Icelake with integrated TB3 controllers completely crushes non-Icelake CPU's with external controllers in performance (up to 40% gains...)
Just saw this and surprised more people aren't talking about it here?
The difference in performance is so massive, despite being a "weaker" processor, that I'd argue it's foolish to invest into a high-end eGPU setup without an icelake CPU. I'm really regretting my RB15 purchase I did last year now. I'll need to replace it if I want to invest into the 3XXX series and actually take advantage of it. So much for me getting an eGPU + this laptop as a means to have my setup be close to as future proof as a desktop!
I'm on Ice Lake and I am not seeing any dramatic improvements in real world performance. With an RX580 I was getting slightly worse performance than the person who maintains the RX580 gaming on eGPU thread, and I think they are on a 4th or 5th gen CPU.
This video simply highlights the TB3 performance deficits seen in 8th-gen or newer H-CPU systens with unoptimized BIOS. Older U-CPU systems are not affected. Consider some background reading in this thread:
The video compares Ice Lake 1065G7 and Comet Lake 10750H. Ice lake does better with eGPU because it uses iTBT (integrated Thunderbolt) instead of dTBT (discrete Thunderbolt).
Tiger Lake 1165G7 will have iTBT (Thunderbolt 4) so it should also have good eGPU performance.
FYI, OWC is taking preorders for a Thunderbolt 4 hub (3 downstream Thunderbolt ports, one USB type A port). It doe not advertise Mac compatibility but I'll try it anyway (I don't have any Tiger Lake CPUs so I don't know what functionality, if any, will work).
I have the latest mbp 12 2020 4TB and it is not much different than the 2018 model mbp 13 4TB. CPU is about 10% faster but with a 5700xt the performance seems similar to me. However , the new one seems to run a bit cooler.
A) 2020 MacBook Pro, i7-1038NG7, 32GB RAM, 1TB, EGPU Razer Core X, Nitro+ RX5700 xt 8Gb, Samsung 65 Q70R
Mac OS Catalina 10.15.5, Internal Bootcamp Windows 10 2004 pci.sys swap.
B) 2.7 GHz I7 4 Cores, 16Gb, 1TB MBP 13 2018 TB3 , EGPU Razer Core X, Nitro+ RX5700 xt 8Gb, LG 32UK550
Mac OS Catalina 10.15.2, Ext SSD Windows 10 1903 V1 .295
@lycid, Tiger Lake laptops have started to appear. Search for laptops with any of the following CPUs:
U-series Series: Tiger Lake Tiger Lake UP3 cores / threads cache Intel Core i7-1185G7 3.0 - 4.8 GHz 4 / 8 12 MB Intel Core i7-1165G7 2.8 - 4.7 GHz 4 / 8 12 MB Intel Core i5-1145G7 2.6 - 4.4 GHz 4 / 8 8 MB Intel Core i5-1135G7 2.4 - 4.2 GHz 4 / 8 8 MB Intel Core i3-1125G4 2.0 - 3.7 GHz 4 / 8 8 MB Intel Core i3-1115G4 3.0 - 4.1 GHz 2 / 4 6 MB Y-series Series: Tiger Lake Tiger Lake UP4 cores / threads cache Intel Core i7-1180G7 2.2 - 4.6 GHz 4 / 8 12 MB Intel Core i7-1160G7 2.1 - 4.4 GHz 4 / 8 12 MB Intel Core i5-1140G7 1.8 - 4.2 GHz 4 / 8 8 MB Intel Core i5-1130G7 1.8 - 4.0 GHz 4 / 8 8 MB Intel Core i3-1120G4 1.1 - 3.5 GHz 4 / 8 8 MB Intel Core i3-1110G4 2.5 - 3.9 GHz 2 / 4 6 MB
And make sure they have Thunderbolt 4.
Amazon.com and NewEgg.com have search results.
For the 1165G7, www.notebookcheck.net has four reviewed laptops:
- Acer SF514-55T (Intel Iris Xe G7 96EUs, 14")
- Acer Swift 5 (2021) (Intel Iris Xe G7 96EUs, 14")
- Asus ZenBook 14 UX425EA (Intel Iris Xe G7 96EUs, 14")
- Asus ZenBook Flip S UX371 (Intel Iris Xe G7 96EUs, 13.3")
But maybe should wait for someone to compare Tiger Lake and Ice Lake eGPU performance.
@joevt, I'm picking up an HP Spectre x360 2-in-1 15.6" [15-eb1043dx] to test this weekend. It has i7-1165G7 and one Thunderbolt 4 port. Performance should be similar with a well-optimized Ice Lake system. The only performance improvement would come from an eGPU enclosure with a newer Thunderbolt controller implementation. I tested an eGPU prototype [build link] earlier this year and H2D speed was close to 3,000 Mbps. Due to NDA I could not take it apart to see what it was.
As @nando4 has mentioned and linked to his well-researched topic, the H-systems are suffering due to unoptimized Thunderbolt 3 connection. You can see an optimized H-CPU system such as 2019 16-in MacBook Pro beating Ice Lake systems in eGPU performance; @bmagnien shared results from his setup using Mantiz Saturn Pro + RTX 3090 [build link]. Another clear example is @mo_oi's RTX 3080 eGPU performance between HP OMEN 17-cb0065TX vs 2017 27" iMac 5K [build link].
@lycid We don't discuss these YouTube videos because many of them have rather limited exposure to this technology. They often lack depth and use shocking observations for views. The bottom line is Intel has done a shit job in educating people on Thunderbolt in general and eGFX in specific. We try to share information as we learn but our voices don't echo as far.
@itsage, by the device IDs 0x15ef and 0x15f0, the prototype eGPU obviously uses a Titan Ridge controller, but not sure if it's the JHL7540 or the JHL7440. The document at https://downloadmirror.intel.com/28658/eng/TBT_Release_Notes_220.127.116.110.txt seems to suggest JHL7440 but why would they use a chip that has USB-C support? My HP Thunderbolt Dock G2 has USB-C support, but the USB controller has ID 0x15EC while the bridge is 0x15F0. I haven't been able to open it up to find out which chip it uses. Maybe the PCI IDs database needs correction.