Monday, September 13th 2021
ZeroPoint Technologies wants to Compress the Data in your RAM
Some of you might be old enough to remember various "RAM doubling" software software solutions that appeared back in the late 1980's for Apple, as well as in the mid 90's for Windows 95 computers. Most of them never really delivered on their claims, but now it looks like we might be getting something similar, albeit in hardware.
Swedish company ZeroPoint Technology AB has announced that it has raised €2.5 million in a seed round to bring its Ziptilion patented memory compression technology IP to the market. ZeroPoint claims a compression ratio of two to three times depending on the workload, which seems very impressive. Unlike current software compression technologies such as ZSWAP or ZRAM that are used to compress data in RAM at a rate of 1.4 to 1.5 times, ZeroPoint promises that it's hardware IP won't have any real world negative effects on system performance. In fact, they claim it'll only cause one nanosecond of extra latency when writing data and 100 nanoseconds delay when it comes to reading the compressed data from RAM.
Ziptilion has already been implemented into a product using TSMC's 28 nm node, using an AXI bus at 800 MHz, which resulted in a memory throughput of 32 GB/s. It's not clear what product this is, but based on the manufacturing node it suggests it might just be an ARM based test platform. ZeroPoint claims that its IP is easy to integrate with existing hardware designs, as it's placed on the memory access path and thus doesn't require changes to the memory controller itself, although it does require a software device driver element which might limit some potential applications.
In something like a normal desktop computer on a 7 nm node, the Ziptilion IP is expected to only use up about 1.36 square millimetres for a standard dual-channel RAM configuration. On a server CPU with 8-channels of RAM, this is estimated to increase to about 3.02 square milimetres. ZeroPoint is expecting it's IP to find its way into servers, embedded systems and smartphone SoCs to start with and the company is already working on its next generation Ziptiolion+ technology which appears to further increase the compression factor depending on the specific target application. Not all kinds of data compress the same, based on ZeroPoint's whitepaper.
The boldest claim made by ZeroPoint is that they claim you'll see the same system performance gains in most of their tested scenarios, as you would from doubling your system RAM. In addition to this, ZeroPoint also claims to increase the effective memory bandwidth in a system, due to the compressed data taking up less bandwidth when being transferred between the RAM and the CPU. The whitepaper also seems to hint at the possibility of Ziptilion being used with GPUs, at least in something like an SoC, where the RAM is shared between the CPU and GPU.
It all sounds a little bit too good to be true, but the company has been researching its memory compression algorithms for the past 15 years, so it doesn't sound like a money grabbing scam, unlike the RAM doubling software back in the day.
Source:
ZeroPoint Technology AB
Swedish company ZeroPoint Technology AB has announced that it has raised €2.5 million in a seed round to bring its Ziptilion patented memory compression technology IP to the market. ZeroPoint claims a compression ratio of two to three times depending on the workload, which seems very impressive. Unlike current software compression technologies such as ZSWAP or ZRAM that are used to compress data in RAM at a rate of 1.4 to 1.5 times, ZeroPoint promises that it's hardware IP won't have any real world negative effects on system performance. In fact, they claim it'll only cause one nanosecond of extra latency when writing data and 100 nanoseconds delay when it comes to reading the compressed data from RAM.
In something like a normal desktop computer on a 7 nm node, the Ziptilion IP is expected to only use up about 1.36 square millimetres for a standard dual-channel RAM configuration. On a server CPU with 8-channels of RAM, this is estimated to increase to about 3.02 square milimetres. ZeroPoint is expecting it's IP to find its way into servers, embedded systems and smartphone SoCs to start with and the company is already working on its next generation Ziptiolion+ technology which appears to further increase the compression factor depending on the specific target application. Not all kinds of data compress the same, based on ZeroPoint's whitepaper.
It all sounds a little bit too good to be true, but the company has been researching its memory compression algorithms for the past 15 years, so it doesn't sound like a money grabbing scam, unlike the RAM doubling software back in the day.
25 Comments on ZeroPoint Technologies wants to Compress the Data in your RAM
Especially with them talking about adding their IP to the CPU side with memory controller - wouldn't it be bandwidth boost as well as size? GPUs have been using DCC for a while now for similar purpose. Although at a storage level, hardware compression of storage data before sending it over PCIe to NVMe SSD - that both new generation consoles do and DirectStorage will support on PC soon - falls into similar category.
I don't remember RAM compression being a thing on Amiga...
Not sure it was actually a real thing, but neither was the one for Windows 95 and they sold hundreds of thousands of copies.
en.wikipedia.org/wiki/SoftRAM
Whole memory encryption/compression is a hardware feature. Ideally it should be transparent to all upper layers and all computations. No matter inactive/active, OS/user level. It is a hardware architecture design.
A direct contrast: There is one region of memory that Windows 10 can never ever compress: itself.
Just wondering if they are planning to follow intel's skylake/14nm ++++++++++++++++++ path & milk it for every penny they can get, hehehehehe :)
It was very common back in the days to use a util like this. Software for "only" 49.95$ or so instead of a 4MB additional ramstick for 200 to 400$.
But it would add extra latency, cycles etc and back in the days with such hardware it was a huge performance impact compared to todays standards.
Ram compression at hardware level could actually work. You could dense far more data into a stick then without. Imagine if this would work on GPU's too. You could save perhaps half of the bandwidth normally needed for graphics or compute operations. Not something small.