Rambus Explores Future Memory Systems

[btarunr]

Rambus Inc. (NASDAQ: RMBS) today announced it will collaborate with Microsoft researchers in the exploration of future memory requirements for quantum computing. The expertise of Rambus in high-bandwidth, power-efficient memory architectures, combined with Microsoft researchers' knowledge of advanced system and data center design will be applied to drive new technology platforms.

"Existing computer architectures are reaching limits due to the ever increasing demands of real-time data consumption, which is driving the need to explore new high-performance, energy-efficient computer systems," said Gary Bronner, vice president of Rambus Labs. "By working with Microsoft on this project, we can leverage our vast expertise in memory systems to identify new architectural models."



"We've been investing in projects that advance our understanding of quantum computing along with its applications and implementation," said Douglas Carmean, Architect, Quantum Architectures and Computation Microsoft Research. "We are focusing on exploring theoretical and experimental approaches to creating quantum computers, designing software, hardware and other elements that support our research and direction. In this process, we are engaging with partners that have critical expertise, such as Rambus, to build new types of computing platforms."

Rambus has a long history of researching and exploring emerging platforms, developing meaningful technologies with broad applicability. As part of this program, Rambus and Microsoft will pool resources to further examine potential architectures that can greatly enhance memory capabilities in various settings to improve overall system performance. Specific details of the project are confidential.

View at TechPowerUp Main Site

 

[cyneater]

Shouldn't rambus team up with someone like Dwave?

 

[R-T-B]

The expertise of Rambus in high-bandwidth, power-efficient memory architectures

You know, I honestly have no idea how memory efficient rambus was back in the day, mostly because it was paired with the god-awful Pentium 4. Any actual figures to back this up?

 

[Frick]

You know, I honestly have no idea how memory efficient rambus was back in the day, mostly because it was paired with the god-awful Pentium 4. Any actual figures to back this up?

I'm pretty sure they've done other stuff since then. Wikipedia say they're the ones behind XDR RAM.

 

[RejZoR]

I've always been wondering what separates RAM (DDR) from VRAM (GDDR) for it to achieve such ridiculous clocks and bandwidth on graphic cards but not as RAM sticks. I mean, why not stick HBM to RAM sticks and call it a day? X99 has quad channel memory, meaning it has 256bit bus width, surely a lot faster RAM sticks could utilize this better.

 

[Zankza]

I've always been wondering what separates RAM (DDR) from VRAM (GDDR) for it to achieve such ridiculous clocks and bandwidth on graphic cards but not as RAM sticks. I mean, why not stick HBM to RAM sticks and call it a day? X99 has quad channel memory, meaning it has 256bit bus width, surely a lot faster RAM sticks could utilize this better.

GDDR5 has high, way higher latency.

 

[R-T-B]

I'm pretty sure they've done other stuff since then. Wikipedia say they're the ones behind XDR RAM.

I had to google it. Looks like XDR ram was a flop as well, only being used in the Playstation 3...

I'm not trying to be critical, but has this company done anything that isn't a flop? How are they still alive?

 

[RejZoR]

GDDR5 has high, way higher latency.

So, high latency is fine for graphics, but not for general computing (ie RAM)? What about HBM? Same thing as GDDR5 as far as latencies are concerned?

 

[Frick]

I had to google it. Looks like XDR ram was a flop as well, only being used in the Playstation 3...

I'm not trying to be critical, but has this company done anything that isn't a flop? How are they still alive?

I've no idea. Mostly they've been fighting patent wars I think. But if MS partners with them they might have something to show.

 

[Keullo-e]

I had to google it. Looks like XDR ram was a flop as well, only being used in the Playstation 3...

And PS2 used RDRAM, both the memory types being just pure "awesomeness"

 

[KithKhan]

I had to google it. Looks like XDR ram was a flop as well, only being used in the Playstation 3...

I'm not trying to be critical, but has this company done anything that isn't a flop? How are they still alive?

Patent trolling.

 

[Deleted member 67555]

Should change the title to "Rambus explores any and all future memory designs in order to patent troll indefinitely"
Because that is more than likely what they are doing.

 

[KainXS]

So, high latency is fine for graphics, but not for general computing (ie RAM)? What about HBM? Same thing as GDDR5 as far as latencies are concerned?

someone correct me if I'm wrong but

GDDR5 has more bandwidth then DDR3/4 but higher latency so its good for moving big chunks of data but the smaller the chunks get the more latency impacts its performance.

DDR3 and 4 have less bandwidth but lower latency so they are better with the smaller chunks(more general).

HBM has better latency than GDDR5 I think,

 

[Prima.Vera]

Latest news from Google:

http://www.technologyreview.com/news/544421/googles-quantum-dream-machine/

 

[ArdWar]

someone correct me if I'm wrong but

GDDR5 has more bandwidth then DDR3/4 but higher latency so its good for moving big chunks of data but the smaller the chunks get the more latency impacts its performance.

DDR3 and 4 have less bandwidth but lower latency so they are better with the smaller chunks(more general).

HBM has better latency than GDDR5 I think,

You'll get worse with small, repeated and segmented data transfer on HBM. It's basically a ridiculously wide, crawling highway: given enough width, you can move more cars than faster, narrower highway, but it's terrible if you want to move from point A to B ASAP.

And, by definition, how we supposed to implement HBM on RAM stick? Unless you bypass the stick altogether and just bolt the RAM directly into processors.

 

[hellrazor]

So, high latency is fine for graphics, but not for general computing (ie RAM)? What about HBM? Same thing as GDDR5 as far as latencies are concerned?

Most of the memory bandwidth going through a GPU is texture data, so it'd be beneficial to waste a few cycles to start a transfer if it means that the actual transfer is shortened by tens of thousands of cycles, but on your typical CPU you're usually asking for many more chunks of memory that are much smaller in size, thus latency matters a lot more.

 

[Katanai]

Rambus explores future lawsuits < That would have been a more appropriate title for this post.

 

[cyneater]

Rambus explores future lawsuits < That would have been a more appropriate title for this post.

So true...

DEC never used rambus they wanted to but the cost was to much.
SGI looked into rambus but cost...
I think nintendo did on there N64 and intel did...

Other than that it was extremely expensive.
And rambus sued people for using and not using there tech....

 

[Uplink10]

Patent trolling.

Should change the title to "Rambus explores any and all future memory designs in order to patent troll indefinitely"

If I made perpetuum mobile I would give it only to companies which aim to improve the world and I would let others watch their downfall. That is how the world is, it is their right, and they should definetly do it because the patents exist, you don't like that then don't support the idea of patents. I wish they milk the $$$ out of everyone!