HyperCloud to own the 32GB market ?

Marketability as a better RDIMM

UPDATE: 07/09/2012 – buyout
UPDATE: 07/09/2012 – strategic value of RDIMM-compatibility and misconceptions debunked
UPDATE: 07/27/2012 – confirmed HCDIMM similar latency as RDIMMs
UPDATE: 07/27/2012 – confirmed LRDIMM latency and throughput weakness

Is Netlist a buyout candidate ?

Is LRDIMM a dead-end product ?

Is it easier to market a better RDIMM that includes all the features of an LRDIMM ?

I think we have already demonstrated that Netlist will own the 32GB market – due to the constraints operating in the memory space:

– need for load reduction and rank multiplication at higher memory loading, higher speeds, lower voltages (all of these exacerbate the need for that technology)
– need for rank multiplication at 32GB memory size – the non-availability of 8Gbit DRAM die (monolithic) will prevent creation of a true 2-rank 32GB memory module without it

Now we will put together that information to see the strategic position for HyperCloud, RDIMMs and LRDIMMs in late 2012 and going forward into 2013.

This comparison will make obvious the assertion that the 32GB market will belong completely to HyperCloud.

Why HyperCloud will own the 32GB market – revisited

As we have examined, the market for memory is starting to look like:

– 16GB for 2012
– 32GB with 2%-3% attach rate, ramping starting late 2012 into 2013 and beyond
– 64GB ramp after that

HyperCloud is selling into:

– 16GB – 3 DPC market
– 32GB – whole market (dominates RDIMM, LRDIMM in performance, latency, price and IP issues)

RDIMMs are selling into:

– 16GB – 1 DPC and 2 DPC
– 32GB – non-viable

LRDIMMs are selling into:

– 16GB – non-viable
– 32GB – non-viable vs. HyperCloud

Use case for HyperCloud

This suggests that anyone who controls HyperCloud will control memory for 2013.

HyperCloud has a niche at the 16GB level (which LRDIMMs do not).

However, HyperCloud will own at the 32GB memory market – 32GB HyperCloud outperforms both RDIMMs and LRDIMMs in performance, price, latency and IP issues.

The reasons as examined earlier are:

– performance – HyperCloud outperforms both 32GB RDIMM and 32GB LRDIMMs
– latency – HyperCloud has better latency than LRDIMMs
– price – HyperCloud uses 4Gbit monolithic leveraging Planar-X vs. 4Gbit x 2 DDP for 32GB RDIMM and 32GB LRDIMM
– IP – HyperCloud owns the IP on load reduction and rank multiplication (Inphi is infringing and lost patent reexams at the USPTO)
– RDIMM-compatible – LRDIMMs cannot claim this feature

The simplicity of an DDR3 RDIMM standard memory module

Yet HyperCloud looks like an RDIMM – it is inter-operable with RDIMMs, and conforms to the JEDEC RDIMM spec in operation.

The additional simplicity of HyperCloud being RDIMM-compatible should not be taken lightly.

Anyone who controls 32GB HyperCloud, could sell a “32GB RDIMM” product starting late 2012 and going into 2013 – and make it THE dominant 32GB product in the market.

HyperCloud are an RDIMM for all practical purposes – except they have the load reduction and rank multiplication features (that will be essential going forward) built into the memory module in a seamless way so that it still looks like an RDIMM.

The familiarity of DDR3 RDIMM standard

This will be a relief to all those who have yet to fully understand the LRDIMM “standard”.

These users are familiar with DDR3 RDIMMs and still not aware of the pitfalls present in the LRDIMM standard. Those who have not fully understood these problems will buy LRDIMMs and later face problems when they do not live up to the marketing speak. These complications will increase the cost of support for memory module makers.

LRDIMMs as an end-of-life product

There will thus be no need for LRDIMMs – and they will become an “end-of-life” product. The reasons are:

– LRDIMMs will not work with RDIMMs
– if companies stop the manufacture of LRDIMMs (loss of interest or on orders of a court for IP infringement), those who have already bought LRDIMMs will not find replacements (faulty LRDIMMs slots will have to go empty as you can’t use RDIMMs in there)
– LRDIMMs are under threat of injunction to stop their product (USPTO ruling on Netlist patents has sealed the fate for Inphi’s LRDIMM roadmap)
– LRDIMMs are not dovetailing into DDR4 – JEDEC is having to copy HyperCloud in even greater detail for DDR4 (symmetric lines and de-centralized buffer chipset)
– LRDIMMs require a BIOS update to work – does this suggest bug fixes will become a regular feature for users of LRDIMMs in existing data centers – a decidedly unpalatable future ?

VMware memory of choice for virtualization on blade servers

VMware certifies Netlist as the sole memory vendor for it’s products. The Netlist 16GB, 32GB HyperCloud (supplied by IBM/HP) and the Netlist 16GB VLP RDIMM (supplied by IBM) are the only memory products certified for use with VMware:

Memory for VMware virtualization servers
July 5, 2012

I have not seen VMware endorse LRDIMMs anywhere.


For more details, check out these articles:

Infographic – memory buying guide for Romley 2-socket servers
June 29, 2012

Memory buying guide – including 1.35V memory for Romley
June 28, 2012

For more on 32GB RDIMM/32GB LRDIMM use of 4Gbit x 2 DDP memory packages:

DDP vs. monolithic memory packages and their impact
June 25, 2012

Multi-die vs. multi-PCB to increase memory density
July 1, 2012

On the non-viability of 16GB LRDIMMs and 32GB RDIMMs:

Why are 16GB LRDIMMs non-viable ?
June 19, 2012

Non-viability of 32GB RDIMMs
June 20, 2012

On the non-viability of LRDIMMs in general:

Examining LRDIMMs
July 5, 2012

On the viability of HyperCloud:

Examining Netlist
July 3, 2012

On DDR4 borrowing from LRDIMM use of Netlist IP in “load reduction” and “rank multiplication”:

DDR4 borrows from LRDIMM use of load reduction
June 8, 2012

JEDEC fiddles with DDR4 while LRDIMM burns
June 7, 2012

On the risk factors for LRDIMM:

LRDIMMs future and end-user risk factors
June 5, 2012

Why are LRDIMMs single-sourced by Inphi ?
June 15, 2012

Netlist has suggested that LRDIMMs are an “end-of-life” product.

See the section entitled “LRDIMMs – cul-de-sac for an “end-of-life” product” in the article:

Examining LRDIMMs
July 5, 2012

For a glimpse of Inphi’s weak position at the USPTO:

Examining patent docs at USPTO II – sloppy appeals
July 8, 2012

UPDATE: 07/09/2012 – buyout

Buyout possibilities

Why the mention of a buyout ?

As examined in this article, there is a need for HyperCloud to see a second-source for the eventual 32GB mainstream market.

A second-source for HyperCloud ?
July 8, 2012

In addition IDTI has kept it’s powder kegs dry – it has NOT entered the LRDIMM space for Romley rollout (despite it’s importance for creating early relationships with module makers and OEMs). This means it was a very conscious decision on the part of IDTI to avoid LRDIMM for Romley.

Of the buffer chipset makers – Inphi, IDTI and Texas Instruments – IDTI has been non-aggressive in the LRDIMM vs. Netlist battle that Inphi has been occupied with. Though I guess TI could be considered given their early settlement in Netlist vs. Texas Instruments and their subsequent absence from the LRDIMM space altogether.

However, what is clear is that the whole 32GB memory space and beyond (64GB) is up for grabs.

Whoever negotiates a deal with Netlist will be in a position to offer a superior product as a regular DDR3 RDIMM compatible product, that dovetails nicely into DDR4 (skipping LRDIMM in the process of course).

As Netlist suggests, they are already shipping a DDR4 type product. At DD4 the need for load reduction and rank multiplication will increase – with higher frequency and lower voltage.

HyperCloud for it’s part will need a second-source.

Although that need does not exist now, such a source will have to start work NOW in order to deliver mid-2013 (though if they started earlier they could deliver by Ivy Bridge – sounds like IDTI ?).

IDTI has suggested they will deliver a “LRDIMM” solution by late 2012 to address Ivy Bridge. Given IDTI behavior up to this point – a deemphasizing of LRDIMMs after initially being gung-ho – it would not be surprising if there is a further tilt away from LRDIMMs and this product winds up appearing as a “better RDIMM” (i.e. HyperCloud).

Conclusion – the marketability of a better RDIMM

Anyone who controls HyperCloud will have the ability to sell a better RDIMM without labelling it as a new standard LRDIMM.

This is a tremendous advantage as it does not require the per-motherboard fiddling of the BIOS required with LRDIMMs.

And it does not entail BIOS updates to be flashed at a later date, after the server is in operation in the data center.

This should reduce costs for motherboard makers as well – as the motherboard no longer have to maintain an LRDIMM-specific thread in their BIOS development and upgrade roadmap.

UPDATE: 07/09/2012 – strategic value of RDIMM-compatibility and misconceptions debunked

The strategic value of a RDIMM-compatible next-gen memory

Even for folks familiar with HyperCloud and LRDIMM, it is a common misconception that HyperCloud needs to surmount some type of validation at Intel, or some approval at JEDEC.

These misconceptions arise from an extrapolations of the problems evident with LRDIMMs – which force a standardization effort at the memory module level as well as the motherboard level (a BIOS update is required otherwise the LRDIMMs do not work) – and extending them to HyperCloud (which LRDIMMs are presumably copying and thus should have “the same problems”).

The argument being that since a new standard LRDIMM could only succeed because Intel and JEDEC were hand-holding it, how is HyperCloud going to fend for itself without such cover ?

The answer is that HyperCloud is actually a SUPERSET of the LRDIMMs. It includes capability not existant in the LRDIMMs.

HyperCloud memory modules are RDIMM-compatible and RDIMM-interoperable.

This places Netlist pushing HyperCloud in the same situation as any other RDIMM memory module maker.

The only difference is that at the 32GB level, Netlist will be making these much better than any 32GB RDIMM maker could maker their product. So much so that even the new standard LRDIMMs (with all their attendant complexity) cannot compete.

The strategic value of an RDIMM-compatible load reduction and rank multiplication product should not be underestimated

Misconceptions – Intel validation

Intel generally stays out of the memory business.

However since LRDIMMs were a new standard which required changes not only at the memory module level, but ALSO required a BIOS update on the motherboards (without with LRDIMMs cannot work), AND because there was a risk that OEMs and motherboard makers may slack off on the effort, for these reasons partipation by Intel was required to push it through.

Intel therefore had to be especially aggressive about qualifying LRDIMMs – this because there was a risk that without such a push, the OEMs and motherboard makers may not have bothered to make the effort to get these devices to run.

This push by Intel allowed acceptance in the industry of a load reduction and rank multiplication solution (which incidentally was copying Netlist IP :-)).

However, Intel does not generally qualify all memory. Most often memory products get qualified at third-party labs, and are approved by Intel based on those results. The Intel validation page requires that you get your product qualified at one of the Intel-approved labs – like CMTL Labs – and then submit the paperwork.

CMTL has previously qualified HyperCloud – and it is CMTL which has created this analysis of HyperCloud vs. LRDIMMs:

Here is a CMTL labs comparison of HCDIMM and LRDIMM that goes into more detail:

CMTL HCDIMM Outperforms LRDIMM in “Big Data” & “Big Memory” Applications White Paper

Further qualification of HyperCloud does not require Intel consent.

Misconceptions – JEDEC approval

Now coming to JEDEC – JEDEC standarization is required in order to make LRDIMMs work – to ensure consistency at the memory module level, AND to ensure motherboard makers know how to make the BIOS updates required for LRDIMMs to work.

JEDEC standarization ensures that LRDIMMs and DDR4 work.

Licensing of appropriate IP (like HyperCloud IP) is required to ensure LRDIMMs and DDR4 have legal cover. Usually licensing should be secured prior to finalization of the standard, and not after when you have no negotiating power over the patent owner.

If JEDEC does not license Netlist IP, it is their choice and users of LRDIMMs and DDR4 will suffer the consequences later.

It is an issue completely unrelated to the deployment of HyperCloud.

A real problem – qualification on a wider range of servers

The only real problem plaguing HyperCloud is that they are not available on a server platform when customers want to buy it on that platform.

This is something Netlist will have to work on – as qualification takes some money, but a lot of time.

Netlist may be happy generating volume on the top 3 high volume virtualization server lines, but they eventually need to be available on every server platform in order to satisfy customer needs.

Conclusion – LRDIMMs as a subset of HyperCloud capability

Because of the effort required to standardize across memory module makers AND to ensure motherboard makers are implementing the required BIOS upgrades, LRDIMMs would fail if not for the stabilizing hand of Intel and JEDEC.

HyperCloud being an RDIMM-compatible product does not require Intel or JEDEC handholding.

This is a very crucial point in debunking the “how will HyperCloud survive without Intel validation and JEDEC approval – LRDIMMs have that support – what will HyperCloud do” misconception.

There is a presumption that since LRDIMMs are copying HyperCloud, that HyperCloud has the same limitations as LRDIMMs – which require the benevolent hand of Intel and JEDEC to cure.

However this is not the case.

HyperCloud embodies capability that is a SUPERSET of LRDIMM capability.

Notably, HyperCloud are RDIMM-compatible.

LRDIMMs implement a SUBSET of the capability present in the HyperCloud. For example, LRDIMMs cannot interoperate with RDIMMs

So Netlist value is that they have an innovation in hand – and that innovation does NOT require cooperation of Intel or JEDEC or any other entity – Netlist is free to sell HyperCloud, just like Smart Modular and other memory module makers are free to sell DDR3 RDIMM memory modules.

This means it is also a capability which has value for other memory module makers – and when 32GB arrives, this capability will be essential for them to compete with DDR3 RDIMM-compatible 32GB HyperCloud.

And it is this requirement which provokes discussion of buyout or partnership.

The RDIMM-compatibility is a feature we don’t hear of much, but it is a very important strategic advantage, both to Netlist and to any potential buyer or partner.

UPDATE: 07/27/2012 – confirmed HCDIMM similar latency as RDIMMs
UPDATE: 07/27/2012 – confirmed LRDIMM latency and throughput weakness

HyperCloud HCDIMM latency similarity with 16GB RDIMM (2-rank) latency has been confirmed. And LRDIMM latency and throughput weakness vs. HCDIMM – even when running at the SAME lowered speeds of the LRDIMM – confirmed:

Latency and throughput figures for LRDIMMs emerge
July 26, 2012



Filed under Uncategorized

3 responses to “HyperCloud to own the 32GB market ?

  1. Pingback: HyperCloud branding should suggest a better RDIMM | ddr3memory

  2. Pingback: Inphi to report July 25 | ddr3memory

  3. Pingback: Awaiting 32GB HCDIMMs | ddr3memory

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s