Embedded Systems Conference
Breaking News
Comments
Oldest First | Newest First | Threaded View
Page 1 / 2   >   >>
TarraTarra!
User Rank
Author
DDR4 replacement??
TarraTarra!   3/11/2014 1:02:52 PM
NO RATINGS
HMC, HBM and other technologies are certainly very interesting but I dont see how they would replace commodity DDR4 DRAM anytime soon. Am I missing something here?

 

The cost/bit of DDR3/DDR4 will continue to be much lower than what HMC/HBM will offer for a long time. HMC and HBM do offer interesting possibiities with on-die memory stacking, fast look-aide memory but will they replace DDR4 a the main memory for processors?

 

 

alex_m1
User Rank
Author
Re: DDR4 replacement??
alex_m1   3/11/2014 1:47:37 PM
NO RATINGS
Let's look at rought cost estimates of cloud computing:


Power related costs are nearly 40% of total, with saving in memory power could reduce 15% of those 40%, meaning 6% of total.

Since server costs are 50% of total ,those saving comes down to 3% of server cost.

Dell charges for a "performance" server $4100K :

"Dual Intel® Xeon® E5-2620 processors, 16GB memory, 500GB hard drive and 3 year ProSupport warranty"

http://www.dell.com/us/business/p/poweredge-r720/fs

 

3% * 4100 = $123 = $7.6 per GB of power savings (without speaking about other benefits). Considering the fact that DDR(consumer- serever costs more) costs $11.5 per GB , those power savings look like a good reason for adding an HMC to a cloud server even as main memory. 

TanjB
User Rank
Author
Re: DDR4 replacement??
TanjB   3/11/2014 3:52:06 PM
NO RATINGS
Power costs at 10 cents/kWh are about 80 cents per year per watt.  Many data centers have lower contracts than that, but should add in overheads to build the power distribution and substation so it is on balance a fair estimate.  A 200W server (which includes the overhead ratio for a modern DC) is costing about $160 per year for power.  This puts the cost of power under 10% of operating costs.

The bigger cost is opportunity cost.  A DC is generally built out to the max power the site can obtain, so every watt in that 200W must be optimized.  If you add 1W in DRAM then you subtract 1W from something else.  Each W thus has an opportunity cost of about 0.5% of the server which more like $10 per year.

So, you have the right conclusion, but it can be helpful to have a better model of why.

Components which save power in the server market are definitely valuable.  Each W saved is worth dollars, so long as performance is not compromised.

TanjB
User Rank
Author
how does this relate to servers?
TanjB   3/11/2014 3:54:36 PM
NO RATINGS
The size of the modules (2GB) makes this look like a technology for local GPU memory or perhaps even for mobile devices.  But servers?  How does this add up to 128GB or more seen on modern servers?  Is there something else in plan for that market, or does that remain DDR4 territory?

alex_m1
User Rank
Author
Re: how does this relate to servers?
alex_m1   3/11/2014 6:13:18 PM
NO RATINGS
Since each hmc has 8 layers of dram, my guess is that for first version they prefer to use old manufacturing lines and get some money from them, instead of taking a risk a reusing new manufacturing lines.

 

resistion
User Rank
Author
TSV cost reduction
resistion   3/11/2014 9:29:21 PM
NO RATINGS
What is the plan for TSV cost reduction? Is it the TSV fabrication that is the key cost or TSV-related integration into the product (like testing)?

TanjB
User Rank
Author
Re: how does this relate to servers?
TanjB   3/12/2014 1:09:39 AM
NO RATINGS
The point is that the chip stack is overkill for large space.  100GB/s bandwidth per 2GB cube is way overkill for building a server with large memory - 50 modules like that, what host chip will have the interconnect or even need it?  And you can't get much bigger cubes because that is the limit you get multiplying DRAM chip capacity x number of TSV layers possible.  So, the whole thing looks optimized for small-memory scenarios.

Where is the server equivalent, or is this simply not coming to a server any time soon?

 

alex_m1
User Rank
Author
Re: how does this relate to servers?
alex_m1   3/12/2014 7:13:42 AM
NO RATINGS
Multiple HMCs can be chained together to appear as a single, mega-humongous memory.

DougInRB
User Rank
Author
Re: how does this relate to servers?
DougInRB   3/12/2014 2:22:23 PM
NO RATINGS
Even though you can chain them together, you are limited to 8 HMC parts per channel.  So, the CPU will need multiple channels to support a large memory server.  That's no problem - they already support multiple channels with far more signals required for DDR4 than for HMC.

The real problem is the size of the HMC.  They are huge (31mmx31mm)!  You can't cram enough of those on a motherboard or DIMM to get a server with 1.5TB of DRAM like you can with the 64GB DDR4 DIMMs that will be available later this year.

 

 

alex_m1
User Rank
Author
Re: how does this relate to servers?
alex_m1   3/12/2014 2:41:29 PM
NO RATINGS
My guess is that a large enough part of the server market could manage with 16GB which comes down to 4 chained HMC's.  Seems possible.

Even 32GB might work.

Page 1 / 2   >   >>


Radio
NEXT UPCOMING BROADCAST
In conjunction with unveiling of EE Times’ Silicon 60 list, journalist & Silicon 60 researcher Peter Clarke hosts a conversation on startups in the electronics industry. One of Silicon Valley's great contributions to the world has been the demonstration of how the application of entrepreneurship and venture capital to electronics and semiconductor hardware can create wealth with developments in semiconductors, displays, design automation, MEMS and across the breadth of hardware developments. But in recent years concerns have been raised that traditional venture capital has turned its back on hardware-related startups in favor of software and Internet applications and services. Panelists from incubators join Peter Clarke in debate.
Flash Poll
Top Comments of the Week
Like Us on Facebook

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
Special Video Section
The LT®3042 is a high performance low dropout linear ...
Chwan-Jye Foo (C.J Foo), product marketing manager for ...
The LT®3752/LT3752-1 are current mode PWM controllers ...
LED lighting is an important feature in today’s and future ...
Active balancing of series connected battery stacks exists ...
After a four-year absence, Infineon returns to Mobile World ...
A laptop’s 65-watt adapter can be made 6 times smaller and ...
An industry network should have device and data security at ...
The LTC2975 is a four-channel PMBus Power System Manager ...
In this video, a new high speed CMOS output comparator ...
The LT8640 is a 42V, 5A synchronous step-down regulator ...
The LTC2000 high-speed DAC has low noise and excellent ...
How do you protect the load and ensure output continues to ...
General-purpose DACs have applications in instrumentation, ...
Linear Technology demonstrates its latest measurement ...
10:29
Demos from Maxim Integrated at Electronica 2014 show ...
Bosch CEO Stefan Finkbeiner shows off latest combo and ...
STMicroelectronics demoed this simple gesture control ...
Keysight shows you what signals lurk in real-time at 510MHz ...
TE Connectivity's clear-plastic, full-size model car shows ...