Contact Start/Stop – its safe vs. unsafe ending

[LongRunner]

The latter contact start/stop drives from Seagate – and to a (slightly) lesser extent, Samsung – are infamous for premature head crashes, far short of their rated (and industry‑standard) 50,000 start/stop cycles. Here I have pieced together the available evidence to help draw the line, in the correct place:

• Western Digital (always the most-honorable manufacturer besides Seagate in their heyday) last used CSS in the 4×125GB Zeus (WD5000KS/WD5000YS), moving to ramps from Tornado (the first WD5000AAKS) onwards; Tornado was also WD's last LMR series (accompanied by the PMR Jupiter).
• Even Maxtor – despite deliberately falsifying their start/stop counts on most DM+9/ML+II and earlier DM10/MLIII (and all DM11/MLPro AFAIK) production – ultimately adopted ramps in their final native DiamondMax 17 (even though it remained cheaply-made, and still crashed some of the time)…
• Unused model numbers in Seagate's Desk Reference (notably ST3500632AS – this is also why the 300GB 7200.9's model numbers imply 2 platters, instead of the 3 they released with) suggest that Seagate originally wanted to make the Barracuda 7200.9 as an uncompromised 3×167GB platform (continuing from their extraordinary 3×133GB 7200.8) to compete directly with Hitachi's Deskstar T7K500, and although not explicitly mentioned, the engineers probably would have adopted ramps if they couldn't pass the specified 50,000 contact start/stops (after all Seagate already used ramps in the Momentus 5400.2 and 7200.1). Under competitive pressure from Maxtor's impending DiamondCrash 11/MaXCrime Con, however, Seagate begrudgingly rushed out the ST3500641AS – basically a (S)ATA version of Blizzard (the NL35 FC, or Barracuda 500LPFC as it would be known under Seagate's earlier honest naming) – so the ramps could wait a bit longer. (Very similar to Seagate's early‑1996 adaptation of the Hawk 2XL into the ST32140A – in fact both used the same type of bistable magneto‑mechanical parking latch, albeit in a different shape.) Enthusiasts already knew Hitachi had the fastest high‑end models (despite their mediocre corner‑cut single‑platter drives), so Seagate preferred to make a reliable DiamondCrash/MaXCrime competitor while it mattered.
  (The 1×160GB 7200.9s could get away with CSS, given their shorter spin-up/down times and hence less wear per-cycle.)
• Fortunately Seagate had perpendicular recording just around the corner, enabling them to raise their flying height and safely retain CSS one generation after their competitors did – the Barracuda 7200.10/ES.1 soon beating the T7K500 to a whopping ≈80MB/s (up from the 7200.8's already-outstanding ≈70MB/s).
  In Superhawk (ST3250310AS & ST3250410AS), the last ST‑10 model, they could even push it to 1×250GB.
  But then the vicious Maxtor managers brought it all down; Moose (the initial 250GB/platter 7200.11/ES.2) still wasn't too bad after updating its firmware (as long as you don't exceed one on/off cycle per day), but its consumer‑only descendants (Garbo, Brinks, Pharaoh & Hepburn) were very substandard indeed. This is why Apple instead ordered 1TB (ST31000521AS) & 500GB (ST3500511AS) Barracuda XTs with proper ramps (Muskie platform, same as the Constellation ES.1), having already learned the hard way from Maxtor…

Judging by Seagate's and WD's examples, the last safe Samsung HDDs would be the P120 (2×125GB), S166 (1×167GB LMR) and S250 (1×250GB PMR) – T133 a known dud (even by Red Hill's admission) and T166 (3×167GB LMR) is past the line too – so it's funny how so many disgruntled IT professionals jumped ship from Seagate to Samsung during the 7200.11 fallout, yet each OK Samsung model still had a solid Seagate equivalent too and the bad Seagates (developed post-merger) had only marginally‑better Samsung equivalents, leaving Western Digital as the last standing honorable manufacturer with decent models in all capacities (which was Seagate's greatest strength in the previous Barracudas).

If anything, even the current 1×1TB Pharaoh Oasis/Hepburn Oasis drives may still be “better” than Pharaoh/Hepburn proper (2×500GB & 4×500GB respectively) were in 2009 – at least on those I've heard, their start/stop sounds again fairly-normal rather than deliberately-rougher.
They're still quite low-end series and not built‑to‑last like the classic Barracudas, but they may at least fail peacefully (head/media degradation) first…

But you can still run the affected drives 24×7 in a non-critical application; just don't stop them (and I would retire Pharaohs once they reach even 1500 cycles).

[LongRunner]

Just a quick update – listening to this sound recording of the HD501LJ, it indeed starts/stops much too roughly for comfort (not as horribly as the DiamondCrashes, but still quite bad ) and it can only have gotten worse from there. This Poseidon doesn't sound so good either.
(Remember that the Barracuda 7200.8 introduced TMR heads, while the Deskstar T7K250 used ordinary GMR but with ramps; Poseidon had neither benefit.)
From the drives I have to hand, the HD250HJs (S250) start/stop reasonably OK, but even the HD082GJ (S166) sounds rougher-than-par given its single head…

For comparison, my 7200.10s start/stop just as smoothly as the 7200.7s or earlier.

This indeed seems to track with percentiles in StorageReview's survey (long defunct, but you can find it in the Wayback Machine):

• Seagate Barracudas, of course, were on top – 90th percentile for the ATA IV, 96th for the ATA V (despite mistakenly duplicating the ST320011A and ST340016A instead of including the correct ST330013A and ST340017A) and 88th for the 7200.7 (which would probably also achieve 90+ if they all used Agere read/write channels, rather than STMicroelectronics). (The ATA III got 92nd, but BB versions are occasionally head-crashed by NRRO…)
  The U5 also achieved 90th percentile (by moving from the U4/U8/U10's inferior STMicroelectronics R/W channel to a reliable NEC one), but U6 regressed somewhat to 75th percentile (since Seagate cut back on the media protection under ever-harder price pressure).
  Barracuda 7200.8 slipped back somewhat to 49th percentile (although probably due to a particularly-bad batch of SMOOTH chips more than anything else), and the 7200.9 to 43rd (the single-platter 160GBs do seem somewhat weaker, though still better than their Hitachi or Samsung competitors).
• Maxtor's MaXCrime Minus II (the fraudulently-upsold DiamondCrash Minus 9 flagship) landed at a woeful 7th percentile – below 9th and 10th for IBM's Deathstar 75GXP and 60GXP. Before that, their 2000-model DiamondMax Plus 40 was reliable (for once even better than the Barracuda ATA II).
  Multi-platter 180GXP onward were quite solid (IBM's flawed head connector notwithstanding), although the single-platters (and P7K500) cut their corners.
  The 7K400 (33rd) and 7K500 (22nd) also scored below-average, perhaps since their heavy spindle load made them more-vulnerable to shock seizure.
• Quantum did great until the Fireball Plus AS, which dropped to 43rd percentile (from the Plus LM's stellar 98th).
• Samsung's earlier models (V20400, P20, V30, P40 & P80) were moderately OK, but the Poseidon (P120) regressed to 34th percentile (corroborated by its audibly rougher start/stop) and although the T133/T166 had insufficient data, we can be sure they were worse again.

Of course their data was biased somewhat towards the short-term, so I'd still trust the 7200.8 over U6 any time. With the possible exception of RAM manufacturing, Samsung never did much of value, other than making slightly cheaper products (subtly enough that you don't notice until it's too late) for the naïve…