Samsung HBM2E ‘Flashbolt’ Memory for GPUs: 16 GB Per Stack, 3.2 Gbps
by Anton Shilov on March 20, 2019 11:00 AM EST
Samsung has introduced the industry’s first memory that correspond to the HBM2E specification. The company’s new Flashbolt memory stacks increase performance by 33% and offer double per-die as well as double per-package capacity. Samsung introduced its HBM2E DRAMs at GTC, a fitting location since NVIDIA is one of the biggest HBM2 consumers due to their popular GV100 processor.
Samsung’s Flashbolt KGSDs (known good stacked die) are based on eight 16-Gb memory dies interconnected using TSVs (through silicon vias) in an 8-Hi stack configuration. Every Flashbolt package features a 1024-bit bus with a 3.2 Gbps data transfer speed per pin, thus offering up to 410 GB/s of bandwidth per KGSD.
Samsung positions its Flashbolt KGSDs for next-gen datacenter, HPC, AI/ML, and graphics applications. By using four Flashbolt stacks with a processor featuring a 4096-bit memory interface, developers can get 64 GB of memory with a 1.64 TB/s peak bandwidth, something that will be a great advantage for capacity and bandwidth-hungry chips. With two KGSDs they get 32 GB of DRAM with an 820 GB/s peak bandwidth.
| Samsung's HBM2 Memory Comparison | ||||||||
| Flashbolt | Aquabolt | Flarebolt | ||||||
| Total Capacity | 16 GB | 8 GB | 8 GB | 4 GB | 8 GB | 4 GB | ||
| Bandwidth Per Pin | 3.2 Gb/s | 2.4 Gb/s | 2 Gb/s | 2 Gb/s | 1.6 Gb/s | 1.6 Gb/s | ||
| Number of DRAM ICs per Stack | 8 | 8 | 8 | 4 | 8 | 4 | ||
| DRAM IC Process Technology | ? | 20 nm | ||||||
| Effective Bus Width | 1024-bit | |||||||
| Voltage | ? | 1.2 V | 1.35 V | 1.2 V | ||||
| Bandwidth per Stack | 410 GB/s | 307.2 GB/s | 256 GB/s | 204.8 GB/s | ||||
To increase DRAM transfer speed per pin to 3.2 Gbps, Samsung probably had to employ various methods to reduce collateral clock interference between the 5000+ TSVs and ensure clean signals, yet the company does not discuss this in its current announcement. Last year the company did disclose some of the tricks used by its Aquabolt HBM2 DRAMs to increase bandwidth per pin to 2.4 Gbps, so most of these methods have probably evolved in case of the Flashbolt.
In fact, Samsung’s announcement does not state that that the company has started mass production of its Flashbolt HBM2E memory, so it looks like the company has finished development of the technology, but is not yet ready to start shipments of such chips in mass quantities.
Related Reading:
- JEDEC Updates HBM Spec to Boost Capacity & Performance: 24 GB, 307 GB/s Per Stack
- Samsung Starts Production of HBM2 “Aquabolt” Memory: 8 GB, 2.4 Gbps
- JEDEC Publishes HBM2 Specification as Samsung Begins Mass Production of Chips
- SK Hynix Adds HBM2 to Catalog: 4 GB Stacks Set to Be Available in Q3
Source: Samsung
Facebook
Twitter
RSS
25 Comments
View All Comments
JoeyJoJo123 - Wednesday, March 20, 2019 - link
Pretty cool. But it'll be a while before memory like this shows up in consumer GPUs.Targon - Wednesday, March 20, 2019 - link
It will probably be ready for the high end Navi video cards that will come with HBM2 memory(November 2019-January 2020 I expect).FreckledTrout - Wednesday, March 20, 2019 - link
I really hope your wrong because then we are saying Navi is just a tweak of Vega. I absolutely hope they redesigned Navi to use GDDR6 and GCN is gone.SaturnusDK - Wednesday, March 20, 2019 - link
Not to worry. AMD has already stated Navi can use both GDDR and HBM2.darkswordsman17 - Friday, March 22, 2019 - link
@SaturnusDK: Vega supported GDDR and HBM, but I think we only saw it with GDDR5 in that one Chinese console. Everywhere else Vega was using HBM (well except in the small APUs where it used the CPU's DDR4). I expect Navi 10 to use GDDR6, and then Navi 20 will use HBM2. I won't be surprised if AMD goes with just two stacks for the consumer versions of Navi 20 (750-800GB/s I think would be enough for gaming GPUs - as Vega 20 doesn't seem to benefit that much from having more than double the bandwidth of Vega 10 in most gaming scenarios, and it'd still be ~50% more bandwidth than I'd expect Navi 10 using GDDR6 will have).StevoLincolnite - Wednesday, March 20, 2019 - link
Navi is likely a Polaris replacement, not a Vega replacement.I.E. Designed for the Mainstream and not the high-end.
Santoval - Thursday, March 21, 2019 - link
Quite a few Navi versions are rumored, so it might be a replacement for both Polaris and Vega.darkswordsman17 - Friday, March 22, 2019 - link
Its hard to know what is what since AMD overhauled their roadmaps after Vega 10's release (and Raja no longer being there). I think the plan was somewhat similar as for Vega, with a Navi 10 initial release that targets consumers (and in this case things have pointed to Polaris' mainstream market segment), and then a Navi 20 that is really targeting the pro markets, but will bring enthusiast level cards based on it that will come in 2020. With the rumors of APUs that pair Zen 2 modules with Navi based GPU modules, I'd guess there's gonna be at least 3 main versions of the GPU (for the PC consumer markets, with sub versions - like Polaris 10 and 11, or the Vega versions where its 10, 11, think 12, 20).darkswordsman17 - Friday, March 22, 2019 - link
Oh and I think it is. Navi 10 is replacing Polaris, and Navi 20 is likely to replace Vega sometime next year. Some I think are expecting something between them, but I have a suspicion that its actually going to be a resurgence of multi-GPU (not sure if AMD will stick with Crossfire branding, as I think this is the beginning of their shift to multi-dice mGPU, where the idea is to get them operating so that to applications they seem to be just a single GPU), where AMD will rely on two (or maybe more) smaller and efficient GPUs to offer enthusiast level performance. I think Navi 20 will be targeting HPC/pro market more so will probably be more compute heavy (it'll still offer more typical GPU power than Navi 10 making it viable for enthusiast level single dGPU, but unless you can utilize the compute capability, that you might be better off with like 2-3 Navi 10 cards for gaming).Santoval - Thursday, March 21, 2019 - link
Navi is 100% GCN based, apparently the last GCN based GPU. The first post-GCN AMD GPU is not Navi, it is Arcturus, which is to be released in late 2020 or 2021.