Intel's Tiger Lake NUC11: Panther Canyon for Asia Alone

27 Comments

View All Comments

• erinadreno - Friday, February 5, 2021 - link

  I bought one without wireless charging lid. The fan seems ramping more aggressively than previous gen nuc, and for some reason the default pl1 is 45w which the cooler obviously cannot handle.

• AdrianBc - Friday, February 5, 2021 - link

  That seems disappointing. The last 2 NUCs that I have bought were two NUC8, including a NUC8i7BEH and those had much better coolers than all the previous NUCs, so they were almost always practically silent.
  The NUC8i7BEH had the short-time power limit set to 50 W and the long term power limit set to 30 W.
  Its cooler handled those powers easily, with little noise, unlike older NUCs. If Intel has downgraded the coolers of the latest NUCs, that is not good.

• erinadreno - Friday, February 5, 2021 - link

  I do also have a NUC8 unit and for sure it's far more quiet. IDK what Intel has been skimped on, but even install some apps got the fan goes up to 3600rpm and give you gaming notebook like sound

• AnybodyM - Monday, February 15, 2021 - link

  Things are getting ridiculous with Intel these days.
  
  Feb 4: Intel announces Panther Canyon will be Asia-only and not come to the EU
  Feb 14: Panther Canyon i7 arrives at Amazon in Europe with excellent availability, two day delivery and seemingly a lot of stock.
  
  Mr Gelsinger, please make the lying stop. This is so disgraceful.

• abufrejoval - Wednesday, February 17, 2021 - link

  I really have only one complaint: What do I do with my NUC8 and NUC10 now?
  (Actually, I'll put them in to a TB3 network cluster :-)
  
  Writing this on the i7 NUC11 freshly rebooted...
  
  Immediate wins: Dual TB3 (too bad it's not TB4, wanted to test that...)
  2.5GB Ethernet saves a Realtek USB dongle and perhaps having to recompile CentOS driver source on every kernel update.
  
  Just a bit surprised to see it kick up to 5.33GHz on Turbo...

• abufrejoval - Monday, February 22, 2021 - link

  A couple of updates:
  
  New turbo record is 5.42 GHz recorded on two 2 of 4 cores via HWiNFO somewhere during 3DMark runs: I'd call it a glitch turbo, because it obviously isn't doesn't stay there long enough to do any significant work. I've never seen these turbos during synthetic CPU loads (4.7GHz until it runs out of thermal headroom).
  
  But the Geekbench5 scores for the i7-1165G7 are astoundingly close to my Ryzen 5800X on precision overclock, at least on Linux, where they tend to be higher than on Windows anyway. So, yes, these cores are nothing to laugh about, 4 cores of i7-1165G7 are almost exactly the same as 6 cores of i7-10700U, while single core, iGPU, thermals or plain everything else (except backward compatibility) are better.
  
  The Xe iGPU certainly is a significant uplift over the NUC8 Iris 655, even without the eDRAM. 3DMark Night Raid is 25FPs at 4k. Don't know how they have done it, but it's impressive, even it's still just as good for games as the NUC10: None whatsoever. But for anything 2D on 4K or Steam remote gaming, all three work equally well. The iGPU receives max 22Watts and seems to sustain that while the fan is allowed to rev. The CPU seems much more thermally limited and will yield clocks to match the power budget settings configured in the BIOS. Without an iGPU workload CPU clocks will remain relatively high, if you mix Prime95 with Furmark, the latter will keep all 22 Watts of the TDP pie.
  
  TB3 networking: Direct links on CentOS8 work out of the box, 9.50Gbit or 1.1Gbyte/s a nice change from the 110MByte/s you get on Gbit or the 330Mbyte/s 2.5GBit links will yield... and vastly better latencies, should you be running HF-trading on these NUCs.
  
  But I need three nodes for a minimal oVirt HCI cluster and that's where I had high hopes for the OWC TB4 hub. Nope, thunderbolt networking doesn't survive having that hub somethere along the path. I'd guess a Linux code issue, but who am I to know?
  
  But with the Tiger Lake NUC having two TB3 ports, I can connect the NUC8 on one side, the NUC10 on the other and have the NUC11 route any traffic between those two...
  
  I was quite shocked to see 21Mbit/s forwarded traffic, when 9.5Gbit/s where possible to either side without the "hop": Linux routing surely can't be that bad?
  
  Then I remembered how important increasing MTU was when I tried to do similar things with Mellanox CX5 VPI hybrid Infiniband/Ethernet adapters, where this routing is actually done on the chip, not the Linux kernel, while we're talking at 100Gbit/s there.
  
  500Mbit/s at 9k MTU seemed encouraging, but still not quite up to snuff, so I went all out with 32K MTU, quite ready to face kernel panics or other types of meltdown...
  
  Instead I got 1.11 Gbyte or 9.5 Gbit/s any which way, direct or forwarded via Linux...
  I immediately sacrificed a couple of innocent coffee beans to Linus and his kernel brethren
  
  Of course that setup isn't redudant with a critical path on the NUC11, but that's only because the other two nodes aren't dual TB.
  
  If you were to take three Tiger Lake NUCs, you can get a really neat and nice 3 node HCI setup with nothing more than 6 short TB cables and an NBase-T switch: run two good old rings between the three NUCs, such that any two NUCs will have at least one direct connection and set up routing with primary on the direct link and secondary via forwarding. That gives you 10Gbit via TB for all Gluster-sync and vMotion traffic while the external Gluster-FS access and GUI access can aggregate 3x 2.5Gbit/s links to almost the same level of bandwidth.
  
  You could use 5GBit Acquantia USB3 NICs and I have tried, but those NICs have unfortunately been killed after the Mellanix acquisition and driver support is dying quickly.

• lindflake79 - Sunday, May 2, 2021 - link

  Does anybody know when it will be possible to buy the phantom canyon in the UK? Currently using the hades canyon and it is time for an upgrade? Are we talking weeks or months? Any information would be very much appriciated!