made up of several pieces. - One piece of this work has been enabling SVE intrinsics. These allow users to write vector code in C, rather than having to use inline assembly, which should be easier to maintain and may allow the compiler to perform minor optimizations to the code. SVE Intrinsics are not yet supported by MSVC, Microsoft’s C++ Toolchain, but are supported by both GCC and Clang and are widely used for writing high-performance SVE programs. - Another piece of this work has been enabling the SVE auto-vectorizer which allows the compiler to turn existing scalar loops into loops that use SVE instructions. Underlying both pieces of work, we have been working with Microsoft to document and implement details of the MSVC ABI for these features. This has included ensuring that LLVM correctly implements the Windows calling convention; that Clang correctly mangles SVE type names when using C++; and that LLVM emits the correct C++ unwinding information for functions that use SVE. The final key piece is to generate the right debug information on Windows, so that users can understand their code.
September 19, 2025 at 4:53 AM
ruen.bsky.social
browser.geekbench.com/v6/cpu/13863...
Screenshot of the CPU information of a geekbench run performed on the assumed to be Xiaomi 17. It supports the following Instruction Sets: neon aes sha1 sha2 neon-fp16 neon-dotprod sve i8mm sme-i832 sme-f32f32
September 17, 2025 at 8:48 PM
ruen.bsky.social
I guess it's true. People suspect geekerwan or someone else ran it while cooling the phone with extreme measures (for a phone) as most of the other results are under 3900

wccftech.com/a19-pro-poss...
Geekbench result for an iPhone 18,2 (iPhone 17 Pro Max). It scores 4019 points single core and 11054 points multi core.
September 13, 2025 at 6:49 PM
ruen.bsky.social
This is a bit of a special situation since I'm pretty sure the pricing is this way because of trade-in offers for the Pixel 10 but I think it still supports my point
Screenshot of Swappa's Google Pixel 10 unlocked page The price is "Starting at $615"
September 11, 2025 at 3:25 PM
ruen.bsky.social
I really didn't like the Bad Guys' 2 ending

Spoilers of course
The bad guys 2's ending is awful, it's just so random. Let's start off with the scene prior. They are locked in a warehouse and the police find them. And then as they are being transported they somehow convince the commissioner that they're good and innocent? With JUST a sticky note they left on her after Mr Shark accidentally knocked her out???? *Afterwards* she sees the "MoonX" rocket starting launch. So then they are in space to turn off a gold magnet because it's gonna destroy the world, but I don't think they elaborated on this. I'm guessing it's referring to gold not mined out yet coming out of the ground??? I'm not sure. They still put the world in harm and probably destroyed the economy because they didn't stop the villain from turning off the magnet in time and lots of gold went into space or in the air (and gold is one of the densests metal). Not to mention the bandits knew how to operate the ISS? We as viewers never get any background on these gals outside of Snake's relationship with Suan/Doom, Kitty Kat's first crime. Which is a shame considering they managed to do multiple crimes successfully and without a trace, it would've been cool to see them in action in an individual mission. How did they get past security all those times for example. On the station Kitty Kat lashes out at Doom for wanting to stop the magnet to save Snake, despite there being no clear explanation for her to explode like that? It just felt random. We got a bit of her backstory: she started by stealing a bag of lollipops from her teacher, but she didn't really express much negative emotions there. And when Dianne fought them she didn't fight her back with that rage because she knew that she could bring down Dianne psychology, so I imagine she just explodes at the sight of her losing control but it was never elaborated to be on that level. When they infiltrated that one rich person's wedding and the plan went sideways she got upset but not to that level and it was against Mr Wolf, not her own crew. Anyway. They stop the magnet and then shot the shuttle into the accumulated gold... To try to return it to earth I guess??? Even though it's a terrible idea because: the gold would probably melt and kill a few people?? Their version of the ISS goes down with them inside and then The Bad Guy's death is faked because they were hired by a secret government agency. Boom, everyone declares The Bad Guys as heroes (even though there's so much gold still in space? And it probably didn't return to the rightful owners if it didn't burn away. Not to mention the loss of the space station???? Where was any of the crew?) and feels bad for not trusting them after changing their way. Which is such a terrible way to end the movie. You're telling me, the message of the movie is: if you aren't accepted by society, even after you try to fit in, you're only good to society if you're dead? I really liked how Moon Girl handled her supervillains, she heard them out and worked with the major to give them a second shot and reintroduce them to society. The first Bad Guys movie kinda played with that idea. Professor Marmalade kinda helped them out to try to re-introduce them into society. But looking back at it, I guess that movie's ending was flawed also because Marmalade only got in prison because they believed he was the Crimson Paw. Also Dianne, the literal governor before getting exposed, couldn't help them out? Couldn't guide them or better yet, fund organizations that work to get criminals back into society? Oh and the extra credits scene is just so out of place. Professor Marmalade gets out of jail because Dianne is exposed, so he gets into a car made out of gold and then ends up in space because of the magnet. But then the car turns into a space shuttle...? Why??? Somehow he had planned this ALL along but nothing lead to that except for his interest in the meteorite in the first movie??
September 10, 2025 at 4:53 AM
ruen.bsky.social
AYN also announced the Odin3 that's also powered by an 8 Elite, the biggest difference I think is that it has an OLED screen.

Pricing and configurations are similar, but overall a tiny bit more expensive.
Pricing chart for the AYN Odin3 The "base" model with 8GB+128GB will start at $299 for 48 hours once it launches. Then it will rise to $329 during the early bird time period. Finally it will cost $349 at retail. The "pro" model has 12GB+256GB and will start at $369, then $399, and finally $419. The "max" model has 16GB+512GB and will start at $419, then $449, and finally $479. The "ultra" model has 24GB+1TB and will start at $499, then $519, then $549
September 7, 2025 at 12:08 AM
ruen.bsky.social
The pricing for this is up now. The 8 Elite is confirmed as the reason it's called "elite"

16GB + 512GB being $459 at retail is a really good value.

Qualcomm please you need to let your Android chips run Windows or bare metal Linux there's so much potential locked away behind Android here.
Table of Ayeneo KONKR Pocket Fit configuration options and pricing. 8GB + 128GB will sell for $269 the first week, then $309 during the early bird price period and $330 at retail 12GB + 256GB will sell for $339 first, then $379 and finally $399 at retail 16GB + 512GB will sell for $389 first, then $429 and finally $459 at retail 24GB + 1TB will sell for $499 first, then $519 and finally $549 at retail The special purple graffiti color will come with 24GB + 1TB and sell for $539 first, then $559, and finally $599 at retail
September 7, 2025 at 12:01 AM
ruen.bsky.social
I/O is probably the super skinny one, which looks even longer than on Lunar Lake (that SoC even had a filler tile to fill in the gap by the too short I/O tile)
Intel slide showing the construction of their Lunar Lake SoC. Starting from the bottom there is the package everything sits on, memory lives next to the SoC itself. The SoC is composed of a base tile, foveros (Intel's interconnect technology) layer and at the top are the compute tile, platform controller tile and a filler tile. The Compute tile visually looks to take up roughly 60% of the area, followed by 30% by the Platform Controller tile and 10% by the filler tile.
September 4, 2025 at 12:36 AM
ruen.bsky.social
The GB10 (what the N1X is rumored to be based on) seems like it's mostly just a Mediatek SoC with a Nvidia GPU attached which is interesting. I guess the NPU seen in the Geekbench OpenCL run for the N1X is probably a Mediatek one (it'd be inefficient to cross to the GPU

wccftech.com/nvidia-gb10-...
Slide with a title of GB10: A Successful Collaboration Between NVIDIA and Mediatek Merging of two architectures - GB10 combines MTK's power efficient CPU & Memory Subsystem implementation with NVIDIA's high-performance GPU - Possible through implementation of industry defined standard interfaces and protocols - Extensive performance modeling of GPU memory traffic into MTKʼs memory subsystem NVIDIA IP Integration - C2C and other NVIDIA IP delivered for integration by MTK through a well-defined IP model Design and Verification flow alignment - Hierarchical functional verification approach with heavy reliance on BFMs as well as co-simulation for complex end-to-end features - Heavy reliance on emulation to validate MTK's + NVIDIA's HW, FW and SW working together The slide includes a diagram showing that every single component you find in a SoC is provided by Mediatek except the GPU
August 27, 2025 at 1:28 AM
ruen.bsky.social
Oh Deer Diner Coffee Thermos (AI73-UE) CONTAINMENT PROCEDURE: No unique procedures required. Altered status is suspected, but unverified. DESCRIPTION/ALTERED EFFECT: A bleu coffee thermos with a white plastic cap and a logo of the "Oh Deer Diner" located in Bright Falls, Washington. The cap doubles as a mug. The item responds to Formula I.16, yet no discernable altered effect has been discovered, other than keeping liquids warm for surprisingly long periods of time. It's also worth noting that the coffee from the thermos is always refreshing and strong, no matter its quality before being poured into the item. BACKGROUND: The thermos was discovered on the shore of Cauldron Lake, outside the town of Bright Falls. Both are known sites of a recurring AWE (see Bright Falls AWE case file for further information). According to witness statements, Alan Wake (see Prime Candidate Program entry for further information), who went missing shortly before the discovery of the thermos, had been seen [censored] collecting similar thermoses for an [censored] purpose.
August 24, 2025 at 1:20 AM
ruen.bsky.social
Since I figured out how to get the extra details out of a Geekbench result (you just add .gb6 at the end of the URL) I have found out that the N1X, Nvidia's upcoming ARM CPU, has a NPU class device. Whether that's just a wrapper for CUDA I have no idea.
{ "id": 31, "value": "American Megatrends International, LLC. 5.36_9ABKA0038", "ivalue": 0 }, { "id": 32, "value": "", "ivalue": 0 ), { "id": 33, "value": "", "ivalue": 0 ), { "id": 34, "value": "JMJWOA-Generic-NPU", "ivalue": 0 }, { "id": 35, "value": "JMJWOA-Generic-GPU", "ivalue": 0 }
August 20, 2025 at 2:44 AM
ruen.bsky.social
Yup. Glymur is the X2 Elite (or Ultra..?) codename
lore.kernel.org/all/52ytt5ag...
> The case of X1E80100 appears to be an exception-likely influenced by its unique naming convention at the time. > > That said, I'd prefer to stay aligned with the established convention used for earlier chipsets to preserve continuity. I'd appreciate hearing your thoughts on this as well. We're changing the naming model completely, so there is no continuity. In fact the Hamoa "exception" would suit us very well for Glymur. And look how nicely the CLK_X1E80100_* entries are grouped together in the Kconfig. Change to CLK_GLYMUR_* please. Regards, Bjorn
August 19, 2025 at 8:08 PM
ruen.bsky.social
I remembered Dell's XPS roadmap leaked from May 2024 and I went to check it out again and it's interesting.

The most recent leak on the X2 Elite is that it will have an 18 core CPU layout. And I guess this supports that with "Oryon V2 Ul" being a 40W TDP chip in the next XPS 14.
August 19, 2025 at 7:51 PM
ruen.bsky.social
Tne 8 Elite 2 in an S26 Edge apparently showed up on Geekbench and the score was 3393 single core and 11515 multi, with the CPU apparently actually running at 4GHz according to the run's json instead of the CPU's reported max of 4.74 GHz.

wccftech.com/snapdragon-8...
Geekbench result of the samsung SM-S947U. The single core score is 3393 and the multi-core score is 11515. The upload date is August 11, 2025 08:45 PM. It has 341 views. Operating System: Android 16 Model and model ID: samsung SM-S947U Motherboard: canoe Governor: wait CPU name: Qualcomm ARMv8 Topology: 1 Processor, 8 cores Identifier: ARM implementer 81 architecture 8 variant 3 part 2 version 1 Base frequency: 3.63GHz Cluster 1: 6 cores @ 3.63GHz Cluster 2: 2 cores @ 4.74GHz Instruction Sets: neon aes sha1 sha2 neon-fp16 neon-dotprod sve i8mm sme-i8i32 sme-f32f32 sme2
August 15, 2025 at 4:31 PM
ruen.bsky.social
I think this is the first time I've actually seen power metrics for an NPU. Though the graph is absolutely terrible. Where are all the performance numbers. What is the Y-axis actually measuring?

www.qualcomm.com/developer/wi...
Graph from Qualcomm titled "NPU Leadership": "Fastest and most efficient on battery. Intel Core Ultra Series 2 peak performance requires 311% more power. AMD Ryzen AI 300 series does not run Procyon AI on the NPU" The graph shows a relation of NPU performance (that's lacking any numbers) vs Idle-Normalized Platform Power (W). The Snapdragon X Elite X1E-80-100's initial performance starts at ~3W, matching the Intel Core Ultra 7 Series 2 256V consuming ~18W. The curve for it stops at ~10W with some better performance. The Snapdragon NPU uses less power at every point in this graph. The Intel CPU's NPU performance starts at ~10W and is illustrated to be lower than the Snapdragon one. The curve goes all the way to ~18W
July 11, 2025 at 3:26 PM
ruen.bsky.social
Your inclusion of Asriel makes me think that maybe the school in hometown follows something more like what the British do where they finish secondary school (what we call high school in the US) at the age of 16 and then go to college or an adult school afterwards?
Table comparing the naming scheme of American school system on the left to the English school system on the right. They are split down the middle with a column stating the age at how every level. At ages 11-12 students attend 6th grade "Middle school" in the American system while students attend Year 7 "Secondary School" in the English system. At ages 12-13 students attend 7th grade "Middle school" in the American system while students attend Year 8 "Secondary school" in the British system. At ages 13-14 students attend 8th grade "Middle school" in the American system while students attend Year 9 "Secondary school" in the British system. Now at ages 14-15 students attend 9th grade "High school" in the American system while students attend Year 10 "Secondary school" in the British system. At ages 15-16 students attend 10th grade "High school" in the American system while students attend Year 11 "Secondary school" in the British system. Now at ages 16-17 students attend 11th grade "High school" in the American system while students attend "Sixth form" or "Lower sixth" or Year 12 or college in the British system At ages 17-18 students attend 12th grade "High school" in the American system while students continue "Sixth form" into "Upper sixth" or Year 13 or continue college in the British system
July 8, 2025 at 9:37 PM