VOGONS


First post, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Were there any multi-lane PCIe cards for USB 3.0 and SATA 3.0 which utilised more than one lane per available port? When I see photos of PCIe x4 or x8 card, I am noticing that the lanes tend to be dedicated to particular USB 3.0 port or SATA port. Do any PCIe upgrade cards allow for multi-lane traffic on a single port to increase throughput?

The best examples I can find today are of a USB 3.0 card with 4 ports. The slot is x4, but each port only uses one lane. I can find SATA 3.0 cards (Adaptec ASR-6805T) in x8, but I suspect each port gets a dedicated lane. There are also some no-name SATA 3.0 cards which have an x4 connector, but if you inspect the traces more closely, you'll discover that only 2 lanes are connected. Upon reading the specifications, it seems that they are only allowing one lane per port as well. Is it not possible to use multi-lane traffic for USB/SATA like with graphic cards?

It appears to me as if most of these USB 3.0 and SATA 3.0 upgrades are targeting PCIe 2.0 installations, which is unfortunate because PCIe 1.0a was around for 4 years before PCIe 2.0 and many motherboards with PCIe 1.0a already had SATA 2.0/USB 2.0 and could benefit from SATA 3.0/USB3.0. My ASRock 939Dual-SATA2 is one example, and my MacPro 1,1 desktop with dual Xeon X5365, 3 GHz CPUs is another. As SSD prices drop like mosquitos on a scorching hot day in northern Ontario, I see 4 TB SSDs in the forecast for these systems. Proper allocation of lanes could allows these systems to better utilise the SATA 3 and USB 3 protocols.

My MacPro 1,1 has two x4 slots, one x1 slot, and one x16 slot. Currently dual boots Mac OS 10.11 and Ubuntu 16.04, but spends 99% of its time in Ubuntu. I'd like to keep OS X 10.11 driver support though. It currently has 16 GB of RAM, but can be upgraded to 32 GB. I was hoping for a 4-port SATA 3.0 card with each port utilising at least 2 lanes. I was also hoping for a 3- or 4-port USB 3.0 card with each port utilising at leats 2 lanes.

My ASRock 939Dual-SATA2 has three PCI, one x1, one x16 slot, and one AGP 8x slot. It currently dual boots XP and Ubuntu 16.04LTS, but spends 99% of its time in Ubuntu. I'd like to keep XP driver support, so most of those newer Adaptec cards are out of consideration even if they had multi-lane port allocation. Currently the system has 4 GB of RAM and is using an HD4650 AGP graphics card. I was thinking that to optimise the buses, PCI = 1 gigabit ethernet; PCIe-x1 = USB 3.0; PCIex16 = SATA 3.0, but again, with each port utilising at least 2 lanes. Better yet would be a USB 3.0/SATA 3.0 combination card which was keyed for x4 or x8, and has each USB and each SATA port utilising 2 lanes. There is such a combo card, but when inspecting it further, only 2 lanes are connected, and one is for USB 3.0, the other for SATA 3.0. Again, it is targeting PCIe 2.0 upgrades, not really PCIe 1.0a upgrades.

Was PCIe 1.0a thrown under the bus? USB 3.0 came around in 2008, so I'd have figured there would have been plenty of interested parties making multi-lane upgrades for PCIe 1.0a, assuming it is possible. Same for SATA 3.0, which was released in 2009. Didn't users with PCIe 1.0a want to upgrade? Or is there some technical limitation which I do not understand that prevents multi-lane, single-port traffic for SATA 3.0 and USB 3.0?

So that people reading this don't have to look up speed specs:

PCIe 1.0a
x1 = 250 MB/s = 2 Gbit/s
x2 = 500 MB/s = 4 Gbit/s
x4 = 1000 MB/s = 8 Gbit/s
x8 = 2 GB/s = 16 Gbit/s
x16 = 4 GB/s = 32 Gbit/s

PCIe 2.0
x1 = 500 MB/s = 4 Gbit/s
x2 = 1000 MB/s = 8 Gbit/s
x4 = 2 GB/s = 16 Gbit/s
x8 = 4 GB/s = 32 Gbit/s
x16 = 8 GB/s = 64 Gbit/s

PCI (conventional)
PCI (32-bit, 33 MHz) = 133 MB/s = 1.06 Gbit/s
PCI-X (64-bit, 66 MHz) = 533 MB/s = 4.3 Gbit/s

AGP (runs at 66 MHz)
AGP 1X = 266 MB/s = 2.1 Gbit/s
AGP 2X = 533 MB/s = 4.3 Gbit/s
AGP 4X = 1066 MB/s = 8.5 Gbit/s
AGP 8X = 2133 MB/s = 17 Gbit/s

SATA uncoded rates
SATA 1.0 (1.5 Gbit) = 150 MB/s = 1.2 Gbit/s
SATA 2.0 (3.0 Gbit) = 300 MB/s = 2.4 Gbit/s
SATA 3.0 (6.0 Gbit) = 600 MB/s = 4.8 Gbit/s

USB
USB 1.0 = 1.5 MB/s = 12 Mbit/s
USB 2.0 = 60 MB/s = 480 Mbit/s
USB 3.0 = 625 MB/s = 400 MB/s (uncoded) = 3.2 Gbit/s

USB 3.1 1x1 = 625 MB/s = 605 MB/s (uncoded) = 4.8 Gbit/s (1-lane)
USB 3.1 2x1 = 1250 MB/s = 1210 MB/s (uncoded) = 9.7 Gbit/s (1-lane)
USB 3.2 1x1 = 625 MB/s = 605 MB/s (uncoded) = 4.8 Gbit/s (1-lane)
USB 3.2 1x2 = 1250 MB/s = 1210 MB/s (uncoded) = 9.7 Gbit/s (2-lane)
USB 3.2 2x1 = 1250 MB/s = 1210 MB/s (uncoded) = 9.7 Gbit/s (1-lane)
USB 3.2 2x2 = 2500 MB/s = 2420 MB/s (uncoded) = 19.4 Gbit/s (2-lane)

Ethernet 100 mbit = 12.5 MB/s = 0.1 Gbit/s
Gigabit ethernet = 125 MB/s = 1 Gbit/s

Last edited by feipoa on 2021-02-13, 20:45. Edited 1 time in total.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 1 of 22, by Doornkaat

User metadata
Rank Oldbie
Rank
Oldbie

I can't answer your original question of wether there are any USB 3.0 controllers that utilise multiple PCIe lanes.
But really, what would have been the point of those back when PCIe 1.x was relevant? A single PCIe 2.0 lane is enough for a single USB 3.0 connection. Users with only PCIe 1.x would still get an impressive improvement over USB 2.0 and honestly what would you have done over USB back then that would have required that much bandwidth on a single port? Who would have needed this product?
Additionally by requiring two or four PCIe lanes a manufacturer would limit sales of this card to customers with a free x4 (or larger) PCIe slot.
The first USB 3.0 motherboards hit the market in late 2009 or early 2010. PCIe 2.0 had been around for two years by then. If you had any actual need for full USB 3.0 transfer speeds you were likely to also have had a PC that had PCIe 2.0 ports.
That's why I'm doubtful such a card existed.

I think the better SATA 6Gbit/s RAID controllers combine PCIe lanes though. Afaik if they offer hardware RAID functionality they basically have to combine PCIe lanes, right?

Reply 3 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++
weedeewee wrote on 2021-02-13, 13:03:

Yes, I've got one USB3 PCIe card that uses 2 lanes... annoying form since you can't use it in a x1 slot

What is the name/model of it? Can it use 2 lanes on one of the USB ports?

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 4 of 22, by weedeewee

User metadata
Rank Oldbie
Rank
Oldbie

The card is using an ASM1142 chip and is labeled
SW-2042 V0.5
PCIE_2_USB31-0V5.PCB
USB 3.1 Host Controller Card
though from the chip datasheet it seems only to be usable on PCIe gen 2 and gen 3 .

and recently someone posted a PCI USB3 card... kinda baffled at how that is supposed to work properly at speed.

Reply 5 of 22, by Doornkaat

User metadata
Rank Oldbie
Rank
Oldbie

For USB 3.1 there's also an ASUS card that uses four lanes and only has two ports so it has to combine.
But as far as USB 3.0 goes I haven't ever seen a card that uses more than one lane per controller chip. (Me not having seen one doesn't mean much though.)

Reply 6 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I have the Adaptec ASR-6805T, so if I have time, I might try to determine if a single SSD can exceed the PCIe 1.0a single lane bandwidth.

I'd like to understand if there is a technical limitation to using more lanes per device. These x4 cards, using just 2 lanes, is curious. Why not connect the other two lanes for use with PCIe 1.0a and 2.0 slots? Makes me think there is some technical limitation.

Even if we are to take a PCIe 2.0 spec for single lane - 4 Gbit/s - the uncoded USB 3.1 gen 2 rate is 9.7 Gbit/s, so only half the speed can be utilised. Looking at USB 3.1 gen 1 - 4.8 Gbit/s - also cannot be fully utilised.

It looks like the USB 3.2 specification allowed for 2-lane traffic. The naming can be pretty confusing, but USB 3.2 Gen 1 had a 2-lane variant, referred to as 1x2. There is also a 3.2 Gen 2 varient that is 2-lane, called 2x2. Perhaps this is what I should be looking for and hope it works on PCIe 1.0a. However, I am getting the impression that this multi-lane traffic is for the USB-C type cables.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 7 of 22, by weedeewee

User metadata
Rank Oldbie
Rank
Oldbie

It's easier to say that PCIe gen1 could do 250MB/s , gen 2 could do 500MB/s and gen3 can do 1GB/s per lane.

So.. can an SSD fully utilize one lan on a PCIe gen 1. I'd say yes. (if it's at least a good quality Sata2 drive that can reach 300MB/s transfer rate and the CPU & chipset can follow with throwing the data where it needs to go at the speed it's coming in or faster)

and don't confuse the decoded bitrate with the raw bitrate.

Reply 8 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I read that quality SATA 3 SSD drives can reach 560 MBytes/s in real tests. With PCIe 1.0a, I would need to utilise at least 2 lanes to get close to this.

Real world, uncoded, USB 3.0 should reach around 400 MBytes/s, assuming the external drive is a capable SSD. I would also need to utilise 2 lanes of PCIe 1.0a to reach this.

Hence, I am looking for USB 3.x and SATA 3.0 expansion cards which can utilise 2 lanes per port. The two computers I mentioned are our everyday use/work and it would be nice to maximise them. I also backup data to external HDDs for off-network storage, so again, it would be nice to utilise full USB3 speeds without upgrading the whole system.

Those Asus USB 3.1 x4 cards look promising, however it does look as if they are only using 2-lanes total. These are the most zoomed in images I could find though. First two lanes go to the IC, the other two go nowhere?

AsRock_USB3.1.jpg
Filename
AsRock_USB3.1.jpg
File size
81.69 KiB
Views
380 views
File license
Fair use/fair dealing exception
Asus_USB_3.1.jpg
Filename
Asus_USB_3.1.jpg
File size
72.7 KiB
Views
380 views
File license
Fair use/fair dealing exception

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 10 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++

You have this card in person? Are you able to follow the traces to see if Lanes 2 and 3 are connected to the IC?

Could the PCIe 2.0 requirement be one of power? If so, the Asrock card (shown above) has external power.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 11 of 22, by weedeewee

User metadata
Rank Oldbie
Rank
Oldbie

I mentioned which card I have in a previous comment I made here.
I doubt that the PCIe2.0 requirement is one about power.

try this one https://www.ebay.com.my/itm/Startech-PEXUSB3S … on/254773500333

edit: added link

Reply 12 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Sorry, must have got you mixed up with the other poster.

Thanks. The descriptions mentions 5 Gbps per port, which is the coded datarate for USB 3.0 (3.2 Gbps for the uncoded/usable speed). This implies to me that each port gets a single lane only. Bang for buck, this one seems better: https://www.ebay.com.my/itm/254833154246 , but I don't think either of these combine lane traffic per port to account for PCIe 1.0a.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 13 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++
weedeewee wrote on 2021-02-13, 23:46:

I doubt that the PCIe2.0 requirement is one about power.

FYI, from wiki:

PCI Express 2.1 (with its specification dated 4 March 2009) supports a large proportion of the management, support, and troubleshooting systems planned for full implementation in PCI Express 3.0. However, the speed is the same as PCI Express 2.0. The increase in power from the slot breaks backward compatibility between PCI Express 2.1 cards and some older motherboards with 1.0/1.0a, but most motherboards with PCI Express 1.1 connectors are provided with a BIOS update by their manufacturers through utilities to support backward compatibility of cards with PCIe 2.1.

So it may very well be that the PCIe2.0 requirement on the card you reference is due to a power issue with respect to PCIe 1.0a.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 14 of 22, by weedeewee

User metadata
Rank Oldbie
Rank
Oldbie

FYI, from the wiki :

The card's Serial ATA power connector is present because the USB 3.0 ports require more power than the PCI Express bus can supply. More often, a 4-pin Molex power connector is used.

Reply 15 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I was unable to find a photo of your specific SW-2042 V0.5 card. So I was only guessing that your card was missing the SATA/MOLEX power connector, similar to how that Asus card is missing it. The Wiki post I pasted is likely referencing cards without the SATA/MOLEX auxiliary power connector - so cards without this aux power that draw more power than PCIe 1.0a spec and thus will not be supported on PCIe 1.0a systems [without the aux power].

Does your card have a SATA or MOLEX power connector?

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 17 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++
weedeewee wrote on 2021-02-14, 16:32:

From what I can tell there has been no change in max power delivery on any of the PCIe standards and that seems to be beside the point.

I'm not sure what your point is; I am trying to determine why your USB 3.1 card explicitely states it needs PCIe 2.0 to function. Does your card have a SATA or MOLEX power connector?

The link I pasted in wiki hints to there being different slot power requirements for the different PCIe specs. Is there some underlying spec difference which doesn't allow newer PCIe spec cards from running in PCIe 1.0a? It seems that these newer 2-lane USB 3.2 products may fit the bill, but not if they won't work in PCIe 1.0a.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 18 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++

For anyone who stumbles upon this thread, I found the fastest option for multi-lane SATA is the Adaptec 6805T with 512 MB controller memory. It is an 8x device and when in stripe mode with two Samsung 870 EVO SSD drives, it would yield an average read speed of 657 MByte/s and write speed of 985 MByte/s. The sample sized used in these tests was set to 10 MBytes, which is the default for the Disks app in Ubuntu. Write-back cache policy is used in the 6805T BIOS. I have the backup capacitor/battery installed to the controller. When the sample test size increases beyond the 512 MB cache size of the controller, the write rate drops to about the speed of the SSD. I did a benchmark with a 1000 MByte test size and the write rate dropped to 550 MBytes/s, which I think is the limit of the Samsung 870 EVO. Interestingly, the read rate increased to 714 MBytes/s for this large sample size.

Fastest_PCIe_1.0a_SATA-III_throughput_in_stripe.png
Filename
Fastest_PCIe_1.0a_SATA-III_throughput_in_stripe.png
File size
156.01 KiB
Views
81 views
File license
CC-BY-4.0

I was unable to find a 2-lane USB 3.x card which would work well in my system. I settled on a PCIe x1 card. Fastest I could find was that based on a Fresco Logic FL1100. When connected to an external SSD on USB3 adapter, I noticed an average read speed of 208 MByte/s and write speed of 169 MByte/s. I retested it recently and noticed a write speed of 174 MByte/s.

Fastest_PCIe_1.0a_USB3_throughput.png
Filename
Fastest_PCIe_1.0a_USB3_throughput.png
File size
156.13 KiB
Views
81 views
File license
CC-BY-4.0

On different motherboard, an ASUS based on the i915GL, some of my multi-lane PCIe USB3 cards did function in 2-lane mode on PCIe 1.0a. So if your PCIe 1.0a motherboard has a free x4 slot, these adapters might be of some use.

One was based on the ASM1142 chip and demonstrated a read speed of 395 MByte/s and write of 258 MByte/s.

The other was based on the ASM3142 and demonstrated a read speed of 388 MByte/s and write of 264 MByte/s.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 19 of 22, by megatron-uk

User metadata
Rank Oldbie
Rank
Oldbie

That's pretty impressive performance.

I've got a much older technology 3Ware/AMCC 9650SE (PCIe x4 interface) which is only SATA 2 compatible, running 5x 8TB drives and they sustain an average around ~300MB/sec. Peak throughput when measured with hdparm or bonnie is just over 500MB/sec.

3ware.png
Filename
3ware.png
File size
275.49 KiB
Views
55 views
File license
CC-BY-4.0

It shows just how much difference there is in consistency of SSD versus rotational drives! 😁

Ideally I'd like a more modern controller but there's not a lot of choice in x4 interface format.

My collection database and technical wiki:
https://www.target-earth.net