-
Posts
2474 -
Joined
-
Last visited
-
Days Won
57
jaxa last won the day on November 13
jaxa had the most liked content!
Reputation
1078 DeityAbout jaxa
- Birthday 04/01/1982
Contact Methods
-
Website URL
http://t4lg.anodal.org/
Recent Profile Visitors
79208 profile views
-
What if the answer to that question was a spoiler?!
-
jaxa started following The limits of the setting
-
There was a T2 FM set on Mars very loosely connected to the setting. There's nothing stopping anybody from making whatever they want, I guess there's a question of what could possibly be rejected from the in-game downloader?
-
https://wccftech.com/amd-teases-ces-2027-with-possibility-of-unveiling-zen-6-based-olympic-ridge-processors/ Zen 6 consumer desktop parts may be revealed at CES 2027. It would be surprising if they weren't revealed by then, really. I hope we see some leaks nailing down the specifics, especially about the I/O chiplet.
-
I think the future will look very different. Custom, 3D, brain-like accelerators. Processing-in-memory. Maybe gigantic unused portions of 3D chips with some parts activated for mixture of experts. Crazy performance/efficiency gains still plausible. It's better now probably. AMD is really reaping the benefits of AI code lololol.
-
CUDA moat is dead, AMD stands to benefit even without putting in the work: https://wccftech.com/anthropic-jeopardizes-nvidias-cuda-moat-as-its-claude-mythos-model-took-only-3-days-to-figure-out-how-to-run-another-model-on-amds-mi355x-gpus/ All sorts of conspiracy theories flying around: https://wccftech.com/after-nvidia-pushes-out-jensen-huang-alice-huangs-spy-credentials-suddenly-start-to-crack-leading-to-pronouncements-of-a-coup/
-
You can't fire me, I quit! End of an era, amid seismic change for the industry. Months/years before this, Nvidia's was starting to get a stinky reputation among gamers, given the major focus on AI, sky high prices, product discontinuations, driver issues (especially Blackwell), and other problems. Now it's possible that it will be revealed that a decent chunk of gaming GPU supply was diverted to China, even illegally. Direct from Nvidia in the case of Megaspeed, or smuggling of cards like the RTX 5090 bought from retailers. An AI market crash may be imminent, followed by a gradual decline in memory prices. AMD has a chance to step up and meet this moment. Even the "CUDA moat" is crumbling, ironically because of AI. Will they fumble? Maybe. They are good at that.
-
He's going to get fired. He has built up such a mythology around himself, which is crumbling. We're probably going to find that someone else can do just as well. Better even, considering the potential criminal liability. He was already getting a little nutty during his stage talks.
-
The literal direct replacement for the 8-core 9800X3D will be a 12-core X3D CPU with a single CCD. The 50% increase in L3 cache also applies to the cache chiplet, so it will end up with 144 MiB L3 cache instead of 96 MiB of the previous generations. You're asking for a 9900X3D2 equivalent, i.e. 16-20 cores (8, 9, or 10 cores enabled on each chiplet), with 2 CCDs, both with 3D cache. This is questionable because the 9900X and 9900X3D have been seen as unpopular SKUs, and there is no 9900X3D2, only the fully enabled 9950X3D2. Maybe it can exist if AMD is assembling the flagship but still has to disable some cores. Which is why we have 5600X3D/7600X3D. I think it's possible this level of SKU becomes more popular simply because it will have a minimum of 8 cores per CCD, up from 6 cores. Despite being substantially disabled if it goes from 12 cores to 8 per CCD, it won't be inferior to the 9950X in any way.
-
Forget old folks home tech, time for a deep dive into something you can't even buy yet: AMD Zen 6 EPYC ‘Venice’ deep dive and benchmarking – 9686F outclasses Nvidia’s Vera, dual EPYC 9996 with 1,024 threads delivers up to 93% performance gain gen-on-gen A lot of previously known information is confirmed. AMD is claiming 20% IPC uplift between Zen 5 and Zen 6c in "enterprise & cloud server workloads", 31% in "AI workloads" (probably taking advantage of new instructions). They may be playing a numbers game since the comparison is between two 8-core Zen 5 CCXs vs. 16 cores on a single Zen 6c CCX. They also doubled the L3 cache per core for the 32-core Zen 6c chiplet, so that is probably a contributing factor. But what's unmistakeable is that for the first time in a few generations, we will see an increase in client CPU L3 cache, from 32 MiB to 48 MiB per chiplet. Zen 2 debuted with 32 MiB split into 2x16 MiB, Zen 3 unified it into 32 MiB for 8 cores, Zen 4/5 made no changes. Zen 6 desktop CPUs with use 12-core CCXs with 48 MiB L3 cache.
-
Copyright has become a blight on the Internet and society. Time to end it via technical means.
-
He makes ridiculously lopsided builds to test old CPUs and GPUs. Titan V is from the ultimate DDR2 video and a separate video examining the card. https://www.youtube.com/@BudgetBuildsOfficial/videos Keep an eye on the channel, it's right up your alley. Very long videos but you can skip around or watch/listen at 2x.
-
Biowhatever was a dead man walking for years.
-
I've written off using >2 sticks. It's unnecessary for me, and you can get a lot in 2 sticks. I also lean towards SFF for future purchases which usually can only fit 2. My 64 GB DDR4 systems have 2x32GB. That is the way. DDR5 users can technically have 2x64GB or 2x128GB quad-rank now, good luck affording it though. https://www.tomshardware.com/pc-components/ram/256gb-of-dual-channel-ram-hits-mass-market-thanks-to-origin-code-quad-rank-cudimm-packs-128gb-of-ddr5-8000-into-a-single-module