Intel 750 Series 400GB Versus Samsung SM951 512GB

Why you can trust Tom's Hardware Our expert reviewers spend hours testing and comparing products and services so you can choose the best for you. Find out more about how we test.

Technical Specifications


MORE: Best SSDs For The Money
MORE: All Storage Articles
MORE: Storage in the Forums

Before we can compare the SSD 750 400GB to Samsung's SM951 512GB (AHCI), let's look at the two offerings from Intel. There is big capacity gap between the 400GB and 1.2TB models, and we're hoping Intel fills that space with an 800GB version. We think we've figured out why it isn't happening, though. The 1.2TB drive uses 18 channels and the 400GB model uses nine. Also, the 1.2TB model has a 4GB DRAM buffer, while the 400GB comes equipped with 2GB. An 800GB SSD 750 would likely break that perfect split. If an SSD doesn't have sufficient DRAM to buffer table data, then it suffers reduced performance. A theoretical 800GB model would most likely need a large 4GB buffer, which would have an adverse effect on cost. An in-between model in the SSD 750 family just might not fit Intel's pricing structure.

With half the number of channels from the controller to flash and half the DRAM, you might assume that the 400GB model's performance would suffer more than Intel's specifications suggest. But we were surprised to find that the company's numbers are spot-on; the nine-channel SSD 750 400GB only takes a moderate performance hit. Sequential read performance drops to 2200 MB/s, down from the 1.2TB model's 2400 MB/s. Sequential write performance drops a little further, sitting at 900 MB/s compared to 1200 MB/s.

The SSD 750 400GB also delivers high random IOPS throughput. It delivers a capacity class-leading 230,000 4KB read IOPS and 430,000 write IOPS. That performance is truly astonishing on paper. However, Intel's enterprise pedigree means you only get those numbers with lots of commands stacked up. Most users rarely achieve more than a queue depth of eight, so this is an area we'll have to explore in greater depth. 

Intel's 400GB SSD 750 does have an advantage in one area over the 1.2TB model (besides price): it requires one-third the airflow over its heat sink. This makes the product better for small form factor systems that don't ship with large intake fans. In our testing of both SSD 750 models, we didn't run into any thermal issues, and we didn't add additional cooling.

This product series is the first client-based SSD to use the NVMe protocol. We covered NVMe in detail in our SSD 750 1.2TB review. If you missed it, you can read more about NVMe here.

Chris Ramseyer
Chris Ramseyer is a Contributing Editor for Tom's Hardware US. He tests and reviews consumer storage.
  • BoredErica
    Hey, I would love to see a new article about trace-based analysis of hard drive load. I've got some money to splurge on storage I don't strictly NEED, but actually I have no idea what to buy. Just a thought. :)
    Reply
  • Arabian Knight
    agaiiiin

    How about testing the Intel 750 card in Raid 0 ?

    please add another Card and try software raid .
    Reply
  • PaulBags
    RE: Sequential Steady State, what is the percent scale in that graph? I assume one end is writes and one read, but which? Graph needs to be properly labeled.
    Reply
  • PaulBags
    Sequential steady state vs random steady state, switching between IOPs and MB/s, yay that's comparible. Bah, I'm done with this article. Bppppppt.
    Reply
  • jedimindtriks
    flexxmemory.co.uk has the sm951 for 350$
    Reply
  • Blueberries
    I'd take a 750 over the SM951 any day, you just can't beat the latency and random read/write performance. The SM951 is probably hands down the best SSD right now for the typical PC user, all things considered; having neck-and-neck or better performance at low queue depths and of course a much faster boot time.

    Get the 750 if you're an enthusiast and you can afford it. Otherwise the SM951 is going to be the best performance you've experienced in your life.
    Reply
  • Osiricat
    Hi! Anything about temps? I read in earliers reviews that SM951 warms up to 70-75ºC, for some reason intel added passive cooling over theirs memory chips!
    Reply
  • CRamseyer
    The industry standard is to measure random performance in IOPS and sequential performance in throughput. Why would you want to compare sequential IOPS to random IOPS or sequential throughput to random throughput?

    The sequential steady state shows read percentage. 100% read to 0% read. It's mainly an enterprise test I imported a few years ago in my testing to see the bathtub curve of the devices under test.

    As for RAID with these drives. I'm not sure if a RAID Report is really needed. You can't boot from devices in Windows software RAID. If 5% of the market cares about these premium parts to start with then RAID performance has to amount to 5 to 10% of those readers. I don't think there are enough readers to justify the time and expense for that level of testing. If Intel wants to provide the parts I don't mind testing and writing the article.

    Chris
    Reply
  • PaulBags
    15929620 said:
    The industry standard is to measure random performance in IOPS and sequential performance in throughput. Why would you want to compare sequential IOPS to random IOPS or sequential throughput to random throughput?

    ...

    Chris
    So that I can see the difference? How much of a performance drop is there with random vs sequential? Yes I don't get to decide how my data travels, but is random throughput significantly above the sata 6gb ceiling or does interface not _really_ matter yet unless you have a lot of sequential reads/writes to do? Additionally, what kind of performance is there when there is mixed sequential and random; somewhere in between the two or would comparing the two be completely irrelevant?
    Reply
  • CRamseyer
    Random performance is never higher than the limits of SATA 6Gbps unless you have a product like the Memblaze or P320H that can deliver full PCIe bandwidth random 4K reads and writes. Those are both enterprise products that cost more than a used Honda.

    Measuring 4K data in throughput is like telling someone the length of a dollar bill in miles.
    Reply