New Broadcom RAID Controllers - MegaRAID 9760W

[DarkServant]

Broadcom released two PCIe 5.0 RAID-controllers with some interesting changes.
A 16 drive (MegaRAID 9760W-16i) and a 32 drive version (MegaRAID 9760W-32i) is coming.

This time they dropped the external DRAM cache for an on-chip integrated cache with only 24MB. It is unclear if they use eDRAM or (hopefully) SRAM. The BBU is still there, but much smaller and finds space on the card itself.
Of course it is faster, two times in random read and five times in random write RAID performance. The power consumption is in the 27-36 watt range, and the heatsink seems to be smaller.
The controllers use x16 cable connectors, several different cable options are available.

New Features are:
- Onboard energy backup
- CNSA 2.0 compliant Hardware Secure Boot and SPDM Attestation support
- PCIe Embedded Analyzer (PEA) technology

Link: Broadcom 97xx PCIe 5.0


If they just would have added an optional blower-fan, that these controllers could also be used in a workstation without the typical overheating problems.
In times of vapor-chambers, heatpipes etc. a heatsink that could cool the ASIC passively even in a single slot size should be possible, if some airflow in the case is available.

 

[i386]

The 9760W-16i has 8GByte ram according to alternate (Broadcom MegaRAID 9670W-16i, RAID-Karte german), but the information could be wrong (it lists the card as pcie 4.0)

If they just would have added an optional blower-fan, that these controllers could also be used in a workstation without the typical overheating problems.

This depends on the definition of "workstation" I think. "Proper" workstations with decent fans (high static pressure & airflow capabilities) and front to back airflow should have no problems. Workstations built with consumer/prosumer chassis with low noise fans and top exhaust fans don't work very well with add on cards (be it nics, raid controllers/hbas, gpus) like these.

 

[twin_savage]

Thanks for sharing!
That's interesting about the reduced cache size. At least for NVMe hardware raid it doesn't seem to depress the performance figures at all, storagereview has these new cards drastically outperforming basically all other software and hardware raid solutions.

If they just would have added an optional blower-fan, that these controllers could also be used in a workstation without the typical overheating problems.
In times of vapor-chambers, heatpipes etc. a heatsink that could cool the ASIC passively even in a single slot size should be possible, if some airflow in the case is available.

It should be pretty easy to 3d print a fan shroud for the card. We can even keep the fan shroud a single slot like this one:

 

[DarkServant]

The 9760W-16i has 8GByte ram according to alternate (Broadcom MegaRAID 9670W-16i, RAID-Karte german), but the information could be wrong (it lists the card as pcie 4.0)

This depends on the definition of "workstation" I think. "Proper" workstations with decent fans (high static pressure & airflow capabilities) and front to back airflow should have no problems. Workstations built with consumer/prosumer chassis with low noise fans and top exhaust fans don't work very well with add on cards (be it nics, raid controllers/hbas, gpus) like these.

This is a new one, the naming is confusing, and i predict already several wrong deliveries... what you mean is this card:

And it has IMHO a bigger heatsink which is better for passive cooling. And 18 chips of DRAM means it has real ECC -> six are on the backside.


The overheating and resulting catastrophic failure was on another card with a bad joke of a heatsink -> LSI 9266-16i


The overheating troubles started with the RoC (RAID-on-Chip) designs.
On GPU's i saw the first overheating problems on an AGP 3Dfx Voodoo 3 3000, and i added an 40mm fan with "wood-screws" which got rid of the problem, even a schoolmate had the same problem and after he did the same, the "white-screens" stopped too.

I have seen a lot of this 3D-printer mods on SAS-HBA's, nice work.
Sad to see, that even the technology for proper cooling exists, it does not get used on this not quite cheap HBA's and controllers. Ok, without any forced air circulation only through convection, it's nearly impossible in a modular system.

It's now many years ago, and with the rising of SSD's, especially server-grade ones, the failure rate was acceptable for me and my use cases.
I rather look that the SSD's stay below 50° Celsius, which accounts for a half a million hours more in MTBF according to newer datasheets; do not fall for the "an SSD has to be hot to get longer lifetime out of the flash", which is worth nothing with a dead controller, a defective PMIC or other essential parts.



PS: has anyone already an Optane P5800X which died?
The use of aluminum-capacitors instead of the solid tantalum ones is not a promising change. No, space is not always the problem, on the samsung xs1715 are 40 solid tantalum capacitors... on the other side with a reliable monitoring system of the capacitor state, it is much easier to replace such big ones. I saw on pictures that some variants of newer SSD's like the e1.s versions sometimes have the solid-capacitors in contrast to the u.2 versions.