AMD raid kicking Intels behind
Printed From: ASRock.com
Category: Media&User's Review
Forum Name: AMD Motherboards
Forum Description: ASRock AMD Motherboards
URL: https://forum.asrock.com/forum_posts.asp?TID=6419
Printed Date: 22 Dec 2024 at 11:58am
Software Version: Web Wiz Forums 12.04 - http://www.webwizforums.com
Topic: AMD raid kicking Intels behind
Posted By: DooRules
Subject: AMD raid kicking Intels behind
Date Posted: 25 Oct 2017 at 10:30pm
|
" rel="nofollow - Have you seen this parsec? Wholly crap. You probably have but that is some kind of impressive. 27 Gbs read speed. Hope all is well with you and family by the way.
https://www.youtube.com/watch?time_continue=627&v=9CoAyjzJWfw |
Replies:
Posted By: parsec
Date Posted: 26 Oct 2017 at 10:38am
Hey DooRules, how are you doing? Great to hear from you! I'm great and the rest of my clan is just fine. I hope you and yours are the same! Yes, AMD has Intel beaten big time with their X399 NVMe RAID. AMD's use of the processors's PCIe 3.0 lanes for the SSDs makes a big difference. Intel's DMI3 interface in the chipset should be equivalent to PCIe 3.0, but it seems to falter with multiple SSDs in RAID 0. Sure there is some increase in performance in RAID 0, but it seems as if each M.2 slot does not have it's own, full DMI3 x4 bandwidth with multiple SSDs at the same time. The specs for the chipset's DMI3 lanes appear to show each M.2 slot has its own four DMI3 lanes, as an SSD in a PCIe 3.0 slot would have four PCIe 3.0 lanes. But something apparently causes a bottleneck, or the specs are not what they appear to be. I don't think that is a driver issue, since there have been many updates since the first IRST 14.x version gave us NVMe RAID. An interesting experiment would be to have two or three NVMe SSDs not in RAID 0, in the M.2 slots connected to the Intel chipset. Then run a benchmark on all of the SSDs at the same time and check the results. If each benchmark showed reduced performance compared to a single drive, single benchmark test, that would reveal in general how it happens, but not why. We can play games with CrystalDiskMark, setting multiple threads for the Sequential high queue depth test to get a better result out of Intel's RAID, which I wanted to believe was real, but we'll see what the X399 results are soon. All is not fun and games in AMD's NVMe RAID, a lot of users are having problems with it, simply creating the RAID 0 arrays. AMD's RAID is different and not as easy to work with as Intel's, and you really need some experience with it before using the NVMe RAID. Plus AMD's RAID documentation is poor, their "All in One" single download for multiple drivers style is carried over to their RAID manual. Every chipset's instructions in one manual, and there are differences between them. I wish I had an X399 board to try this out myself, it's driving me crazy not being able to figure out the details of the NVMe RAID array creation process. Meanwhile, I'm running a crazy RAID 0 of three 32GB Intel Optane SSDs on a Z270 board. Back to an 80GB OS drive, like the old days with one of Intel's G2 SATA II SSDs! Just want to show you how they do, the 4K Random Read result at the bottom is the most interesting. FYI, the Optane SSDs run at DMI3 x2, not x4:  ------------- http://valid.x86.fr/48rujh" rel="nofollow"> 
|
DooRules wrote:Posted By: DooRules
Date Posted: 26 Oct 2017 at 5:41pm
|
" rel="nofollow - Those 4k numbers sure do jump out at ya. Just nice to see someone to start pushing intel on many different fronts. They have had it too easy for too long. So not much bandwidth for those Optane ssd's to fly, interesting. Do you find it very snappy with those 4k numbers? I think you must. Is this the Optane you are using? https://www.newegg.ca/Product/Product.aspx?Item=N82E16820167427 Still using my two 950's here. I have not really seen anything to make me change my mind.
|