Flashbolt! Ah-ah! King of the impossible

Subject: General Tech | March 22, 2019 - 01:43 PM |
Tagged: flashbolt, Samsung, HBM2, HBM2E, aquabolt, flarebolt

Last year Samsung released Aquabolt, their HB2 modules which offered 2.4 Gbps per pin data transfer rates at 1.2V, rather impressive since the specification stated 2.0Gbps per pin was the top transfer rate.  Today they announced the next generation, Flashbolt which uses HBM2E and unlike certain other standards which recently bolted an E onto the end of their name this offers an actual improvement.

To make sure it you are confused this Friday, the new chips are rated for 410GB/s of bandwidth and 16GB of capacity; go yell at The Inquirer but it isn't a mistake.  To put it more helpfully, HBM2E can accommodate 16Gb per die, with transfer rates of 3.2Gbps per pin.  That is an immense amount of bandwidth for GPUs and HPC cards, though we won't be seeing any products in the near future nor are you likely to be able to insert it into your system any time soon.  AMD adopted a winning technology first, but as often seems to be the case, perhaps they moved a little too early.

View Full Size

"The name follows the same conventions as Samsung's previous memory products, Aquabolt and Flarebolt, but relies on a radically different architecture: HBM2E can accommodate 16Gb per die, twice as much as HBM2 used in Aquabolt, and offer transfer rates of 3.2Gbps per pin."

Here is some more Tech News from around the web:

Tech Talk

 

Source: The Register

Video News


March 23, 2019 | 07:02 PM - Posted by MoreThanApplelaptopsNeedDiscreteMobileVegaWith4GBofHBM2 (not verified)

I'd galdly pay more for an Discrete Mobile Vega AMD GPU that made use of one of these Samsung HBM2E stacks if AMD could begin offering Discrete Mobile Vega GPU SKUs for the non Apple laptop market. I do not want to have to wait for Navi as I'm more interested in getting a laptop with an AMD diecrete mobile GPU that can make use of Vega's HBCC/HBC IP to turn that HBM2/HBM2E into a last level GPU VRAM Cache.

If you are working with Blender 3D/Other 3d software then having Vega's ability to leverage the HBM2 as Cache(HBC) for some larger amount of Virtual VRAM paged to and from regular system DRAM then that's what's great about Vega's HBCC/HBC IP and Discrete Mobile Vega/HBM2 on laptops for non gaming 3d graphics workloads that are not really FPS oriented.

As far as I'm concerned AMD's too much focused on Gaming and Navi and really should be paying more attention to Discrete Mobile Non Gaming Graphics users who do not necessarily need to worry about any FPS oriented metrics. And really there needs to be some non Apple Discrete Mobile Vega/4GB-HBM2 offerings because Navi is currently not released! And even when Navi is released it will be a good long while before any Navi products are available in laptops.

Vega should have had discrete mobile GPU options within 3-6 months after the Desktop Vega SKUs where RTM.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

CAPTCHA
This question is for testing whether you are a human visitor and to prevent automated spam submissions.