Short answer: if your workload is heavy random writes — virtualization hosts, databases, frequent RAID 5/6 rebuilds — take the 4GB version. If you run file serving, backup targets and mostly sequential reads, the 2GB version does the same job for less money. Both are the same Broadcom Tri-Mode controller; only the cache differs.
What the two versions share
The 2GB and 4GB 9460-8i are the same card at the hardware level. The controller, ports and firmware are identical:
| Item | 9460-8i (both versions) |
|---|---|
| Internal ports | 8 × 12Gb/s SAS / SATA / NVMe (Tri-Mode) |
| Host interface | PCIe 3.1 x8 |
| RAID levels | 0, 1, 5, 6, 10, 50, 60 |
| Power-loss protection | CacheVault (supercapacitor) module |
| Form factor | Low-profile MD2, full-height bracket on most listings |
So the decision is not “which card is better”. It is “how much write cache does your workload actually use”.
What cache size actually changes
The cache on a RAID controller is a write buffer. In write-back mode the controller acknowledges a write as soon as it lands in cache, and flushes to disk in the background. This is what makes RAID 5 and RAID 6 usable at all — without cache, every parity update means multiple reads and writes across the stripe.
A bigger cache helps in three specific situations:
- RAID rebuilds. Rebuilding a degraded RAID 5/6 array is parity-heavy random I/O. On a 4GB version the rebuild windows we see in practice are noticeably shorter than on 2GB units with the same drives.
- Virtualization hosts. Many small VMs generate scattered writes. More cache absorbs the peaks; less cache means the array feels the spikes directly.
- Databases and mail servers. Small-block random writes are exactly what cache is for.
If none of those describe you, the extra 2GB mostly sits idle.
When the 2GB version is enough
File serving, backup targets, media storage, surveillance boxes, and any workload that is mostly sequential reads and writes. A 2GB cache absorbs write bursts fine, and sequential throughput on both versions is limited by the drives and the PCIe 3.1 link, not by cache size. If you are dropping a 9460-8i into a TrueNAS-style build to present one big volume, 2GB is the sensible pick — spend the difference on drives instead.
When you want the 4GB
RAID 6 arrays over eight NVMe or mixed NVMe/SAS drives, Proxmox or VMware hosts with a dozen-plus guests, and anything doing sustained parity writes. The 4GB cache also buys headroom for a future drive upgrade: a card that feels fine with six SSDs can fall behind when the array grows.
Do not run it without CacheVault
This is the part most used listings get wrong. The 9460-8i holds unflushed writes in cache; if power is cut with no CacheVault module attached, that data is gone and the controller falls back to write-through, which cuts performance hard. If the CacheVault supercapacitor is missing, aged, or reported failed in storcli, treat the card as incomplete — a replacement module is not optional. We list the controller with its CacheVault as one kit for exactly this reason.
Three things to check before you order a used one
-
Cache size in the firmware. Some listings round up. Run
storcli /c0 showand read the value the controller itself reports. - CacheVault health. Check the module state and capacity retention, not just that the connector is populated.
- Port condition and bracket. The 9460-8i uses two mini-SAS HD SFF-8643 internal connectors (four lanes each); verify all eight lanes negotiate, and note that SFF-8644 is the external variant used on the -8e and -16e parts. Confirm you get the bracket that matches your case.
Every 9460-8i we list is bench-tested on our own test line, and cache size plus CacheVault state are part of that test — ask for the test notes on any unit before you order.
Comparing across the family instead of within one model? The full MegaRAID buyer’s guide covers where the 9460 sits against the 9361, 9500 and 9560 series, and our MegaRAID selection guide goes deeper into workload matching.
FAQ
Is the 4GB cache version physically different?
Only in the memory chips and part marking. Ports, connector, PCIe interface and firmware are the same, and a 2GB unit can be told apart from a 4GB unit only by what the firmware reports or the part number.
Can I upgrade a 2GB 9460-8i to 4GB later?
No. The cache is soldered, not a DIMM slot. Buy the size you need up front.
Does cache size affect read performance?
Marginally. The cache helps most on writes. Read performance depends mainly on your drives and the PCIe 3.0 link.
What CacheVault do I need for the 9361-8i instead?
For the 9361-8i family the matching module is the CVM02 supercapacitor kit. The 9460-8i uses its own module generation; they are not interchangeable.
Send the part number and quantity and we will check availability. WhatsApp +86 13068616914 · zhongxindata.hk
What the controller cache actually does
- Read cache absorbs repeated reads, which matters for workloads with a hot working set
- Write cache lets the controller acknowledge writes before they reach the media, protected by the supercap
- The benefit scales with the size of the hot set, not with the array size
When 4GB is worth the difference
- Large read-mostly working sets benefit most; a 2GB cache handles a smaller working set equally well
- Write-heavy workloads benefit from cache depth because more can be absorbed during a burst
- For a modest array in a general-purpose server, the difference is usually not the deciding factor
Checking the cache on a used controller
- The part number states the cache size — 9460-8i variants ship with different cache configurations
- The cache is soldered on most 9460 models, so the cache size is fixed by the card you buy
- The supercap condition is the more important thing to verify on a used unit, and we report it per unit
Related reading
- MegaRAID 9560-16i — the next generation controller
- Data center RAID strategy — controller selection context
- CacheVault supercap warning — the related supercap article
- RAID controllers — category hub
Frequently Asked Questions
Does the 9460-8i need 4GB of cache?
Not in most cases. The larger cache helps when your hot working set exceeds the smaller cache, which usually means large read-mostly datasets or burst-heavy write loads. For a modest array with ordinary workloads, 2GB is adequate and the difference rarely shows up outside benchmarks.
Is more controller cache always faster?
No. Cache benefit scales with the size of the hot working set. If your working set fits in the smaller cache, the extra memory does nothing. Where it does help is a workload that repeatedly reads a dataset larger than the cache — then the hit rate and therefore the read latency improve.
Can I upgrade the cache on a 9460-8i later?
On most 9460 models the cache is soldered to the board rather than socketed, so the cache size is fixed by the card you buy. That makes the part number important — 9460-8i variants exist with different cache configurations, so we confirm the exact variant against the number you send us.
What is the supercap and why does it matter?
The supercap is the small rechargeable unit that preserves the write cache during a power event. Without a healthy one, the controller drops to write-through mode and any unwritten cache data is at risk. On a used controller it is the component most likely to be degraded, which is why we test and report its status on every unit we list.
How do you verify a used 9460-8i?
We clean, re-flash and bench-test on our own test line, then confirm the part number, cache size, connector count, port mapping and firmware level, and we read the CacheVault supercap status. We report the supercap state rather than leaving it to be discovered at install time. Ask for photos of the actual card before you commit.
Does cache size affect rebuild time?
Indirectly, and positively. Write cache with battery protection lets the controller drain outstanding writes at power-off rather than forcing a flush to media, which shortens the shutdown and recovery sequence around a rebuild. The dominant factor is still array size and drive speed.