{"id":92841,"date":"2025-08-24T05:07:10","date_gmt":"2025-08-24T05:07:10","guid":{"rendered":"https:\/\/www.newsbeep.com\/ca\/92841\/"},"modified":"2025-08-24T05:07:10","modified_gmt":"2025-08-24T05:07:10","slug":"kioxias-new-5tb-64-gb-s-flash-module-puts-nand-toward-the-memory-bus-for-ai-gpus-hbf-prototype-adopts-familiar-ssd-form-factor","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/ca\/92841\/","title":{"rendered":"Kioxia\u2019s new 5TB, 64 GB\/s flash module puts NAND toward the memory bus for AI GPUs \u2014 HBF prototype adopts familiar SSD form factor"},"content":{"rendered":"

Kioxia<\/a> has developed a prototype of the 5TB high-bandwidth flash memory module with a bandwidth of 64 GB\/s. It’s essentially NAND-based memory for GPUs. Compared to HBM, High Bandwidth Flash<\/a> (HBF) adapts the concept to NAND flash, offering 8-16x the capacity of DRAM-based HBM. By combining speed with persistent storage, HBF enables large AI datasets to be accessed efficiently while using less power. One of these HBF modules, which Kioxia has pushed to 64 GB\/s, is what allows this capability.<\/p>\n

When you hear \u201cflash storage,\u201d you usually think in terms of capacity first, speed second. Even the fastest PCIe 5.0 SSDs <\/a>today\u201414 GB\/s class drives like Samsung\u2019s 9100 Pro<\/a>\u2014are dwarfed by the bandwidth demands of modern GPUs and CPUs. Kioxia\u2019s new prototype turns that expectation on its head: a single flash module delivering 5 TB of capacity and 64 GB\/s of sustained bandwidth over PCIe 6.0<\/a>. To put that into perspective, that\u2019s over 4x faster than the fastest PCIe 5.0 drives shipping today, and within striking distance of HBM2E\u2019s per-stack throughput<\/a>.<\/p>\n

The key is how the system scales; instead of one central controller trying to manage an entire bank of NAND\u2014which quickly becomes a bottleneck as more dies and channels are added\u2014Kioxia gives each module its own controller. That controller sits right next to its NAND and links to others in a daisy-chain layout. This reduces crosstalk and eliminates the complexity of wide parallel buses, which become increasingly challenging to manage as speeds increase. Instead, data is passed along in series, with each link pushing 128 Gbps using PAM4 signaling.<\/p>\n

You may like<\/p>\n

\"Kioxia's<\/p>\n

(Image credit: Kioxia)<\/p>\n

PAM4 (Pulse Amplitude Modulation with four levels) doubles the data rate per symbol compared to traditional NRZ signaling, but it\u2019s also more sensitive to noise and bit errors. To maintain signal integrity, Kioxia relies on equalization, error correction, and stronger pre-emphasis\u2014similar to what PCIe 6.0 itself requires.<\/p>\n

This helps explain the move to PCIe 6.0 as the host interface, as x16 lanes of PCIe 6.0 can theoretically handle around 128 GB\/s bidirectional. Kioxia\u2019s 64 GB\/s target sits just under half that limit, leaving enough headroom for error correction and overhead without maxing out the bus.<\/p>\n

\"Actual<\/p>\n

(Image credit: Kioxia)<\/p>\n

As you might be expecting, latency is the main tradeoff. HBM memory works in hundreds of nanoseconds, almost like an extension of GPU registers. NAND flash\u2014even with advanced controllers\u2014still accesses data in tens of microseconds, which is magnitudes slower. Kioxia counters this with aggressive prefetching and controller-level caching, so sequential workloads are less affected. It doesn\u2019t make NAND as fast as DRAM, but it narrows the gap enough that for streaming datasets, AI checkpoints, or large graph analytics, bandwidth matters more than raw latency.<\/p>\n

<\/p>\n

Power is another crucial factor here, as Kioxia claims under 40W per module, which appears impressive when compared to traditional Gen5 SSDs that can draw up to<\/a> 15W for ~14 GB\/s<\/a>. On a GB\/s per Watt basis, this module is dramatically more efficient. That matters because in a hyperscale rack, a few hundred drives can easily consume multiple kilowatts. AI datacenters\u2014already ballooning in power budgets thanks to H100 clusters<\/a>\u2014need every watt saved at the storage layer.<\/p>\n

Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.<\/p>\n

These modules also open up new system design options. With daisy-chained controllers, adding more modules doesn\u2019t consume additional bandwidth, so performance scales linearly with capacity. A complete set of 16 could reach 80 TB of flash and over 1 TB\/s of throughput\u2014numbers once limited to parallel file systems or DRAM scratchpads. This makes it possible to treat storage as near-memory, sitting directly on the PCIe fabric alongside accelerators, rather than being stuck in back-end I\/O.<\/p>\n

\"NAND<\/p>\n

(Image credit: Micron)<\/p>\n

This isn\u2019t Kioxia\u2019s first foray into bandwidth-heavy flash, either. The company has been experimenting with long-reach PCIe SSDs <\/a>and GPU peer-to-peer flash links, including research with Nvidia into XL-Flash drives<\/a> tuned for 10 million IOPS. Pair those efforts with its newly announced fab expansions in Japan<\/a> \u2014 driven by an expectation that flash demand will nearly triple by 2028\u2014and it\u2019s clear this prototype isn\u2019t a one-off. It\u2019s a roadmap<\/a> hint toward NAND not just being bigger, but to be faster, and fast enough to sit closer to the compute stack.<\/p>\n

For now, the module remains in prototype stage, and there are unanswered questions: how it handles mixed random workloads, how ECC scaling impacts latency, and what real-world throughput looks like under AI training conditions. However, the bigger message here is that flash is breaking out of its role as slow, deep storage and moving up the hierarchy. If Kioxia\u2019s vision (as outlined in their press release) pans out, the next generation of data centers<\/a> <\/a>might see storage modules competing for bandwidth bragging rights alongside GPUs themselves.<\/p>\n

Follow Tom’s Hardware on Google News<\/a> to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.<\/p>\n","protected":false},"excerpt":{"rendered":"Kioxia has developed a prototype of the 5TB high-bandwidth flash memory module with a bandwidth of 64 GB\/s.…\n","protected":false},"author":2,"featured_media":92842,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[62,276,277,49,48,61],"class_list":["post-92841","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-ca","tag-canada","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/posts\/92841","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/comments?post=92841"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/posts\/92841\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/media\/92842"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/media?parent=92841"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/categories?post=92841"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/tags?post=92841"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}