AMD Ryzen Threadripper NVMe RAID 0 Demonstrated

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
For those who missed it, Roman Hartung (der8auer) has re-uploaded his video from last week where he puts Threadripper’s many PCIe lanes to task with an NVMe RAID 0 setup: the test, which incorporated multiple adapter cards and 960 PRO drives, resulted in a pretty crazy benchmark result (27GB/s I/Os per second). Thanks cageymaru.
 
I wanna see an official "how much faster does it load different games/programs than a standard HDD or SSD."

I know the answer, but I want to see it included in future real world use benchmarks. You know, cause while we like big numbers we also like to know if it actually does something for us besides make our e-peens hard.
 
I wanna see an official "how much faster does it load different games/programs than a standard HDD or SSD."

I know the answer, but I want to see it included in future real world use benchmarks. You know, cause while we like big numbers we also like to know if it actually does something for us besides make our e-peens hard.

This is not for gamers. This is for people doing research, 8k video editing work and so on.
 
This is nuts. Too bad the disks are not $100 yet. Kind of makes me think intel optane will still be more expensive than this in raid 3 or 6!!!
 
Honestly, call me odd but if I had the type of $$$ to put down on the card & drives I'd like to have raid 5+0 support across multiple cards.
 
I wanna see an official "how much faster does it load different games/programs than a standard HDD or SSD."

I know the answer, but I want to see it included in future real world use benchmarks. You know, cause while we like big numbers we also like to know if it actually does something for us besides make our e-peens hard.

I know for a fact that SSD's definitely load games faster than HDD's. When you go from loading Dragon Age: Inquisition from 2 minutes down too 20 seconds, it makes a big difference. :) (No, the hard drive was not going bad.)
 
This is pretty cool, but haven't we reached the point of severely diminishing returns at this point?

Go from any hard drive to almost any SSD, and you'll see a HUGE improvement in boot times, program and game level load times, and overall system responsiveness.

After that, any incremental SSD improvement is much less of a big deal.

When I went from my SATA Samsung 850 Pro to my PCIe Intel 750 drive, sure it benched 4 times faster in sequential reads and writes, but in practical real world use scenarios it was a huge meh.

Using RAID0 to put a bunch of M.2 drives together will certainly boost sequential read times, but will actually reduce seek times, which is where most of the every day benefit from SSD's come from.

Outside of a few corner cases where people need extreme sequential read and write speeds, I don't see this being much of a big deal.

Well, unless you start thinking of it from a redundancy perspective. Mirroring a couple of M.2 drives to increase reliability and failure redundancy would be pretty badass.
 
This is not for gamers. This is for people doing research, 8k video editing work and so on.


Ummm ...... really?

I'm a gamer, I play a lot of games. I like to have a lot of games installed. I haven't used a mechanical hard drive as my boot drive since 2009.

My games can't have the benefit of speed or mass storage because ..... you said so or think otherwise?

Gamers often edit video, edit photo's, search for files, etc.

PC enthuists usually do more with their PC than just game.

You're comment is not very well thought out.
 
Ummm ...... really?

I'm a gamer, I play a lot of games. I like to have a lot of games installed. I haven't used a mechanical hard drive as my boot drive since 2009.

My games can't have the benefit of speed or mass storage because ..... you said so or think otherwise?

Gamers often edit video, edit photo's, search for files, etc.

PC enthuists usually do more with their PC than just game.

You're comment is not very well thought out.


I think his comment means that raided m.2 drives won't see much benefit - if any - in typical system responsiveness and game level loading times.

And he is likely right.

We hit diminishing returns here some time ago.
 
Ummm ...... really?

I'm a gamer, I play a lot of games. I like to have a lot of games installed. I haven't used a mechanical hard drive as my boot drive since 2009.

My games can't have the benefit of speed or mass storage because ..... you said so or think otherwise?

Gamers often edit video, edit photo's, search for files, etc.

PC enthuists usually do more with their PC than just game.

You're comment is not very well thought out.

He's not saying SSD's are not for gamers. He (and I will concur) was saying that RAIDed SSDs are not for gamers.

This is like speaker quality vs price. There's a point at which cost/performance is nearly 0.

Going from a HDD to a SSD is a great performance leap for the cost. Going from one SSD to RAID 0/10/etc SSD will be very little to latterly no performance improvement for the cost.
 
Unless you have workloads with large queue depths (aka not games) you will just be wasting your money doing this. Now if your workloads DO have large queue depths or benefit from very high I/O, this will more than pay for itself in time saved.
 
He's not saying SSD's are not for gamers. He (and I will concur) was saying that RAIDed SSDs are not for gamers.

This is like speaker quality vs price. There's a point at which cost/performance is nearly 0.

Going from a HDD to a SSD is a great performance leap for the cost. Going from one SSD to RAID 0/10/etc SSD will be very little to latterly no performance improvement for the cost.
And that's why I want to see that shown in reviews of these fast, new drives and the new systems and possible RAID configs being touted.

The vast majority of reviews ignore this or gloss over it - to the detriment of the majority of gamers.
 
And that's why I want to see that shown in reviews of these fast, new drives and the new systems and possible RAID configs being touted.

The vast majority of reviews ignore this or gloss over it - to the detriment of the majority of gamers.

Not sure how that's a detriment to gamers... it's not hard to know that loading a level - even if it's at the extreme of a few hundred megabytes - will not have any noticeable difference between a single SSD and RAIDed SSDs. The single SSD will already be able to send all the needed data because the response time is very low and the bandwidth is very high. A second drive will not increase either of those in this case because the limiting factor - the bandwidth - isn't capped in the single SSD case.

If you were loading game levels that were a few gigs in size, then you'd see a slight difference (at a point the GPU, memory, or cpu will become bottlenecks).


In my mind, asking for a review of RAIDed SSDs in gaming is like asking for a review of an nVidia 1080ti for playing solitare where the screen is capped at 60 fps.
 
Last edited:
And that's why I want to see that shown in reviews of these fast, new drives and the new systems and possible RAID configs being touted.

The vast majority of reviews ignore this or gloss over it - to the detriment of the majority of gamers.


Agreed.

Most reviews of SSD's just include benchmark data without a commentary on real world outcomes.

Most fast new NVMe drives have very limited real world benefits in desktop workloads where queue depths are relatively low.

Using raid to speed them up may have some benefits where an increase in sequential speeds outweighs a slight loss in seek times, but likely there will not be an appreciable difference.

For client workloads it will be more about bragging rights than anything else.
 
So one benchmark in inconsistent for only this brand new storage setup and the other isn't, so it's the benchmark fault.....yeah I smell BS. I'm sure at some point AMD will get the bugs worked out. Today is not that day.
 
While this is fun

Could anyone tell me why especially gamers are always so loud about needing more lanes?

Especially a lot?
Especially after they bought into threadripper or skylake-x (which is especially fun because of its artificial restrictions)

I have yet to receive an answer from anyone saying they need those lanes for gaming


Still fun to see


Hmm
Now onto spending 10 grand on a diskstation that can handle that kind of data transfer
Cat videos n stuff ya know :D
 
And that's why I want to see that shown in reviews of these fast, new drives and the new systems and possible RAID configs being touted.

The vast majority of reviews ignore this or gloss over it - to the detriment of the majority of gamers.

There already exists many youtube videos and reviews where the newer fast NVMe drives are compared to their modern SATA SSD counterparts. While they are like 3-5x faster in raw benchmarks, real world load times are within 5-10% in the more significant tests. These are simply raided NVMe drives. You're talking about quintupling your costs or more for 5-10% load time decrease in a BEST case scenario, many tests they were nearly equal. As a comparison the SATA HDD vs SATA SSD were more like 50-80% faster which IS a worthwhile upgrade for gamers. YMMV, burn your money at will, I sure won't stop you, it will functionally work fine.
 
I wanna see an official "how much faster does it load different games/programs than a standard HDD or SSD."

I know the answer, but I want to see it included in future real world use benchmarks. You know, cause while we like big numbers we also like to know if it actually does something for us besides make our e-peens hard.

Kinda like asking how much you SQL database is gonna run faster on a GPU. There ARE other heavy loads than games for a computer to run.
 
I went from SSD to NVME and the difference is completely worth while to me. Perhaps I'm the odd man out.
 
There already exists many youtube videos and reviews where the newer fast NVMe drives are compared to their modern SATA SSD counterparts. While they are like 3-5x faster in raw benchmarks, real world load times are within 5-10% in the more significant tests. These are simply raided NVMe drives. You're talking about quintupling your costs or more for 5-10% load time decrease in a BEST case scenario, many tests they were nearly equal. As a comparison the SATA HDD vs SATA SSD were more like 50-80% faster which IS a worthwhile upgrade for gamers. YMMV, burn your money at will, I sure won't stop you, it will functionally work fine.

Oh I've done my own testing and posted the results over on the SSDs and Data Storage sub forum - I pointed this out because reviewers rarely hit that point, just glorifying the massive sequential read, write, and IOPS numbers. Kinda like the dude in the video with his impressively high results.
 
I went from SSD to NVME and the difference is completely worth while to me. Perhaps I'm the odd man out.

It would appear so.

I saw hardly any difference day to day in going from 550MBps to 3500MBps. Machine booted roughly the same time, Fallout 4 loads up the same, Firefox etc. I guess I don't perceive time or have the reaction times of a housefly. :D


Like I said before there are diminishing returns for 80% of current computing with data rates. The benefits are largely for the benchmarks.

This is a niche hardware application and those that can really use it will get a good boost. Gamers...not really.

Get the best single NVME SSD you can buy and stick with that. Spend the hard earned elsewhere.
 
Last edited:
+1 on not a whole lot of difference, game-wise. I went from an msata to a 960pro for boot and games and while the system does run faster, I wouldn't be screaming from the rooftops over it.

Move files around or buy a new game on Steam at it's REALLY fast, but for everyday stuff? Not so much.
 
I went from SSD to NVME and the difference is completely worth while to me. Perhaps I'm the odd man out.

You are definitely not the odd man out, my boot drive is now NVME and it makes a very large difference. That said, I saw a larger difference going from 2 hard drives to 2 x 1TB Sata SSD's late last year for my games and they load considerably faster. :)
 
You know, cause while we like big numbers we also like to know if it actually does something for us besides make our e-peens hard.

This is not for gamers. This is for people doing research, 8k video editing work and so on.

This is pretty cool, but haven't we reached the point of severely diminishing returns at this point?

Unless you have workloads with large queue depths (aka not games) you will just be wasting your money doing this.

[H]ard|OCP goes {S}oft...what has become of this community?

It used to be about fast for the sake of being fast, damn the price, damn the "diminishing returns".
 
[H]ard|OCP goes {S}oft...what has become of this community?

It used to be about fast for the sake of being fast, damn the price, damn the "diminishing returns".

Yeah but this isn't 2007 any more. Most of us here don't even upgrade our CPUs every 6 months or 6 years in some cases.

Folks just don't have the cash or the need so much. Those of us that are older now have family, divorces or lower wages to contend with. Those younger just don't have the jobs or need to save up for a home. We are now content to let YouTube tech sites that get given the gear masturbate over it instead. Fair enough I say.

The main fact is probably a lot of us have experienced the somewhat less than staggering real world leap from SATA to NVME and it's tempered our enthusiasm.

Tech is mainstream, most of it now is 'good enough' and the performance ratio for the cost no object stuff just isn't all that now.
 
[H]ard|OCP goes {S}oft...what has become of this community?

It used to be about fast for the sake of being fast, damn the price, damn the "diminishing returns".

Shhhhh! "We" know what's best for you. If it's good enough for me then you don't need it and shouldn't buy it.
Take your enthusiasm for the latest and greatest tech somewhere else.

/sarcasm off
 
[H]ard|OCP goes {S}oft...what has become of this community?

It used to be about fast for the sake of being fast, damn the price, damn the "diminishing returns".


I don't think the [H] was ever about wasting money on expensive dyno-queen technology.

It was always about getting the best real world experience for the money. That's why Kyle was the first to ditch the canned benchmarks and focus on the real in-game experience rather than crowning a winner over Q3A benchmarks of 315fps vs 300fps, in an era where the CRT monitors we were using rarely worked with a refresh rate of above 100hz.

It's the tech equivalent of focusing on the quarter mile figures rather than looking at peak whp on the dyno.

I don't have any data to go by right now, but everything I know from past SSD experience is that multiple raided NVME SSD's will do pretty much jack shit for the things that matter to most of us, system responsiveness, OS boot times, and game/map load times, as compared to a single good NVMe drive. Save the money and spend it on a Titan instead. This will be a much bigger bang for the buck.
 
This is pretty cool, but haven't we reached the point of severely diminishing returns at this point?

Go from any hard drive to almost any SSD, and you'll see a HUGE improvement in boot times, program and game level load times, and overall system responsiveness.

After that, any incremental SSD improvement is much less of a big deal.

When I went from my SATA Samsung 850 Pro to my PCIe Intel 750 drive, sure it benched 4 times faster in sequential reads and writes, but in practical real world use scenarios it was a huge meh.

Using RAID0 to put a bunch of M.2 drives together will certainly boost sequential read times, but will actually reduce seek times, which is where most of the every day benefit from SSD's come from.

Outside of a few corner cases where people need extreme sequential read and write speeds, I don't see this being much of a big deal.

Well, unless you start thinking of it from a redundancy perspective. Mirroring a couple of M.2 drives to increase reliability and failure redundancy would be pretty badass.

Try a compile benchmark of chrome or mozilla and get back to me. :)
 
You haven't fully read through this thread have you?

Benchmarks mean jack when it comes to real world experience.

I did read through it. Chrome and mozilla compile times ARE real world test and would be reflective of large compiles.
 
I think that kind of I/O, while great to have, exposes other bottlenecks. Is it necessary for most of us? Likely not. But the idea of the capability opens the doors for more innovative applications in the future. Today's top end concept is tomorrow's mainstream
 
So just when would 99.999999% of the computing population ever need to do that?;)

Seriously? I can't tell if you are just egging me on, or being serious.

Look if Chrome/Mozilla/open source (7zip) compiles benefit 50% from the throughput increase (which is what most compiles are limited by) then chances are equally large projects would benefit on a similar scale.
 
Seriously? I can't tell if you are just egging me on, or being serious.

Look if Chrome/Mozilla/open source (7zip) compiles benefit 50% from the throughput increase (which is what most compiles are limited by) then chances are equally large projects would benefit on a similar scale.

Yeah so largely niche as I mentioned a while back. What most of us are looking for nowadays is tangible leaps in performance that are readily noticeable during pretty much the entire computing experience.

If we spend $XXXX on a upgrade, it had better deliver the "wow factor" in more than just benchmarks.
 
Seriously? I can't tell if you are just egging me on, or being serious.

Look if Chrome/Mozilla/open source (7zip) compiles benefit 50% from the throughput increase (which is what most compiles are limited by) then chances are equally large projects would benefit on a similar scale.


Most people, even in an enthusiast forum like our own have never even once run a compiler on their machines.

Software compilation is a corner case used by an extreme minority of even computer hardware enthusiasts.
 
Back
Top