New Intel 7260 drivers, 17.1.0.19

[code65536]

These are the signed, WHQL drivers that were posted to station-drivers.com. They have not yet appeared on Intel's website. (Update: It's now on Intel's website; see post #10.)

A diff of the INFs for these new 17.1.x vs. the old 17.0.x drivers reveals a new toggle option in the advanced driver settings pane, called "Throughput Booster". The default setting is "Disabled".

I wanted to see just what this Throughput Booster does, so I ran some benchmarks on 11ac (I didn't bother with 11n on 2.4GHz).

Base station: Buffalo WZR-1750DHP with DD-WRT 24865M. Explicit and implicit beamforming enabled, 80MHz channel in U-NII-1.
Client: Laptop with a 7260 card. Approx. 1 meter away from base station. Laptop was kept in the exact position and angle throughout.
Test: Each test done used iperf2 for 15 seconds and a 1MiB TCP window. 10 tests per run, with the highest and lowest results discarded. The lowest, highest, and average of the remaining 8 tests are shown.

Version 17.0.6.1 (old driver)
Downlink: 447-467 Mbps, average of 461 Mbps
Uplink: 367-377 Mbps, average of 374 Mbps

Version 17.1.0.19, "Throughput Booster" off (new driver, default settings)
Downlink: 429-433 Mbps, average of 431 Mbps
Uplink: 300-310 Mbps, average of 305 Mbps

Version 17.1.0.19, "Throughput Booster" on (new driver)
Downlink: 429-432 Mbps, average of 430 Mbps
Uplink: 360-381 Mbps, average of 371 Mbps


After the tests, I did some switching back and forth between the drivers and settings to make sure that these results were consistent, and yes, they were. They're also consistent across hard power-offs (to clear the firmware that is loaded into the card's volatile memory).

So... basically, the throughput booster in 17.1 affects only uplink... and it just restores the uplink performance to where it was with 17.0.

 

[theveterans]

What about actual file transfer speeds from a NAS or hardwired PC to the Intel 7260.

 

[azazel1024]

What about actual file transfer speeds from a NAS or hardwired PC to the Intel 7260.

Agree. I trust iPerf zero. The results I've gotten with iPerf and similar TCP and UDP testing tools for wireless testing have not backed up actual file transfers results in my experience, which is why I never use them with the sole exception that they can be useful to setup a continuous test and run wireshark when attempting to diagnose issues with Tx/Rx wired or wireless.

Also I'd be curious if this is supposed to be something with toggling 256QAM on/off. So if anyone has the card and has an AC1900 router, I'd be interested to see if it either obtains different link rates. Might need to be something we need to wait for and check out in Intels driver notes once it is officially on their site.

 

[code65536]

Nominal link speed was 867Mbps in all cases. It's highly unlikely that it's a QAM256 thing, since the performance discrepancy from that should be far higher. Edit: Oh, you meant the non-standard Broadcom-only QAM256 for 11n on 2.4GHz? But that wouldn't explain the difference in uplink on 11ac.

I like ttcp/iperf because of its consistency and because it allows for better experimental control--you don't have to worry about the source disk and the destination disks keeping up. And it gives you a nice average-over-this-time-period number instead of having to keep your eye on transfer speed graph in Windows. Also, because I've found ttcp/iperf test results to be fairly consistent with what I get copying files to and from my file server over SMB.

Anyway, I did do a quick file copy test last night between a Windows 8.1 client and a file server (also running Windows 8.1) as a "sanity check" (just in the downlink direction, though). I could do an uplink test later if people really care, but, as I said, I've always found ttcp/iperf and SMB to be consistent with each other (just one is far easier to use for testing than the other).

Downlink, old driver, reading the transfer speed during the highest sustained peaks (ones lasting for at least a few seconds) in the graph: ~59-60 MiB/s. (compare with 55 MiB/s iperf2's average; oh, and the iperf2 tests were to and from this same file server).

Downlink, new driver, ~56-57 MiB/s (compare with 51 MiB/s iperf2 average)


As for the performance discrepancies that you guys see between "real-world" and iperf, um, iperf should be real-world performance. Keep in mind that high-speed transmission with TCP--because TCP does require that ACK from the other end--can be very sensitive to various factors. A ttcp/iperf test with a large TCP windows (e.g, 1MiB) generally yield substantially better results (closer to what I get with Windows SMB) than one with a small window. And Windows-to-Windows SMB produces far, far better results--something on the order of 2x as good--than anything involving Samba (which is very poorly-optimized)--this is part of the reason why my file server uses Windows 8.1 using Storage Spaces (4x4TB with 1 disk worth of parity) instead of something like FreeNAS.

 

[theveterans]

Very fast! My laptop with OEM internal realtek 2x2 802.11ac can only get 50 MB/s. Nothing really beats the Intel 7260 on price and performance.

 

[cc666]

Download

Can someone provide a download link?

CC

 

[code65536]

Can someone provide a download link?

CC

The first post of this thread stated that they were downloaded from station-drivers.com, and if you had gone there, you would have noticed a link to it in the news feed on the front page.

 

[azazel1024]

Well, depending on how well or crappy Intel ended up doing things, it litterally could be a toggle between 256QAM on both 2.4 and 5GHz, though I suspect not if it is telling you it is linking at 867Mbps on 5GHz with it toggled on and off.

Probably some other optional feature in 11ac that Intel is toggling on and off with it.

I do love my Intel 7260ac as well, but whether it is card specific or what, with the newer drivers it does NOT play well with my TP-Link WDR3600. I get around 3-4MB/sec transfer speeds with the newer 16.x.x.x branch and all of the 17.x.x.x branch I have tested, though I'll give the 17.1.x.x branch a try. Maybe they've fixed whatever the issue was. The only driver I can use is the original Windows 8.1 16.x.x.x driver and the ONE version newer of the 16.x.x.x driver branch. then I'll get ~23MB/sec on 2.4GHz 40MHz and ~25MB/sec on 5GHz 40MHz.

On my AC1750 router I'll hit around 50MB/sec down and 38MB/sec up on 5GHz 80MHz. I haven't tested the newer drivers with it because the newer drivers are so broken with my WDR3600.

I guess time to break out the latest drivers and do some testing (and some finger crossing. Also, I have a second WDR3600...which I actually haven't tested to see if it'll work okay with the newer drivers. I guess it is always possible it is specific to that one single WDR3600, which in some ways would be nice if it was).

Disk throughput isn't a limitation for me with wireless testing due to my RAID array, though it can be a bit with my wired testing (due to my effectively bonded gigabit ports on my server and desktop).

 

[azazel1024]

Okay, I finally grabbed the latest drivers, still no joy with my TP-Link WDR3600, but it got me playing with it even more. I tracked it down to U-APSD apparently not playing well with the WAP in the newer Intel firmwares. I disabled it and the wireless speed jumped up to the same as it was with the very early 7260ac driver versions.

I did find that the 17.1.x.x is slower than the 17.0.x.x drivers. Not by much, but a little. With my AC1750 and the "accelerator" option toggled on I found that uploads took longer to hit max speed and the same with downloads. Speed was within about 1-2MiB/sec of the 17.0.x.x drivers, but still just a hair slower (peak, about 51.5MiB/sec down, I can't recall what it was up, versus about 53.5MiB/sec down with the last 17.0.x.x drivers).

2.4GHz seemed to be the same with the AC1750 in both driver versions (about 28MiB/sec down and 26MiB/sec up).

With the WDR3600 5GHz was also slower. 2.4GHz was the same 23.5MiB/sec down and 25MiB/sec up, but 5GHz it was 22.5MiB/sec down and 24MiB/sec up. With the 17.0.x.x branch it was 25.5MiB/sec down and 26MiB/sec up.

So, whatever Intel did in the latest branch, it is a bit slower. With the WDR3600 I noticed no difference with the accelerator enabled or disabled. With it disabled on my AC1750 I got around 10% slower than with it enabled (only hit around 47MiB/sec down).

No real noticable difference between the 16.x.x.x driver I had been using and U-APSD on and the 17.0.x.x and U-APSD off in terms of performance (with either the WAP or router), but at least a nice feeling that I am on a later driver branch.

 

[code65536]

17.1.0.19 has been posted on Intel's website.

The driver package also contains updated drivers in the two 15.x branches for legacy cards, but Intel inexplicably stripped out support for the 6200, 6230, 6250, and a number of other cards, while retaining support for other 6xxx cards like the 6205 and 6300. I mean, WTF, isn't the whole point of legacy driver branches to... um... support legacy devices?

Also, I would recommend that people avoid Intel's Bluetooth drivers newer than 17.0.1401. I upgraded a couple of 6xxx cards to 7260's this weekend, and found that there were a lot of issues with Bluetooth devices failing to reconnect properly. At first, I thought it was a problem with the 7260, but I quickly determined that it was a driver issue. 17.0.1405 as well as drivers in the new 17.1 branch (including 17.1.1407 that hasn't yet been posted on Intel's website as of the writing of this post) all exhibit the same reconnection problems on both 7260 and 6235 cards (tried it with three different cards on three different computers).