Remember those LSI SAS-12 chips SemiAccurate told you about last year? Last summer, LSI was only talking about them, but yesterday they were showing some new parts off.
LSI was showing off both the SAS-12 ROCs and a SAS-12 expander during their Asian sales conference. The demo is said to be an improved version of their IDF showcase one, basically a SAS-12 card connected to a SAS-6 expander. This was then plugged in to a rack full of SSDs, and the total throughput was paltry 3022MBps.
I know, you are thinking to yourself that if you only get a bit over three GBps from your drives, is it even worth the bother? It is far too slow for the lone enthusiast let alone performance sensitive applications, but help is on the way. The bottleneck in the IDF demo appears to be the SAS-6 expander because that is the main bit that LSI changed between demos. Yesterday’s version of the demo had not just the LSISAS3008 controller but also an LSISAS3x48 12Gbps expander in place of the 6Gbps one.
Does it blend? Err, no, but it wasn’t meant to either. That said, the throughput of the array, this time connected to 12 SSDs from various manufacturers, pushed over 6GBps sustained read speed. That is much closer to snappy, especially when you consider it all goes through a single PCIe3 card. When these chips finally end up in products you can buy, and the SAS-12 drives flood the market to use with them, even the most complex game levels should load in tolerable times.
In case you can’t see the sarcasm dripping from the above descriptions, 6GBps, not Gbps as we keep pointing out, is seriously fast. The mere fact that you can pump this level of I/O through a single PCIe3 card is nothing short of astounding. Even if you just have one of the inevitable PCIe3 LSI SAS-12 RAID cards with single digit number of drives connected, you should be able to support some very heavy workloads in a DAS configuration. It is unlikely to be cheap for the next year or two but good luck getting this level of sustained performance from SATA. SAS-12 can’t come soon enough.S|A
Latest posts by Charlie Demerjian (see all)
- Globalfoundries 7nm process isn’t even close to the name - Sep 26, 2016
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016
- Everspin and Globalfoundries team up for embedded ST-MRAM - Sep 15, 2016
- Intel’s Xpoint is pretty much broken - Sep 12, 2016
- ARM adds 2048-bit vectors to v8A with SVE - Sep 7, 2016