USB 3 Slower on nMP

Eziggy

macrumors newbie
Original poster
Sep 7, 2010
5
1
Hi All,

First time poster, long time MacRumors reader. A quick thanks to everyone here for their great insights over the years.

I just received the nMP (6-core/16GB/512GB/D700) and started setting it up. I noticed something interesting - I tested the speed of my external USB 3.0 drive and noticed it ran at least 50-60MB/s slower than on my 13" rMBP (2013).

The external drive is an Intel 520 480GB SSD in an Oyen Digital USB 3/eSATA enclosure. On the rMBP it works amazing, actually was faster than the Seagate thunderbolt adapter, which I sent back.

AJA: ~400MB/s read/write on the rMP vs. mid-300s on the nMP
BlackMagic: 400+ read on the rMBP, mid-300s on the nMP

Anyone else noticed this or would know why this is?

Thanks!
 

Cubemmal

macrumors 6502a
Jun 13, 2013
824
1
Hi All,

First time poster, long time MacRumors reader. A quick thanks to everyone here for their great insights over the years.

I just received the nMP (6-core/16GB/512GB/D700) and started setting it up. I noticed something interesting - I tested the speed of my external USB 3.0 drive and noticed it ran at least 50-60MB/s slower than on my 13" rMBP (2013).

The external drive is an Intel 520 480GB SSD in an Oyen Digital USB 3/eSATA enclosure. On the rMBP it works amazing, actually was faster than the Seagate thunderbolt adapter, which I sent back.

AJA: ~400MB/s read/write on the rMP vs. mid-300s on the nMP
BlackMagic: 400+ read on the rMBP, mid-300s on the nMP

Anyone else noticed this or would know why this is?

Thanks!
Yup, two things. The chipset still doesn't include USB 3, so that's on an external USB controller. Two, by the time you get to USB the system is running out of PCIe lanes. Check the Anandtech article, I forget how many the USB system gets but it's not a lot. Regardless he warns you not to use USB in the hopes of it being a poor mans Thunderbolt.

If you have USB disks you want to use, you'd probably be better served by putting them on the Thunderbolt chain via a USB adapter or dock. USB3 on the nMP should only be light duty like keyboards and the like.
 

paulrbeers

macrumors 68040
Dec 17, 2009
3,961
120
Anyone else noticed this or would know why this is?

Thanks!
As already stated, technically the USB bus only gets 1 PCIE 2.0 lane which only works out to 4 gb/s instead of 5. I wouldn't go as far as to say the USB 3.0 is only good for keyboards, but it will be slower for SSDs and RAIDs. Regular hard drives won't have any issues.
 

Anim

macrumors 6502a
Dec 16, 2011
614
22
Macclesfield, UK
Doesn't USB3 share lanes with one of the TB controllers? If your using a TB port for your monitor try it in a diff socket and see if that makes a difference? No idea if that idea has any merit just a wild suggestion.
 

paulrbeers

macrumors 68040
Dec 17, 2009
3,961
120
Doesn't USB3 share lanes with one of the TB controllers? If your using a TB port for your monitor try it in a diff socket and see if that makes a difference? No idea if that idea has any merit just a wild suggestion.
Anandtech believes it has it's own PCIE lane BUT one PCIE 2.0 lane wouldn't supply enough bandwidth for full USB 3.0

The TB ports are off of the CPUs PCIE 3.0 ports

http://anandtech.com/show/7603/mac-pro-review-late-2013/8
 

Eziggy

macrumors newbie
Original poster
Sep 7, 2010
5
1
Yup, two things. The chipset still doesn't include USB 3, so that's on an external USB controller. Two, by the time you get to USB the system is running out of PCIe lanes. Check the Anandtech article, I forget how many the USB system gets but it's not a lot. Regardless he warns you not to use USB in the hopes of it being a poor mans Thunderbolt.

If you have USB disks you want to use, you'd probably be better served by putting them on the Thunderbolt chain via a USB adapter or dock. USB3 on the nMP should only be light duty like keyboards and the like.
Thanks! That makes sense, looks like Thunderbolt it is then. Bummer on the limited USB 3.0 throughput, looking forward to seeing more thunderbolt enclosures.

----------

As already stated, technically the USB bus only gets 1 PCIE 2.0 lane which only works out to 4 gb/s instead of 5. I wouldn't go as far as to say the USB 3.0 is only good for keyboards, but it will be slower for SSDs and RAIDs. Regular hard drives won't have any issues.
Thanks, it's unfortunate there aren't many decent Thunderbolt 1 or 2 enclosures. Looks like a dock is the way to go.
 

bcuzawd

macrumors member
Dec 25, 2013
98
0
Great info!

USB 3.0 external drive has now officially been removed from my "is this a good alternative" list.
 

Dranix

macrumors 65816
Feb 26, 2011
1,007
410
left the forum
Each day we find another one to whine on it. If it doesn't fit your reqs get something different. This whining leeds to nothing.
 

vmacrob

macrumors newbie
Jan 8, 2014
16
2
Anandtech believes it has it's own PCIE lane BUT one PCIE 2.0 lane wouldn't supply enough bandwidth for full USB 3.0

The TB ports are off of the CPUs PCIE 3.0 ports

http://anandtech.com/show/7603/mac-pro-review-late-2013/8
I don't think this is correct. Anandtech states "A single PCIe 2.0 lane offers a maximum of 500MB/s of bandwidth in either direction (1GB/s aggregate), which is enough for the real world max transfer rates over USB 3.0"

USB 3 is 5 Gbps bi-directional, PCIe 2.0 is 1 GBps (or 8 Gbps). So, there shouldn't be a limitation, at least when one device is used.
 

Bear

macrumors G3
Jul 23, 2002
8,088
4
Sol III - Terra
I don't think this is correct. Anandtech states "A single PCIe 2.0 lane offers a maximum of 500MB/s of bandwidth in either direction (1GB/s aggregate), which is enough for the real world max transfer rates over USB 3.0"

USB 3 is 5 Gbps bi-directional, PCIe 2.0 is 1 GBps (or 8 Gbps). So, there shouldn't be a limitation, at least when one device is used.
Shouldn't you be comparing the single direction speed of PCIe to the speed of USB 3.0. The reason I suggets this is if you're doing mostly reads or mostly writes, it's the maximum single direction speed you need to compare.
 

analog guy

macrumors 6502
Mar 6, 2009
358
0
Anandtech believes it has it's own PCIE lane BUT one PCIE 2.0 lane wouldn't supply enough bandwidth for full USB 3.0

The TB ports are off of the CPUs PCIE 3.0 ports

http://anandtech.com/show/7603/mac-pro-review-late-2013/8
small tangent: has anyone seen a similar breakdown for the late-2013 iMacs?

this information has been good as i plan my external storage strategy for my nMP, but i wonder if the iMac is similarly affected (i.e. if the storage strategy for an iMac would have the same considerations as one for the nMP).

i suspect the iMac has more lanes open to it because lanes are not taken up by a 2nd graphics card.

thanks!
 

vmacrob

macrumors newbie
Jan 8, 2014
16
2
Shouldn't you be comparing the single direction speed of PCIe to the speed of USB 3.0. The reason I suggets this is if you're doing mostly reads or mostly writes, it's the maximum single direction speed you need to compare.
Ah, you are correct. I was thinking that USB 3.0 was 2.5 Gbps each way - but it's 5 Gbps each way.
 

paulrbeers

macrumors 68040
Dec 17, 2009
3,961
120
small tangent: has anyone seen a similar breakdown for the late-2013 iMacs?

this information has been good as i plan my external storage strategy for my nMP, but i wonder if the iMac is similarly affected (i.e. if the storage strategy for an iMac would have the same considerations as one for the nMP).

i suspect the iMac has more lanes open to it because lanes are not taken up by a 2nd graphics card.

thanks!
IMac wouldn't be affected because the chipset actually has USB 3.0 built in. Because Xeon processors are server chips it has no native USB 3.0 and thus needs to "steal" the bandwidth from PCIE. Also there aren't multiple thunderbolt lanes in an iMac like there are on the Mac Pro.
 

analog guy

macrumors 6502
Mar 6, 2009
358
0
IMac wouldn't be affected because the chipset actually has USB 3.0 built in. Because Xeon processors are server chips it has no native USB 3.0 and thus needs to "steal" the bandwidth from PCIE. Also there aren't multiple thunderbolt lanes in an iMac like there are on the Mac Pro.
thank you. are the two TB ports on the iMac sharing bandwidth, then (similar to how members of each of the 3 pairs of ports on the nMP share a lane)?
 

crjackson2134

macrumors 601
Mar 6, 2013
4,733
1,863
Charlotte, NC
If you are going to use Platter drives, I would think the USB ports would work just fine (save fast Raid 0). I would rather invest in a good TB Dock that supports USB 3 if you need SSD or fast Raid on USB. If you're willing to spend this much on a nMP, then do it right.

Honestly, unless really vested in USB 3.0 devices, I'd rather go with TB anyway.
 

motegi

macrumors regular
May 14, 2009
197
0
sydney.au
If I'm running my RAID5 box of 7200rom disks over USB3, it's not really gonna matter right? Since the drive speed will be the bottleneck not the bus?
 

d-m-a-x

macrumors 6502a
Aug 13, 2011
509
0
If you are going to use Platter drives, I would think the USB ports would work just fine (save fast Raid 0). I would rather invest in a good TB Dock that supports USB 3 if you need SSD or fast Raid on USB. If you're willing to spend this much on a nMP, then do it right.

Honestly, unless really vested in USB 3.0 devices, I'd rather go with TB anyway.


----------

If I'm running my RAID5 box of 7200rom disks over USB3, it's not really gonna matter right? Since the drive speed will be the bottleneck not the bus?
It sounds like It will be fine so long as you use only one usb3 port, for all 4 ports only have one lane. Beyond that a dock may speed things up if configured right
 

Derpage

Suspended
Mar 7, 2012
451
194
Each day we find another one to whine on it. If it doesn't fit your reqs get something different. This whining leeds to nothing.
Yes, yes. Let's embrace poorly executed design at it's best. This is either bad design, or it's a blatant attempt to push TB2.0. Either way, another iToy innovation at it's finest.
 

VirtualRain

macrumors 603
Aug 1, 2008
6,304
115
Vancouver, BC
I'm actually surprised there's a storage solution that can push USB3 to the limit. Hat's off to Oyen Digital. For some strange reason, most USB3 enclosure tests I've seen can't come close to filling USB3's bandwidth. The only other enclosure that could push over 400MB/s is the FirmTek miniSwap that Barefeats tested a while ago.

Personally, I don't care, since I don't have anything that can push USB3 beyond 400MB/s anyway, but if I play arm-chair nMP product manager for a moment... I think it would have been ideal if Apple had opted for two dual-port USB3 chipsets, each on a separate PCIe lane. As I understand it from Anand's article, the USB3 chipset got a single lane along with each GigE port and the Wifi. They could have muxed all the networking onto two lanes (or even a single lane without bottlenecking anything there) and that would have given them two lanes for the USB3. The way it is it's a bit of a lazy/cheap design decision.

EDIT: BTW, almost every USB3 PCIe add-on card I've seen, is PCIe x1 also so this is hardly without precedent.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.