Seagate SandForce SF3500 Performance Data Leaked

After our conversation with Seagate, we really didn't expect to uncover this information until our upcoming briefing in August during the Flash Memory Summit. You can imagine our surprise when the data fell right into our laps from a close source outside of Seagate. 

The upcoming 4-channel SF3500 is the client version of Seagate's upcoming SandForce SF3000 series. The SF3700 controller is firmly at the top of the series and will ship in enterprise products with several additional features designed for enterprise environments. Meanwhile, the upcoming SF3500 will play a large role in the client SSD arena when released. Since their arrival, SandForce controllers have steered the market until slowly fading into the background as the company ping-ponged between LSI and Avago, before finally finding a home at Seagate.

Swipe to scroll horizontally
SpecificationsSF3514SF3504SF3524
ApplicationsMainstream SATA ClientEntry PCIe ClientEnthusiast SATA / PCIe Client
DuraClass TechnologySF3000
Architecture1TB1TB1TB
Host InterfaceSATA 6 Gb/sPCIe 2.0 x2 (NVMe)PCIe 2.0 x2 (NVMe)SATA 6 Gb/s
Max Capacity Supported1TB1TB1TB
Controller Clock Frequency275 MHz275 MHz300 MHz
Performance
Sequential ReadUp to 550 MB/sUp to 900 MB/sUp to 900 MB/s
Sequential WriteUp to 450 MB/sUp to 525 MB/sUp to 525 MB/s
Random ReadUp to 100,000 IOPSUp to 130,000 IOPSUp to 130,000 IOPS
Random WriteUp to 80,000 IOPSUp to 90,000 IOPSUp to 90,000 IOPS
Random 70% Read MixUp to 80,000 IOPSUp to 120,000 IOPSUp to 120,000 IOPS
Flash Memory SupportMLC, TLC, 3D from top flash memory manufacturers4-channels up to 400 MT/s1x nm, 1y nm, 1z nm, ONFi 2.0/3.0, Toggle 1.0/2.0
Sector Size Support512b4KB4KB (PCIe NVMe)512b (SATA)
SecurityDual AES-256 EncryptionTCG Opal 2.0IEEE-1667Windows eDrive
ReliabilityShield Error CorrectionFull end-to-end CRC Protection
Data ProtectionRAISE 1 +Fractional RAISE
Package401-ball FCBGA 11x18 mm
ComplianceRoHSHalogen FreeGreen

The Seagate SF3000 has been plagued with setbacks due to brain drain and acquisitions, but with specifications in hand, we expect developments to come quickly as third-party controller-less companies push to release products. The SF3500 client drive misses the PCIe 3.0 x4 mark, and appears to lack the same in-your-face power as Samsung's SM951 in 4-corner performance. But lying just beneath the surface is where SandForce products perform really well. Mixed workload performance is real workload performance, and even to this day, the previous generation SF2000 line is competitive with other SATA products.

It will be very interesting to test SF3500 under real world conditions, and for some, it may just be an eye opener.

Chris Ramseyer is a Contributing Editor for Tom's Hardware, covering Storage. Follow him on Twitter and Facebook.

Follow Tom's Hardware on Twitter, Facebook and Google+.

Chris Ramseyer
Chris Ramseyer is a Contributing Editor for Tom's Hardware US. He tests and reviews consumer storage.
  • tom10167
    I hope speed isn't the direction SSDs keep going. I don't think anyone is asking to transfer 1PB/s with a queue depth of 512. At this point all anyone's wanting for is lower cost and consistent reliability without fancy software.
    Reply
  • Alec Mowat
    I hope speed isn't the direction SSDs keep going. I don't think anyone is asking to transfer 1PB/s with a queue depth of 512. At this point all anyone's wanting for is lower cost and consistent reliability without fancy software.

    Performance increases ultimately equal lower prices on older technology. I'm looking forward to speed, personally.
    Reply
  • alidan
    I hope speed isn't the direction SSDs keep going. I don't think anyone is asking to transfer 1PB/s with a queue depth of 512. At this point all anyone's wanting for is lower cost and consistent reliability without fancy software.

    there is only so much that can be done to decrease the price, i personally welcome the speed/performance gains as the other guy said, it lowers the price on older tech... get 2 older drives and either raid 1 or 0 them and get fantastic performance (just the risk of losing everything if one fails.
    Reply
  • wtfxxxgp
    I hope speed isn't the direction SSDs keep going. I don't think anyone is asking to transfer 1PB/s with a queue depth of 512. At this point all anyone's wanting for is lower cost and consistent reliability without fancy software.

    That's a mighty serious risk though, isn't it?

    there is only so much that can be done to decrease the price, i personally welcome the speed/performance gains as the other guy said, it lowers the price on older tech... get 2 older drives and either raid 1 or 0 them and get fantastic performance (just the risk of losing everything if one fails.
    Reply
  • danwat1234
    Max capacity of 1TB, really? For a new controller still not out and it doesn't support capacities above the capacity of SSDs we have now? Somehow I think this is a typo.
    Reply
  • Eggz
    My initial thought is "meh," but it'll be good if this thing delivers decent performance at a much lower price than the other NVMe drives. I think most people will appreciate an increase in the minimum speed of SSDs, rather than the max. Hopeful that drives a shift in advertising away from sequential transfers and toward small random access at low queue depths.
    Reply
  • xchoice
    Max capacity of 1TB, really? For a new controller still not out and it doesn't support capacities above the capacity of SSDs we have now? Somehow I think this is a typo.

    Leaked data, may not be accurate!.....but who knows :P
    Reply
  • CRamseyer
    Unlike some of the leaked data you may find from other sources, my leaks come with double confirmation. The capacity information comes from Seagate who we sat down with earlier in the day.

    The maximum capacity may change when IMFT launches 256Gbit 3D NAND. SandForce has a long history of supporting flash released well after the controller hits the market. Also, the modular design of the SF3000 series means that a nip and glue upgrade could increase any specification without a full redesign. PCIe 3.0 support - chop, cut, rebuild. The same thing with the number of addressable flash channels. The core architecture is solid and in theory SandForce could drive future products from this base for many product generations.
    Reply
  • Pailin
    My initial thought is "meh," but it'll be good if this thing delivers decent performance at a much lower price than the other NVMe drives. I think most people will appreciate an increase in the minimum speed of SSDs, rather than the max. Hopeful that drives a shift in advertising away from sequential transfers and toward small random access at low queue depths.

    Totally. My Vertex 4 is still a pretty well performing SSD and I really notice load time on some bigger games.
    Star Citizen for example takes about 50 secs (counted in head, so give or take a few) to load from the launcher into my Asteroid Hangar with currently 10 ships in it.

    I would be Very Happy for my next SSD to have double or triple that performance
    personally :)



    You can almost never have too much performance, there have been Many embarrassing moments of those thinking they had all we could "Possibly" need !!!
    Reply
  • skipperkins
    My initial thought is "meh," but it'll be good if this thing delivers decent performance at a much lower price than the other NVMe drives. I think most people will appreciate an increase in the minimum speed of SSDs, rather than the max. Hopeful that drives a shift in advertising away from sequential transfers and toward small random access at low queue depths.

    Why is that better? For what applications is this a better reference?
    Reply