AMD + HBM to Turbo Charge Scrypt Mining

March 5, 2014 08:26 UTC
AMD will use HBM in their upcoming Pirate Islands GPUs

The gaming industry is doing cryptocoin mining a lot of favours. A joint development project by AMD and SK Hynix promises to boost the capacity of commercially available memory later this year. The chip manufacturer plans to use it’s new generation of memory module called High Bandwidth Memory (HBM) in it’s GPUs and indications are that target delivery may be the highly anticipated ‘Pirate Islands’ series of GPUs.

Thousand Island Park

The island themed code names are by now familiar to miners, who discovered the unparalleled hashing power of the Southern Island chip series when, before the introduction of ASICs, Bitcoin mining was ramped up in early 2012 by the hashing power of AMD’s popular HD 7000 cards.

October 2013 saw the release of the Volcanic Islands chip series as part of a strategic AMD campaign to dominate the Gaming GPU market. However, AMD underestimated the miner demand for these cards and the Volcanic Island Rx series was effectively sold out before it hit the retail shelves – leaving both gamers and AMD stockholders looking for answers. Nvidia found itself alone in the race for a few lucky months. Usurped from its Gaming destiny, the R9 290 is of course the undisputed GPU heavy weight champion of scrypt mining, but – grab the rails – because that could loop-the-loop later this year when AMD introduces HBM 3D to it’s GPU arsenal. HBM promises to give scrypt mining the heavy artillery that everyone perhaps expected would be delivered via hybrid ASICs.

Well, not everyone. The Execoin developers, for example had anticipated the ASIC Invasion and coins like Execoin not only stand to benefit from the HBM GPU race but from the coming ASIC threat in general.

“One more step, Mr. Hands” (from Treasure Island)

The target for HBM 3D technology is likely to be either of the Pirate Islands flagship GPUs, code named ‘Treasure Island’ and ‘Bermuda’, with a third chip, ‘Fiji’, positioned as an entry-level model.

Says Bryan Black, Senior Fellow and 3D Program Manager at AMD:

“Getting 3D going will take a BOLD move and AMD is ready to make that move.”

High Bandwidth Memory

The quest for faster memory had hit a ceiling with DDR5 RAM where neither design nor material changes can accomplish significant speed enhancement – and, mind you, it’s really fast. However, if we must go faster – and we do – enter 2.5 and 3D “Stacked” memory. As with CPU design and manufacture, this level of component design makes rocket science seem self-evident. We’re talking nanotechnology and electronic switching that occurs at voltages and speeds smaller than most of us can comprehend. The design concept of HBM can be comprehended, because it is a logical progression from the flat memory banks of DDR.

Think of conventional RAM as a flat, single storey warehouse with aisles and shelves. When the CPU requests stored data, the memory module determines the location (memory address) of that data on the warehouse map and retrieves it from the relevant aisle and shelf. To increase the size of memory, the warehouse’s floor space must be extended and, as a result, the process of retrieving data stored near the outer edges of the map takes longer. Until recently manufacturers ignored this ‘slow edges’ caveat, in favor of finding ways to speed-up memory by making the shelves and aisles smaller – thereby effectively decreasing floor space. As mentioned, the current technological limit of miniaturization was achieved in DDR5 RAM.

Hynix promises memory module improvements with HBM

Instead of attempting to decrease the nanosize of aisles and shelves, HBM revolutionizes the memory storage model by, instead, extending the warehouse in the vertical plane. Memory storage is stacked in layers and so our warehouse now has more storeys with multiple elevators interconnecting them. This 3 dimensional memory array makes data storage and retrieval more compact and efficient. Although the primary benefit is a boost to memory module speed, a knock-on effect is that the module consumes less energy than a conventional 2D memory map. Smart.

Commercial Availability

The concept of “die stacking” is not new and is currently used in some FPGAs and image sensors. Cost has, so far, prevented manufacturers from including stacked memory in commercial products. However, Samsung released the first generation of 3D solid state storage in 2013 and thereby ushered in a new era. AMD and Hynix aim to extend the concept of memory stacking to an interconnected 3D array. Obvious targets for their HBM offering include GPU, networking devices and high availability computing architectures.

65% Speed Improvement

The question remains: “How fast?” Hynix promotional material suggests they can achieve 40% power reduction and an incredible 65% speed improvement. Whilst these figures are “promised” and not confirmed, whatever the exact speed increase, miners can look forward to greater hashing power and, important for bottom line, power savings.

There are also questions regarding AMD’s commitment to the cryptocoin mining market. Are they even interested? Will AMD develop products aimed specifically at mining application, as opposed to gaming? We have already explored how mining derailed AMD marketing strategy surrounding the Pirate Islands series and it is unsure how they will address mining demand in the future.

AMD has not announced a Pirate Islands release date, other than saying “late 2014”. We will alert readers with any new information.

See also:

Last modified: April 28, 2014 10:34 UTC

More of: miningscrypt

Market analyst and Open source developer with a keen interest in blockchain technology, consensus mechanisms and the decentralizing effect. He has found a solution to the PKI mechanism. Email me to discuss.

Show comments