Samsung 9100 Pro 2 TB

[Selaya]

makes sense; i remember backblaze shucking consumer-grade hdds for their drive pods too; if you have the infrastructure & knowledge for dealing w/ routine hardware failures like that then yeah ...

they probably also stopped making those consumer-grade mlc drives precisely bc they realised alot of them ended up in the prosumer/enterprise sector anyways while being significantly cheaper then actual enterprise-grade products. oh well.

 

[EsaT]

1.56W consumption in idle is one of the worst values (and too much for me personally) and 0.2W in idle L1 ASPM isn't that good either, when other drives have 0.05W. In a mobile device this may make a difference. Is this due to SAMSUNG's process node, PCIe 5.0 or what could be the reasons?

That power draw for doing nothing is far from good power efficiency.
For most home users drive is going to spend almost all the time in practise doing nothing with only tiny tiny fraction of total time spent in burst loads.
So better R/W load efficiencies matter little for battery life.

With Samsung having in house made main components and control over them (instead of relying others) would have thought them to be able to have good idle power consumption.
Now they've been regressing in that since 980 Pro.

While PCIe 5.0 drives in general seem to suck power like "spinning rust" just for keeping powered up, at least Phison E31 drives have quite good power save.

 

[Chrispy_]

makes sense; i remember backblaze shucking consumer-grade hdds for their drive pods too; if you have the infrastructure & knowledge for dealing w/ routine hardware failures like that then yeah ...

they probably also stopped making those consumer-grade mlc drives precisely bc they realised alot of them ended up in the prosumer/enterprise sector anyways while being significantly cheaper then actual enterprise-grade products. oh well.

My friend working as a Google storage architect says they do the same thing - Designing systems that use cheap parts with a high failure rate and designing the overall platform to be failure-tolerant through redundancy and agility.

 

[unwind-protect]

And make sure not to use Arrow Lake for testing as that platform is terrible for testing SSDs. Screenshot below.

View attachment 391806

Is there a technical explanation for this?

Where is the chart from?

 

[angaroo]

Is there a technical explanation for this?

Where is the chart from?

The both charts come from
https://x.com/coolingreviews/status/1890444665654001742 and https://www.reddit.com/r/intel/comments/1d0mhbd
Also check this out https://www.reddit.com/r/hardware/comments/1grq7r2 and

Talks about it.

I've seen in few instances that it is indeed not just a "unlucky" run, it's consistent that SSDs just suck on Arrow Lake, just how it's performance in games, fully tuned 14900ks is beating out 285k in games. Looking at raw performance, not cost or power consumption - as argument for power consumption goes both ways, more power = more performance, less power = less performance.

But what i cared and care mostly is how much degraded loading times are when you are comparing it with top5 fastest ssds in the world in terms of latency and loading times. The performance penalty of using Optane on Arrow Lake is almost 50%.


As for if there is technical explanation for this, you will have to ask Intel engineers or do healthy assumption, some people say it's due io-die.

 

[R-T-B]

Both are consumer grade drives.

The P5800x is marketed as a workstation and DC part and is NOT a consumer drive, even if consumers can buy it.

I picture the reason TPU does not bench these drives is new ones cost a small fortune, and used would not really be fair.

What's the point when nobody can buy Optane drives?

You still CAN buy the 5800x new with warranty, if you feel like burning a few grand.

As for if there is technical explanation for this, you will have to ask Intel engineers or do healthy assumption, some people say it's due io-die.

Yep it's most likely the chiplet io-die. Ryzen chiplet based designs suffer a similar fate.

 

[bug]

The both charts come from
https://x.com/coolingreviews/status/1890444665654001742 and https://www.reddit.com/r/intel/comments/1d0mhbd
Also check this out https://www.reddit.com/r/hardware/comments/1grq7r2 and

Talks about it.
View attachment 392312
View attachment 392313


I've seen in few instances that it is indeed not just a "unlucky" run, it's consistent that SSDs just suck on Arrow Lake, just how it's performance in games, fully tuned 14900ks is beating out 285k in games. Looking at raw performance, not cost or power consumption - as argument for power consumption goes both ways, more power = more performance, less power = less performance.

But what i cared and care mostly is how much degraded loading times are when you are comparing it with top5 fastest ssds in the world in terms of latency and loading times. The performance penalty of using Optane on Arrow Lake is almost 50%.


As for if there is technical explanation for this, you will have to ask Intel engineers or do healthy assumption, some people say it's due io-die.

Intel has dropped Optane support from chipsets since the 700-series. So yeah, Optane tends to underperform on platforms that do not support it.

 

[Fluke]

@W1zzard
I'd love to see a review of the WD SN850X as it's currently dominating the TLC+DRAM SSD charts.
The 850Xs are all over the marketplace.

Best WD offer: 4TB SN850X 256€
Meanwhile, in Samsung Pricing Land: 4TB 9100PRO 462€

Thanks for turning that purchasing decision into a no-brainer, Samsung.

 

[terroralpha]

these results are so impressive that i now understand why it's impossible to find a crucial 2TB T705 anywhere.

 

[FatBoyNL]

The Thermal Analysis of this review is partially incorrect. There are indeed 2 different temp sensors on this NVMe SSD, BUT both are exposed as well - they are NOT the same!

But the problem is that Samsung decided to swap the 2 sensor readings around in their current firmware (against NVMe standards) and therefore mixed up things with the default "Temperature" reading of the SSD.

The result of that is that only the lowest sensor temp is shown in most (if not all) monitoring programs. So the same temp reading is shown, but in duplicate...
Only Samsung can fix this ultimately (by firmware), but AIDA64 will probably include a workaround for this issue in their next beta.

P.S. The other temp sensor (the 'real' #2) reads about 7 degrees higher on my system, so the Samsung SSD 9100 PRO is definitely 'not as cool' as Samsung is currently faking it in all hardware reviews.

EDIT: <attached screenshot>

 

[529th]

So is this an attempt to fool the public to gain reputation & sales, and this is what the CEO was "Do or Die" about?

 

[FatBoyNL]

So is this an attempt to fool the public to gain reputation & sales, and this is what the CEO was "Do or Die" about?

Actually, I think this is just a dumb-ass mistake of Samsung engineers. But possibly waaaaay up from there on.
Fake numbers sell though.
(EDIT: as they are quite convenient of course)

I hope they'll fix this asap. I'm not a happy customer right now, that's for sure

EDIT: This screenshot shows what's actually happening while just copying data between drives. Temps got up to 80 degrees Celsius...