Questions about SSD

[anandtechreader]

Hello, I have some questions about SSD.

1. Besides the cost, within the same series, is a 2T memory module faster than 1T memory module?

2. In a high performance i9-7900X workstation, how are the following SSD ranked? Will I see noticeable differences? I am not a gamer but a researcher doing research that involve big data and GPU CUDA computations.

a) Samsung 860 EVO M.2 1TB SSD (MZ0N6E1T0BW)
b) Samsung 960 PRO Series - 1TB PCIe NVMe - M.2 Internal SSD (MZ-V6P1T9BW)
c) Samsung 960 EVO Series - 1TB PCIe NVMe - M.2 Internal SSD (MZ-V6E1T0BW)

3. The connector of the above three SSD look different, can they be plugged in the same slot? Motherboard manufacturer mentioned: "The numbers 2242/2260/2280/22110 denote the size of the Hard Drives that can be used. M.2_1 goes up to 22110 while M.2_2 does not." Don't know how these numbers related to the above three Samsung's SSDs.

4. Can I turn a SSD/NVMe - M.2 SSD into RAM? Will it be faster than DDR4-3200C14 or DDR4-3600C17? Can I do it in Linux?

Thanks.

 

[Yuriman]

2) I would advise reading Anandtech's review on those drives, and draw your own conclusions:

https://www.anandtech.com/show/10833/the-samsung-960-evo-1tb-review

3) I'm not terribly familiar with the variations of the M.2 slot, but there is backwards compatibility. Any modern board should accept any of the three.

4) No, you can't turn them into RAM, and no, they're not faster than RAM. Any of these drives will be far slower than RAM.

 

[IntelUser2000]

The numbers 2242/2260/2280/22110

Here's a good guide: https://www.ramcity.com.au/blog/m.2-ssd-guide-compatibility-list

Now, M.2 SATA and M.2 NVMe may not be compatible, even though it may be physically. You have to check with the board. The 860 EVO is SATA and 960 Pro and EVO is NVMe. NVMe is faster than SATA.

4. Can I turn a SSD/NVMe - M.2 SSD into RAM? Will it be faster than DDR4-3200C14 or DDR4-3600C17? Can I do it in Linux?

SSDs are about a 1000 times slower in latency. Your OS would be absolutely unusable if it was running completely on an SSD.

 

[dlerious]

If you're research project involves a lot of I/O, you might be taking a big hit on Intel with the Spectre/Meltdown patches.

 

[Kartajan]

Various types of M.2 devices are denoted using the "WWLL-HH-K-K" or "WWLL-HH-K" naming schemes, in which "WW" and "LL" specify the module width and length in millimeters, respectively.
The "key" states the class of module, with B or M used for SSD's

As far as ranking the SSD's you list, in order of IOPS best to worst:

b) Samsung 960 PRO Series - 1TB PCIe NVMe - M.2 Internal SSD (MZ-V6P1T9BW) M.2 2280 <PCIE 3.0 x4>
c) Samsung 960 EVO Series - 1TB PCIe NVMe - M.2 Internal SSD (MZ-V6E1T0BW) M.2 2280 <PCIE 3.0 x4>
a) Samsung 860 EVO M.2 1TB SSD (MZ-N6E1T0BW) M.2 2280 <SATA3>

the 2280 means they are all the "gum stick" form factor, 22m wide and 80mm long...

You really need to confirm if the motherboard supports both PCIE and SATA in their m.2 slot. Newer boards tend to support both, older boards typically only support the sata3 only.

see: http://www.te.com/commerce/Document...m=1-1773702-1NGFFQRG-EN&DocType=DS&DocLang=EN

 

[anandtechreader]

If you're research project involves a lot of I/O, you might be taking a big hit on Intel with the Spectre/Meltdown patches.

Do you have more info on this?

 

[Kartajan]

a lot depends on specific workloads, but here is one focused on VM's
https://www.altaro.com/hyper-v/meltdown-spectre-hyperv-performance/

 

[Kartajan]

Longer story on meltdown/ spectre:
https://arstechnica.com/gadgets/201...e-and-meltdown-patches-will-hurt-performance/

 

[anandtechreader]

Do you recommend going ahead to build a very expensive i9-7900X system with several GPUs now or just get a temporarily i7-8700K with only 1GPU until PCIe 4.0 components come and those meltdown/spectre issues are removed at hardware level?

 

[Kartajan]

as far as GPU's go, I recommend holding off until prices normalize. (SWAG of 6-12 months).
I would not be concerned with PCIe 4.0 or meltdown/ spectre impact on the 8th Gen parts.
Earlier gen parts appear to be impacted more with the m/s patches than 8th gen.
New PCIe spec looks to be of low impact at the current state of GPU progression.

as far as i9-7900X vs 8700K-
7900X is better for multi-threads, 8700K for single.
And if your workload loves multicore- did you check out AMD Threadripper? (Cheaper than 7900X)

Thus, buy what you <need> when you need it, but save the <wants> for better pricing conditions...

 

[anandtechreader]

Thanks. I considered the Threadripper earlier but decided not to choose it because of possible compatibility issues. Aren't threadripper also affected as well? Should I give it a second through?

As far as I know, GPU's performance is much more important than CPU so I guess 8700K is OK if I only use 1 GPU. 8700K can only support 1 PCIe 3.0x16 anyway.

Due to the cost of the GPU, I will just get a inexpensive mid-range one until I need the power later. So, I will use only 1GPU regardless of going for the 7900X or 8700K.

In any case, I need a decent machine for work now. Does not have to be highest end until at least half a year later. Spending that much money on a system with performance hit may not be a good idea.

 

[dlerious]

Anandtech did a recent article looking at an Intel NUC (Kaby Lake). GPU testing was ok, but SSD testing took a good hit.
https://www.anandtech.com/show/12566/analyzing-meltdown-spectre-perf-impact-on-intel-nuc7i7bnh

 

[anandtechreader]

Thanks. Am I correct that storage's performance is dropped by about 30% while GPU performance is dropped by about 1.5%? Am I reading it correctly?

So, if I use GPU to do CUDA computations most of the time and if at most 1.5% performance loss is acceptable, then it is OK to go ahead with the 7900X system. However, if I do a lot of moving data between the SSD and the CPU/GPU, then better not to spend all these money on a problematic system and just get the less expensive 8700K for temporary use. Am I correct?

Do you think the performance loss due to these issues are similar between the 8700K and 9700X? Some people suggested going for the Threadripper. Isn't it affected as well?!

 

[Kartajan]

Storage performance takes a hit due to CPU losses. On the mobile chips (as tested in the anandtech article) that did result in a large hit on NVMe and SSD.
The Desktop chips take a hit as well, but it is mitigated somewhat by the larger core count and clockspeed.
Threadripper (if my memory is serving me) should have similar effects as desktop 8th Gen (present but mitigated by cores & clocks).

As far as which way to buy, after considering future upgrade-ability for GPUs, "which CPU best fits my workload", then finally on which SSD best fits my workload, and if the performance hit from Spectre/Meltdown warrants the next higher tier of SSD to compensate for the losses....

If CPU load is not a major impactor, you may even want to consider the 2000 series Ryzen chips as well...

 

[Kartajan]

keep in mind that the s/m patches can be skipped if you are in the right environment (For business purposes, an air-gapped machine with no potential physical access by bad actors)

 

[Kartajan]

the tl,dr from the altaro article:
What we learn from these charts is that the full range of updates causes CPU performance degradation for some functions but leaves others unaffected. Normal storage I/O does not suffer as much. I imagine that the higher impacts that you see in the Storages Spaces Direct benchmarks shared by Ben Thomas are because of their extremely high speed. Since most of us don’t have storage subsystems that can challenge a normal CPU, we aren’t affected as strongly.

Overall, I think that the results show what could be reasonably inferred without benchmarks: high utilization systems will see a greater impact. As far as the exact impact on your systems, that appears to largely depend on what CPU functions they require most. If your pre-patch conditions were not demanding high performance, I suspect that you’ll suffer very little. If you have higher demand workloads, I’d recommend that you try a test environment if possible. If not, see if your software vendors have any knowledge or strategies for keeping their applications running acceptably.

 

[anandtechreader]

The workstation is for home use. In this case, can I skip the s/m patches?

 

[Yuriman]

I imagine Spring Creator's update will have these rolled into it, so good luck with that. Microsoft very aggressively pushes updates now, so it will become increasingly hard to avoid as time goes on.

 

[anandtechreader]

Jut received a message from a colleague. He mentioned that I do have to worry about such vulnerabilities. The slowdown could be negligible or as much as 20%. We use Ubuntu Linux, not Windows. What suggestions do you have? Stop the plan to spend a lot of money to build a supposed to be high performance system and wait until Intel/AMD fixed the bugs at hardware level (will that be by the end of this year?), build a lower cost system using 8700K or just go ahead to order components for the 7900X system?

 

[dlerious]

AMD is only vulnerable to one of the exploits (forget which right now). Performance isn't affected as much as Intel. Hardware level fixes will require new chips - not sure when they'll be released. I don't have any experience with Intel HEDT platforms, so can't make a recommendation there. For Linux, maybe have a look around Phoronix
https://www.phoronix.com/scan.php?page=home

 

[Kartajan]

The big question is still whether or not your workload prefers cores or clocks.
If your workload is more "single fast core" oriented- I would get the fastest single core setup w/ a single GPU and the option to add more GPU later.
If your workload is more "large thread count" oriented- I would go with threadripper w/ a single GPU and the option to add more GPU later.

I still maintain that the performance drop from the m/s bug software fixes are somewhat mitigated by having a stronger/faster CPU than required, so I would not hold off on CPU just because a fix in silicon is not available yet.

 

[anandtechreader]

As it is more like GPU computing, having 6-8 cores in the CPU would would be sufficient. No need to have more than that.

About those m/s bug software fixes, aren't then included in system OS updates? In other words, when I update the OS, those fixes are automatically applied then the system slows down.

My current computer is 10 years old. Cannot even do CUDA computing. As I type, there is a long pause in between each series of keystrokes.

 

[Kartajan]

re: m/s fixes; some patching is done in OS, but some requires BIOS update.
re: GPU compute; assuming a focus on CUDA encode/decode I looked up someone elses builds:
1) i7 Build (Intended as a Houdini FX Build): https://pcpartpicker.com/b/TVFtt6
2) threadripper build (Intended as a "machine learning" build): https://pcpartpicker.com/b/Y4P323

For either, I would only get 1 GPU until the prices improve.
and ask for advice in a new thread for whichever build strikes your fancy...

 

[anandtechreader]

Yes, GPU price is an issue. Since I just get 1 GPU now, I don' benefit from having a high-end CPU-MB that have many lanes.

In case of going for the temporary 8700K route, should I just get 32GB dual channel RAM and buy a new set of faster quad channel RAM later or 64GB quad channel faster RAM for future proof? I know 8700K is a dual channel CPU.

 

[Kartajan]

I say pick one of the builds I linked, but only get 1 GPU.
And if 8700k, just get the appropriate dual channel kit and stay there...