Any future for SRAM?

[m25]

According to this article, this stuff on 65nm packs 5Mbit/mm2 and 10Mbit/mm2 on 45nm but runs at a slower 400MHz. Is it any future for large, on die (but slower) L3.
http://www.eetimes.com/news/semi/showArticle.jhtml?articleID=196601127

 

[gOJDO]

What they were speaking about is Z-RAM, not S-RAM. The difference between them is huge.

 

[pmr]

Here's the difference!

ZRam

and

Sram

To answer to the question, I think Sram stills growing on the mtb market, since it owns Rock-Shox, Avid and Truvativ, it can compete with Shimano

(Sorry, I had to do it)

 

[m25]

What they were speaking about is Z-RAM, not S-RAM. The difference between them is huge.

Yes, however, is there any benefir of having some extra 400MHz on-die RAM?!

 

[gOJDO]

Depends on what do you compare 400MHz to.
If compared to 800MHz, it is a huge slowdown. If compared to 4GHz than it is almost unnoticable slowdown.

 

[Julian33]

Presumably latency if not bandwidth will be much lower with it being on die.... this is why it would make sense to use it for the L3 cache, since it would serve the purpose of hiding the latency of DDR2.

400MHz does seem rather slow though. Dosn't current SRAM run at the full clockspeed of the chip? I guess it would be possible to run at 400MHz and still achieve higher bandwidth than external RAM if they used a wide connection - 256 or 512 bits as opposed to the 2x64 you have for dual channel RAM? Don't know if it would really be feasible to have such a wide bus though.

 

[elpresidente2075]

Depends on what do you compare 400MHz to.
If compared to 800MHz, it is a huge slowdown. If compared to 4GHz than it is almost unnoticable slowdown.

As much as I'd like to say I understand what you're saying, I just can't. That just don't make no kinda sense! [/hick] But seriously, I would definitely notice a slowdown from 4GHz to 400MHz. What am I missing here?

 

[m25]

With much lower latency, it should serve as an intermediate memory between motherboard RAM modules and CPU speed cache (at least theorically)

 

[gOJDO]

we are talking about CPU cache memory. I bet on my ass that you will not notice the difference between 4GHz and 3.6GHz cache.

 

[m25]

Also found this:
http://techreport.com/onearticle.x/11371
Don't know how this could affect AMD's chips but seems to help more Intel's arch, which benefits from large cache; 625KB/mm2, you could get 30MB out of 50 mm2 and that (32M) used to be standard SDRAM equipment some time ago.

 

[gOJDO]

Well, if it clocks up to 400MHz, than it can be used for making fast RAM memory modules. For example DRAM cells on the DDR-400 and DDR2-800 are operating at 200MHz. There are faster RAM types with more agressive clock, like DDR-600 or DDR2-1200, but are very expencive and are consuming much more energy/dissipating much more heat.
So by changing the regural DRAM with ZRAM it is possible to get twice fast RAM(DDR ZRAM-800, DDR ZRAM2-1600) with extremly high power efficiency. I wonder what kind of latencies such memory will have and how many operations per second they will be able to complete.

 

[ajfink]

AMD owns the rights to use it for a reason, you know.

 

[elpresidente2075]

Because they bought it. It says so in the article.

@gOJDO: Also according to the article it says it clocks up to 400MHz, so I don't see where the difference between 4 and 3.6 GHz is coming from. How is this a subtractive thing? Am I just missing the point?

 

[exit2dos]

Intel does not use Silicon On Insulator (SOI) technology. Z-RAM requires SOI, which AMD does use.

There was talk of Intel utilizing SOI at some point in the future, But now it looks like they have advanced their High-k Metal Gate technology, so SOI will probably not be needed.

 

[ajfink]

Intel does not use Silicon On Insulator (SOI) technology. Z-RAM requires SOI, which AMD does use.

There was talk of Intel utilizing SOI at some point in the future, But now it looks like they have advanced their High-k Metal Gate technology, so SOI will probably not be needed.

I thought I heard that most experts agree High-K isn't practical until 32nm. Then again, stuff like this gets said all the time, then someone does it.

 

[exit2dos]

That may be true.

Intel already is taping out 45nm, of which I haven't heard of them utilizing High-K, so it may indeed wait until their next step - which will be 32nm.

 

[m25]

Well, if it clocks up to 400MHz, than it can be used for making fast RAM memory modules. For example DRAM cells on the DDR-400 and DDR2-800 are operating at 200MHz. There are faster RAM types with more agressive clock, like DDR-600 or DDR2-1200, but are very expencive and are consuming much more energy/dissipating much more heat.
So by changing the regural DRAM with ZRAM it is possible to get twice fast RAM(DDR ZRAM-800, DDR ZRAM2-1600) with extremly high power efficiency. I wonder what kind of latencies such memory will have and how many operations per second they will be able to complete.

I am very enthusiast about it, if it makes RAM cheaper However, the use is commented to be for on die cache, though keeping it clocked right takes an independent clk generator.