#hbm — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #hbm, aggregated by home.social.
-
RT @kyzoroXX: Enterprise-AI-Server haben seit Jahren etwas vermisst: HBM über PCIe. AMD ändert das endlich mit dem Instinct MI350P. • 144 GB HBM3e • 4,0 TB/s Speicherdurchsatz • PCIe Gen5 x16 Dies sind nicht nur größere Zahlen. HBM-Speicher war weitgehend auf rackbasierte KI-Systeme beschränkt. Die Integration in einen Standard-PCIe-Beschleuniger könnte größere LLMs auf bestehenden Enterprise-Servern ohne Umstellung auf spezialisierte Infrastruktur praktikabel machen. Bei vielen Inferenz-Workloads werden Speicherkapazität und -durchsatz zum Engpass, lange bevor die reine Rechenleistung limitiert wird. Manchmal ist das wichtigste Upgrade nicht mehr TFLOPS, sondern die Beseitigung des Speicherengpasses. Video KyzoroX (@kyzoroXX) Artikel DGX Spark vs. AMD Strix Halo vs. Mac Mini: Der ehrliche 3-Wege-Benchmark. Ein 600-Dollar-Mac-Token generiert Token fast genauso schnell wie ein 4.700-Dollar-DGX-Spark. Diese eine Tatsache verändert die Art und Weise, wie jeder lokale KI-Hardware kauft – bis man die in Benchmarks verborgene Kennzahl kennt. Alle testen – https://nitter.net/kyzoroXX/status/2076791492778287387#m
mehr auf Arint.info
-
#Samsung and #Broadcom join forces for advanced #AI technologies: #HBM memory and 2nm chips
https://gadgetflux.eu/samsung-si-broadcom-extind-colaborarea-pentru-ai/
-
#Samsung and #Broadcom join forces for advanced #AI technologies: #HBM memory and 2nm chips
https://gadgetflux.eu/samsung-si-broadcom-extind-colaborarea-pentru-ai/
-
RT @wccftech: Chinas CXMT setzt auf maßgeschneiderten DRAM, um US-Sanktionen gegen HBM zu umgehen, während koreanische Lieferanten, die auf Nummer sicher gehen, eine große Chance verpassen könnten. wccftech.com/cxmt-custom-dra… Link Chinas CXMT setzt auf maßgeschneiderten DRAM, um US-Sanktionen gegen HBM zu umgehen; Koreanische Lieferanten... Durch US-Sanktionen von HBM-Tools ausgeschlossen, wendet sich Chinas CXMT dem kundenspezifischen 3D-DRAM zu und eröffnet ein hohes Risiko, das vorsichtige koreanische Lieferanten verpassen könnten wccftech.com
mehr auf Arint.info
-
https://www.wacoca.com/life/431734/ NH Investment & Securities、SK hynixから5,000億ウォン調達へ 証券業界で「半導体マネー」獲得競争が激化 #ai #HBM #investment #KBSecurities #MiraeAssetSecurities #NHInvestment&Securities #SKHynix #toshi #キム・ドンウォン #投資 #流動性管理規制 #直接公募 #長期供給契約
-
https://www.wacoca.com/life/431734/ NH Investment & Securities、SK hynixから5,000億ウォン調達へ 証券業界で「半導体マネー」獲得競争が激化 #ai #HBM #investment #KBSecurities #MiraeAssetSecurities #NHInvestment&Securities #SKHynix #toshi #キム・ドンウォン #投資 #流動性管理規制 #直接公募 #長期供給契約
-
Memory earnings season opened with numbers that would have been typos three years ago: $61 billion in quarterly operating profit at Samsung, $41.5 billion in revenue at Micron against $9.3 billion a year earlier. Q3 contract talks point to DRAM rising another 13–30%, NAND 10–40%.
The more interesting shift is the forecast horizon. On the day of its $26.5 billion Nasdaq debut — the largest ever US listing by a foreign company — SK hynix called 2027 the worst supply year in the industry's history and said demand will outrun capacity beyond 2030. Customers reportedly asked for five to six times as much supply; some offered to buy the fab equipment outright. Meanwhile, wafer allocation keeps tilting toward HBM, leaving standard DDR4/DDR5 undersupplied and pushing buyers into the secondary market.
The full Q3 market update, including the case for skepticism about that 2030 forecast:
https://www.buysellram.com/blog/market-update-dram-contract-prices-are-set-to-jump/
#DRAM #MemoryPrices #MemoryShortage #DDR5 #DDR4 #Semiconductors #AIInfrastructure #DataCenter #ServerHardware #HBM #EnterpriseIT #ITAD
-
NVIDIA KoreaとSK hynix、800名が集結し「AI半導体」の研究成果を共有 — BigGo ファイナンス https://www.yayafa.com/2841631/ ##HBM #AgenticAi #AI #AIファクトリー #ArtificialGeneralIntelligence #ArtificialIntelligence #LGElectronics #NVIDIA #NVIDIAKorea #SamsungElectronics #SKHynix #エージェント型AI #チョン・ソヨン #人工知能 #先端パッケージング #汎用人工知能 #韓国ナノ総合技術院 #韓国科学技術情報通信部
-
NVIDIA KoreaとSK hynix、800名が集結し「AI半導体」の研究成果を共有 — BigGo ファイナンス https://www.yayafa.com/2841631/ ##HBM #AgenticAi #AI #AIファクトリー #ArtificialGeneralIntelligence #ArtificialIntelligence #LGElectronics #NVIDIA #NVIDIAKorea #SamsungElectronics #SKHynix #エージェント型AI #チョン・ソヨン #人工知能 #先端パッケージング #汎用人工知能 #韓国ナノ総合技術院 #韓国科学技術情報通信部
-
I Token AI Sono Appena Diventati Ancora Più Economici. Risolveranno la Carenza di Chip per l'AI… o la Alimenteranno?
I token dell'AI diventano sempre più economici, ma la memoria diventa sempre più costosa. I costi dell'inferenza sono diminuiti di 280 volte in soli due anni; nonostante ciò, la spesa delle imprese per l'AI è triplicata e i prezzi contrattuali della DRAM hanno appena registrato un aumento record del 90-95% in un solo trimestre. William Stanley Jevons aveva già osservato questo fenomeno nel 1865 con il carbone. Uno sguardo a ciò che riequilibrerà davvero il mercato dei chip e al motivo per cui la risposta sono le nuove fabbriche di semiconduttori previste per la fine del 2027, non modelli più intelligenti o ASIC.
buysellram.com/blog/cheaper-to…
#AI #Semiconduttori #DRAM #CarenzaDiMemoria #DataCenter #HBM #GPU #SSD #ITAD #BuySellRam #ParadossoDiJevons #Hardware #Tecnologia
-
I Token AI Sono Appena Diventati Ancora Più Economici. Risolveranno la Carenza di Chip per l'AI… o la Alimenteranno?
I token dell'AI diventano sempre più economici, ma la memoria diventa sempre più costosa. I costi dell'inferenza sono diminuiti di 280 volte in soli due anni; nonostante ciò, la spesa delle imprese per l'AI è triplicata e i prezzi contrattuali della DRAM hanno appena registrato un aumento record del 90-95% in un solo trimestre. William Stanley Jevons aveva già osservato questo fenomeno nel 1865 con il carbone. Uno sguardo a ciò che riequilibrerà davvero il mercato dei chip e al motivo per cui la risposta sono le nuove fabbriche di semiconduttori previste per la fine del 2027, non modelli più intelligenti o ASIC.
buysellram.com/blog/cheaper-to…
#AI #Semiconduttori #DRAM #CarenzaDiMemoria #DataCenter #HBM #GPU #SSD #ITAD #BuySellRam #ParadossoDiJevons #Hardware #Tecnologia
-
#SKHynix, #SouthKorea’s second-most valuable company, rose 13% in its Nasdaq debut, raising $26.5 billion for expansion plans. The company, a leader in high-performance #memory for #AIchips, is confident in sustained demand due to the AI boom. While some #HBM will be packaged in the US, most expansion will occur in South Korea. https://www.cnbc.com/2026/07/10/sk-hynix-skhy-stock-nasdaq.html?eicker.news #tech #media #news
-
#SKHynix, #SouthKorea’s second-most valuable company, rose 13% in its Nasdaq debut, raising $26.5 billion for expansion plans. The company, a leader in high-performance #memory for #AIchips, is confident in sustained demand due to the AI boom. While some #HBM will be packaged in the US, most expansion will occur in South Korea. https://www.cnbc.com/2026/07/10/sk-hynix-skhy-stock-nasdaq.html?eicker.news #tech #media #news
-
The cheaper AI tokens get, the more expensive memory becomes. Inference costs fell 280-fold in two years; enterprise AI spending tripled anyway, and DRAM contract prices just posted a record 90-95% quarterly jump. Jevons saw this pattern in 1865 with coal. A look at what actually rebalances the chip market...
https://www.buysellram.com/blog/cheaper-tokens-ai-chip-shortage/
#AI #Semiconductors #DRAM #MemoryShortage #DataCenters #HBM #GPU #SSD #ITAD #BuySellRam #JevonsParadox #TechHardware -
The cheaper AI tokens get, the more expensive memory becomes. Inference costs fell 280-fold in two years; enterprise AI spending tripled anyway, and DRAM contract prices just posted a record 90-95% quarterly jump. Jevons saw this pattern in 1865 with coal. A look at what actually rebalances the chip market...
https://www.buysellram.com/blog/cheaper-tokens-ai-chip-shortage/
#AI #Semiconductors #DRAM #MemoryShortage #DataCenters #HBM #GPU #SSD #ITAD #BuySellRam #JevonsParadox #TechHardware -
The cheaper AI tokens get, the more expensive memory becomes. Inference costs fell 280-fold in two years; enterprise AI spending tripled anyway, and DRAM contract prices just posted a record 90-95% quarterly jump. Jevons saw this pattern in 1865 with coal. A look at what actually rebalances the chip market...
https://www.buysellram.com/blog/cheaper-tokens-ai-chip-shortage/
#AI #Semiconductors #DRAM #MemoryShortage #DataCenters #HBM #GPU #SSD #ITAD #BuySellRam #JevonsParadox #TechHardware -
The cheaper AI tokens get, the more expensive memory becomes. Inference costs fell 280-fold in two years; enterprise AI spending tripled anyway, and DRAM contract prices just posted a record 90-95% quarterly jump. Jevons saw this pattern in 1865 with coal. A look at what actually rebalances the chip market...
https://www.buysellram.com/blog/cheaper-tokens-ai-chip-shortage/
#AI #Semiconductors #DRAM #MemoryShortage #DataCenters #HBM #GPU #SSD #ITAD #BuySellRam #JevonsParadox #TechHardware -
Claude Sonnet 5 launched at less than half the flagship's price with ninety percent of its benchmark performance. The natural conclusion: AI is getting cheaper, so chip demand should ease. Every data point says otherwise.
Inference costs fell 280-fold between 2022 and 2024 (Stanford AI Index). Enterprise AI spending tripled to $37B in 2025 anyway. DRAM contracts rose a record 90-95% in Q1 2026, NAND is projected up 70-75% in Q2, and SK Hynix has sold out its entire 2026 output. William Stanley Jevons described the mechanism in 1865: make a resource cheaper to use, and total consumption rises.
The article walks through what this means chip by chip — DRAM, NAND, GPUs, and why CPUs are the calm exception — plus the strongest counterarguments (ASICs, MoE, small models) and why they fall short. One practical takeaway: retired server memory is appreciating for the first time in years.
#AI #Semiconductors #DRAM #MemoryShortage #DataCenters #HBM #GPU #SSD #ITAD #ITAssetManagement #JevonsParadox #TechHardware #technology
-
Claude Sonnet 5 launched at less than half the flagship's price with ninety percent of its benchmark performance. The natural conclusion: AI is getting cheaper, so chip demand should ease. Every data point says otherwise.
Inference costs fell 280-fold between 2022 and 2024 (Stanford AI Index). Enterprise AI spending tripled to $37B in 2025 anyway. DRAM contracts rose a record 90-95% in Q1 2026, NAND is projected up 70-75% in Q2, and SK Hynix has sold out its entire 2026 output. William Stanley Jevons described the mechanism in 1865: make a resource cheaper to use, and total consumption rises.
The article walks through what this means chip by chip — DRAM, NAND, GPUs, and why CPUs are the calm exception — plus the strongest counterarguments (ASICs, MoE, small models) and why they fall short. One practical takeaway: retired server memory is appreciating for the first time in years.
#AI #Semiconductors #DRAM #MemoryShortage #DataCenters #HBM #GPU #SSD #ITAD #ITAssetManagement #JevonsParadox #TechHardware #technology
-
The cheaper AI tokens get, the more expensive memory becomes. Inference costs fell 280-fold in two years; enterprise AI spending tripled anyway, and DRAM contract prices just posted a record 90-95% quarterly jump. Jevons saw this pattern in 1865 with coal. A look at what actually rebalances the chip market...
https://www.buysellram.com/blog/cheaper-tokens-ai-chip-shortage/
#AI #Semiconductors #DRAM #MemoryShortage #DataCenters #HBM #GPU #SSD #ITAD #ITAssetManagement #JevonsParadox #TechHardware -
The cheaper AI tokens get, the more expensive memory becomes. Inference costs fell 280-fold in two years; enterprise AI spending tripled anyway, and DRAM contract prices just posted a record 90-95% quarterly jump. Jevons saw this pattern in 1865 with coal. A look at what actually rebalances the chip market...
https://www.buysellram.com/blog/cheaper-tokens-ai-chip-shortage/
#AI #Semiconductors #DRAM #MemoryShortage #DataCenters #HBM #GPU #SSD #ITAD #BuySellRam #JevonsParadox #TechHardware #tech -
Dentro la gerarchia della memoria delle GPU: come i server AI spostano i dati dagli SSD alla HBM
Ti sei mai chiesto come i modelli di intelligenza artificiale trasferiscono i dati alla GPU? Questo articolo spiega in modo semplice i diversi livelli di memoria all'interno di un server AI, perché la memoria della GPU è diventata un collo di bottiglia e quali nuove tecnologie potrebbero migliorare le prestazioni dell'intelligenza artificiale.
buysellram.com/blog/inside-the…
#HBM #HBM4 #GPUMemory #AIInfrastructure #DataCenter #CXL #HighBandwidthFlash #AIHardware #MemoryHierarchy #AIInference #NVMe #ITAD
-
Dentro la gerarchia della memoria delle GPU: come i server AI spostano i dati dagli SSD alla HBM
Ti sei mai chiesto come i modelli di intelligenza artificiale trasferiscono i dati alla GPU? Questo articolo spiega in modo semplice i diversi livelli di memoria all'interno di un server AI, perché la memoria della GPU è diventata un collo di bottiglia e quali nuove tecnologie potrebbero migliorare le prestazioni dell'intelligenza artificiale.
buysellram.com/blog/inside-the…
#HBM #HBM4 #GPUMemory #AIInfrastructure #DataCenter #CXL #HighBandwidthFlash #AIHardware #MemoryHierarchy #AIInference #NVMe #ITAD
-
A modern AI server runs five layers of memory, from nanosecond on-chip SRAM to petabyte-scale SSDs, and keeping the GPU fed is the whole engineering game. This piece walks the full hierarchy: why HBM became the bottleneck, why adding more isn't simple, and where HBF, CXL, and PIM fit into the next generation.
#HBM #HBM4 #GPUMemory #AIInfrastructure #DataCenter #CXL #HighBandwidthFlash #AIHardware #MemoryHierarchy #AIInference #NVMe #ITAD #buysellram
-
Why can't a GPU just carry more HBM? Interposers max out in size, stacking 12 to 16 DRAM dies compounds yield losses, and every stack sits beside a kilowatt-class package that hates sharing heat. So capacity climbs in careful steps — 80 GB on the H100, 141 on the H200, 192 on the B200, 288 on Blackwell Ultra — while KV caches for long-context inference balloon past 40 GB per request.
That gap between what models demand and what packaging permits is reshaping server design. NVIDIA's Rubin platform treats CPU memory and HBM as one coherent pool. SanDisk and SK hynix are standardizing High Bandwidth Flash as a capacity tier under HBM, with first samples due this half. CXL 4.0 pooling hardware is landing in racks now.
This article walks the whole memory hierarchy, from on-chip SRAM to NVMe, and explains what each emerging technology actually solves — and what it doesn't.
#HBM #HBM4 #GPUMemory #AIInfrastructure #DataCenter #CXL #HighBandwidthFlash #AIHardware #MemoryHierarchy #AIInference #NVMe #ITAD #technology
-
Why can't a GPU just carry more HBM? Interposers max out in size, stacking 12 to 16 DRAM dies compounds yield losses, and every stack sits beside a kilowatt-class package that hates sharing heat. So capacity climbs in careful steps — 80 GB on the H100, 141 on the H200, 192 on the B200, 288 on Blackwell Ultra — while KV caches for long-context inference balloon past 40 GB per request.
That gap between what models demand and what packaging permits is reshaping server design. NVIDIA's Rubin platform treats CPU memory and HBM as one coherent pool. SanDisk and SK hynix are standardizing High Bandwidth Flash as a capacity tier under HBM, with first samples due this half. CXL 4.0 pooling hardware is landing in racks now.
This article walks the whole memory hierarchy, from on-chip SRAM to NVMe, and explains what each emerging technology actually solves — and what it doesn't.
#HBM #HBM4 #GPUMemory #AIInfrastructure #DataCenter #CXL #HighBandwidthFlash #AIHardware #MemoryHierarchy #AIInference #NVMe #ITAD #technology
-
A modern AI server runs five layers of memory, from nanosecond on-chip SRAM to petabyte-scale SSDs, and keeping the GPU fed is the whole engineering game. This piece walks the full hierarchy: why HBM became the bottleneck, why adding more isn't simple, and where HBF, CXL, and PIM fit into the next generation.
#HBM #HBM4 #GPUMemory #AIInfrastructure #DataCenter #CXL #HighBandwidthFlash #AIHardware #MemoryHierarchy #AIInference #NVMe #tech
-
A modern AI server runs five layers of memory, from nanosecond on-chip SRAM to petabyte-scale SSDs, and keeping the GPU fed is the whole engineering game. This piece walks the full hierarchy: why HBM became the bottleneck, why adding more isn't simple, and where HBF, CXL, and PIM fit into the next generation.
#HBM #HBM4 #GPUMemory #AIInfrastructure #DataCenter #CXL #HighBandwidthFlash #AIHardware #MemoryHierarchy #AIInference #NVMe #tech
-
South Korea unveiled $550 billion in new memory fabs from Samsung & SK Hynix, part of a much larger $3.1 trillion corporate spending plan. Most of the new capacity targets server DRAM & HBM, not the commodity memory in ordinary PCs and servers. Industry forecasts point to real relief no earlier than late 2027, with some analysts already flagging oversupply risk...
https://www.buysellram.com/blog/will-samsung-and-sk-hynixs-550-billion-in-new-fabs-ease-the-memory-shortage/#DRAM #Semiconductors #MemoryShortage #SKHynix #Samsung #Micron #HBM #SupplyChain #DataCenter #buysellram
-
Samsung and SK Hynix just pledged $550 billion for new memory fabs — the concrete piece of a far larger $3.1 trillion long-term investment plan unveiled alongside South Korea's government on June 29. The number is easy to cheer and harder to interpret: most of the new capacity is aimed at server-grade DRAM and HBM, the highest-margin corners of the memory market, not the commodity memory in ordinary PCs and low-end servers.
Micron is running its own expansion on a similar timeline. China's CXMT is growing DRAM share faster than anyone but still years from qualifying for server and HBM business. Industry forecasters disagree on exactly when relief arrives — some point to late 2027, others flag oversupply risk once all this capacity actually lands.https://www.buysellram.com/blog/will-samsung-and-sk-hynixs-550-billion-in-new-fabs-ease-the-memory-shortage/
#DRAM #Semiconductors #MemoryShortage #SKHynix #Samsung #Micron #HBM #SupplyChain #DataCenter #CXMT #technology -
Samsung and SK Hynix just pledged $550 billion for new memory fabs. Most industry forecasts still don't expect real DRAM/NAND relief before late 2027 — some analysts warn of a glut right after. The actual math behind the timeline, broken down.
https://www.buysellram.com/blog/will-samsung-and-sk-hynixs-550-billion-in-new-fabs-ease-the-memory-shortage/#DRAM #Semiconductors #MemoryShortage #SKHynix #Samsung #Micron #HBM #SupplyChain #DataCenter #ITAD #CXMT #AIInfrastructure #technology
-
Samsung and SK Hynix just pledged $550 billion for new memory fabs. Most industry forecasts still don't expect real DRAM/NAND relief before late 2027 — some analysts warn of a glut right after. The actual math behind the timeline, broken down.
https://www.buysellram.com/blog/will-samsung-and-sk-hynixs-550-billion-in-new-fabs-ease-the-memory-shortage/#DRAM #Semiconductors #MemoryShortage #SKHynix #Samsung #Micron #HBM #SupplyChain #DataCenter #ITAD #CXMT #AIInfrastructure #technology
-
Kartell-Vorwürfe gegen DRAM-Hersteller: In einer Klage stehen Speicherpreise und die Frage im Fokus, ob HBM-Nachfrage nur vorgeschoben ist. Kritisch: Könnte die Drosselung von Consumer-Speicher gezielt höhere Preise ermöglichen? Was steckt wirklich dahinter? 📈🤔 https://www.golem.de/news/kartellbildung-klage-gegen-dram-hersteller-wegen-speicherpreisen-2606-210300.html #DRAM #HBM #Kartell #TechNews
-
Kartell-Vorwürfe gegen DRAM-Hersteller: In einer Klage stehen Speicherpreise und die Frage im Fokus, ob HBM-Nachfrage nur vorgeschoben ist. Kritisch: Könnte die Drosselung von Consumer-Speicher gezielt höhere Preise ermöglichen? Was steckt wirklich dahinter? 📈🤔 https://www.golem.de/news/kartellbildung-klage-gegen-dram-hersteller-wegen-speicherpreisen-2606-210300.html #DRAM #HBM #Kartell #TechNews
-
#Micron, a US #memorychip maker, has seen its stock soar due to the #AI driven supply crunch for memory chips, particularly High-Bandwidth Memory #HBM. The company’s recent earnings report showed a significant increase in revenue and profits, driven by strong demand from AI data centre builders. https://techcrunch.com/2026/06/28/why-wall-street-thinks-us-memory-maker-micron-is-the-next-nvidia/?eicker.news #tech #media #news
-
#Micron, a US #memorychip maker, has seen its stock soar due to the #AI driven supply crunch for memory chips, particularly High-Bandwidth Memory #HBM. The company’s recent earnings report showed a significant increase in revenue and profits, driven by strong demand from AI data centre builders. https://techcrunch.com/2026/06/28/why-wall-street-thinks-us-memory-maker-micron-is-the-next-nvidia/?eicker.news #tech #media #news
-
Micron posted record DRAM revenue but can fill only ~two-thirds of demand. Three reasons supply won't catch up: HBM eats the wafers, new fabs are years out, and output is pre-sold. Pulled server memory is now real supply.
https://www.buysellram.com/blog/micron-q3-dram-shortage-sell-server-memory/
#DRAM #ServerMemory #ITAD #DataCenter #MemoryShortage #DDR5 #HBM #Micron #SKHynix #ITAssetDisposition #EnterpriseIT #SemiconductorIndustry -
AI inference keeps getting pricier, and a lot of that cost is memory. High Bandwidth Flash (HBF) aims to change that: HBM-class speed, up to ~16x the capacity, far less per GB. What it is, how it could cut inference cost, and why it isn't shipping yet.
https://www.buysellram.com/blog/high-bandwidth-flash-ai-inference-cost/
#HBF #HighBandwidthFlash #AIinference #HBM #AImemory #GPU #DataCenter #Semiconductors #NAND #SKhynix #SanDisk #MemoryWall -
If you run AI models in production, you've probably noticed how much of the cost traces back to memory.
Every token a model generates is served from fast memory sitting right next to the GPU. That memory — High Bandwidth Memory, or HBM — is scarce, expensive, and effectively sold out for 2026. It's the same crunch that pushed DDR5 prices several times higher this past year. For long-context models and agents, memory is now the single largest cost driver in inference.
High Bandwidth Flash (HBF) is built to attack that. The idea: take the cheap, dense flash from SSDs and rebuild it to keep pace with a GPU — much of HBM's bandwidth, up to 8–16x the capacity, at a fraction of the cost per GB. In one vendor simulation, an HBM+HBF hybrid hit up to 2.69x higher throughput per watt than HBM alone.
The catch: HBF isn't shipping. Samples are due in late 2026, first systems in 2027 — the savings so far live in simulations, not invoices.
New explainer covers what HBF is, how it would cut cost, the alternatives (CXL, processing-in-memory, LPDDR modules), and how seriously to take the claim.
https://www.buysellram.com/blog/high-bandwidth-flash-ai-inference-cost/
#HBF #HighBandwidthFlash #AIinference #HBM #AImemory #GPU #DataCenter #Semiconductors #NAND #SKhynix #SanDisk #MemoryWall
-
Micron just turned in a record fiscal quarter, but the number that matters for anyone who runs servers wasn't the revenue.
DRAM average selling prices jumped into the low-60% range from the prior quarter while bit shipments barely moved. Price is doing the work because there aren't enough bits to ship. Micron says it can fill only about half to two-thirds of customer demand, with tight conditions past calendar 2027.
So why can't the makers just build their way out? Three supply-side forces:
HBM eats the wafers. It needs roughly 3x the silicon per bit of DDR5 (4x for HBM4), so as its share climbs, total bit output sags. 2026 DRAM bit growth is only about 16%.
New fabs are years away. Micron's Idaho plant won't ship until late 2027; its New York fabs land around 2029 to 2030.
What's made is pre-sold. Multi-year deals (Micron–Anthropic, plus Samsung and SK Hynix) lock volume off the open market.For IT teams retiring hardware, the DDR4 and DDR5 in decommissioned servers doesn't wait for a 2027 fab or sit behind a contract. It's available now, and one of the few sources of conventional DRAM many buyers can get.
https://www.buysellram.com/blog/micron-q3-dram-shortage-sell-server-memory/
#DRAM #ServerMemory #ITAD #DataCenter #MemoryShortage #DDR5 #HBM #Micron #SKHynix #ITAssetDisposition #EnterpriseIT #SemiconductorIndustry #technology -
Micron just turned in a record fiscal quarter, but the number that matters for anyone who runs servers wasn't the revenue.
DRAM average selling prices jumped into the low-60% range from the prior quarter while bit shipments barely moved. Price is doing the work because there aren't enough bits to ship. Micron says it can fill only about half to two-thirds of customer demand, with tight conditions past calendar 2027.
So why can't the makers just build their way out? Three supply-side forces:
HBM eats the wafers. It needs roughly 3x the silicon per bit of DDR5 (4x for HBM4), so as its share climbs, total bit output sags. 2026 DRAM bit growth is only about 16%.
New fabs are years away. Micron's Idaho plant won't ship until late 2027; its New York fabs land around 2029 to 2030.
What's made is pre-sold. Multi-year deals (Micron–Anthropic, plus Samsung and SK Hynix) lock volume off the open market.For IT teams retiring hardware, the DDR4 and DDR5 in decommissioned servers doesn't wait for a 2027 fab or sit behind a contract. It's available now, and one of the few sources of conventional DRAM many buyers can get.
https://www.buysellram.com/blog/micron-q3-dram-shortage-sell-server-memory/
#DRAM #ServerMemory #ITAD #DataCenter #MemoryShortage #DDR5 #HBM #Micron #SKHynix #ITAssetDisposition #EnterpriseIT #SemiconductorIndustry #technology -
Running AI in production? A lot of the bill is memory. Every token a model generates is served from fast memory next to the GPU — and that memory, HBM, is scarce and expensive.
HBF is built to attack that cost: much of HBM's bandwidth at a fraction of the price per GB. A new explainer covers what HBF is, how it could lower inference costs, and how close it actually is.https://www.buysellram.com/blog/high-bandwidth-flash-ai-inference-cost/
#HBF #HighBandwidthFlash #AIinference #HBM #AImemory #GPU #DataCenter #NAND #SanDisk #MemoryWall
-
Running AI in production? A lot of the bill is memory. Every token a model generates is served from fast memory next to the GPU — and that memory, HBM, is scarce and expensive.
HBF is built to attack that cost: much of HBM's bandwidth at a fraction of the price per GB. A new explainer covers what HBF is, how it could lower inference costs, and how close it actually is.
https://www.buysellram.com/blog/high-bandwidth-flash-ai-inference-cost/#HBF #HighBandwidthFlash #AIinference #HBM #AImemory #GPU #DataCenter #NAND #SanDisk #MemoryWall #tech
-
Micron posted a record quarter, but the real signal for IT teams isn't revenue. It can fill only about half to two-thirds of demand, with tight conditions past 2027. Three supply-side forces keep DRAM short: HBM consumes far more silicon per bit, new fabs are years out, and much of what's made is pre-sold to hyperscalers. The memory in retired servers is now real, available supply.
https://www.buysellram.com/blog/micron-q3-dram-shortage-sell-server-memory/
#DRAM #ServerMemory #ITAD #DataCenter #MemoryShortage #DDR5 #HBM #Micron #SKHynix #tech -
Micron posted a record quarter, but the real signal for IT teams isn't revenue. It can fill only about half to two-thirds of demand, with tight conditions past 2027. Three supply-side forces keep DRAM short: HBM consumes far more silicon per bit, new fabs are years out, and much of what's made is pre-sold to hyperscalers. The memory in retired servers is now real, available supply.
https://www.buysellram.com/blog/micron-q3-dram-shortage-sell-server-memory/
#DRAM #ServerMemory #ITAD #DataCenter #MemoryShortage #DDR5 #HBM #Micron #SKHynix #tech -
TSMC's roadmap shows AI chip progress is becoming as much about packaging as process nodes. CoPoS may scale package size, but CoWoS still looks key for dense HBM links.
-
RT @wccftech: Qualcomms HBC-Technologie stapelt die Rechenleistung direkt unter den DRAM-Chip, um die Speichermauer im KI-Bereich zu durchbrechen, und liefert pro Watt die sechsfache Bandbreite von HBM. wccftech.com/qualcomm-hbc-st… Link Qualcomms HBC stapelt Rechenleistung unter DRAM, um die KI-Speichermauer zu knacken und die sechsfache Bandbreite pro Watt zu erzielen... Qualcomm stellt seinen Durchbruch für den KI-Rechenzentrumsmarkt vor: HBC (High-Bandwidth Compute), entwickelt, um die Speichermauer zu durchbrechen. wccftech.com
mehr auf Arint.info
#DataCenter #HBC #HBM #KünstlicheIntelligenz #Qualcomm #TechNews #arint_info
-
Samsung и SK Hynix. Как Корея подмяла всю мировую память
Какой главный корейский продукт? Многие назовут автомобили, электронику и бытовую технику, сериалы, косметику или даже K-Pop. Но главный корейский продукт другой. Без него не работали бы смартфоны и ноутбуки, не было бы ИИ-бума, а весь мировой прогресс серьёзно бы замедлился. Эта корейская суперсила - память. В этой статье мы разберемся, как небольшая азиатская страна, отнюдь не сверхдержава, подмяла под себя главную драгоценность мировых технологий.
https://habr.com/ru/articles/1051552/
#память #чипы #полупроводники #hbm #dram #искусственный_интеллект #южная_корея #samsung #sk_hynix #nvidia
-
Micron Technology、Anthropicと提携 AIメモリ供給で協業、Series Hにも出資 https://www.yayafa.com/2828923/ ##HBM #AgenticAi #AI #Anthropic #ArtificialGeneralIntelligence #ArtificialIntelligence #claude #MicronTechnology #SeriesH #エージェント型AI #ストレージ #メモリ #人工知能 #汎用人工知能 #生成AI
-
Für 90+ Prozent Marge: SK Hynix baut mehr DDR5 statt HBM während Samsung lauert https://www.computerbase.de/news/arbeitsspeicher/fuer-90-plus-prozent-marge-sk-hynix-baut-mehr-ddr5-statt-hbm-waehrend-samsung-lauert.98049/ #semiconductor #Samsung #Skhynix #DRM #HBM
