Intel Iris Xe vs UHD 700 series GPUs

[bcm00re 18]

I currently use a HP SFF PC with an i7-6700 as my Emby server.  I am looking to do an upgrade and surprisingly finding that Intel's mobile processors appear to have notably superior GPUs (at least on paper).  For instance https://www.cpu-monkey.com/en/compare_cpu-intel_core_i5_13500h-vs-intel_core_i5_13400 shows the Iris Xe in the mobile unit handily beats the UHD 730 in the desktop one.  To be specific I am looking at Execution Units and Shader values as well as GFLOPS values in the benchmarks.  Should I be concerned with those values and/or are there others I should be paying attention to?  Another appealing plus with going laptop or mini PC would be the reduced power consumption since the server is running 24/7 -- but I don't want to go that route if I am misreading things (and it'll really be a compromise).

 

Edited August 4, 2023 by bcm00re

[bcm00re 18]

BTW I do plan to keep my existing HP PC running using it to host security camera software for my TP Link and Tapo cameras.  And that SFF unit also houses my two 3.5 HDD data drives as well as the optical drive I use to rip my DVDs (both of which I wouldn't have with a laptop or mini PC).

[Luke 37095]

Hi, I would give the slight edge to the XE but considering they're both from the same generation I would expect overall performance for server transcoding to be very similar.

[bcm00re 18]

Thanks!  Other websites (like https://nanoreview.net/en/cpu-compare/intel-core-i5-13500h-vs-intel-core-i5-13400) also seem to indicate notable differences in the GPU performance, but it seems like they don't really translate to real world (at least not for Emby).  Do things like type and/or size of the SSD system drive matter?  What about the amount and/or type (DDR4/DDR5) of RAM?  Are there "minimum" values for either?

 

 

[RanmaCanada 345]

If you're using it for encoding, the ASICS are what matter, and the ASICS across a generation are typically the same, be it desktop or mobile.  I use a laptop for my server, an i3-1115g4 (Tiger Lake), and have no issues with encoding for Emby.  The most people I've had online has been 14, with at least 10 of them transcoding stuff, and no one complained.  All the shaders and what not have no bearing on encoding.  See our discussion here and see the wikipedia article with it's sources here.

[bcm00re 18]

Thank you for the link!  Has Emby done anything with Intel's Deep Link when using an Iris Xe MAX 100 discrete card?  From what I read it is supposed to be geared to encoding not gaming.

 

[Luke 37095]

2 hours ago, bcm00re said:

Thank you for the link!  Has Emby done anything with Intel's Deep Link when using an Iris Xe MAX 100 discrete card?  From what I read it is supposed to be geared to encoding not gaming.

 

It's just part of our general Intel transcoding support.

[bcm00re 18]

I found this discussion:  

But they are talking about using the 2nd generation card not the first gen one that I am considering.  It would be paired with an i5-12400 FWIW.

 

 

Edited August 5, 2023 by bcm00re

[bcm00re 18]

So I settled on a new Lenovo desktop with the i5-12500 (which has the UHD 770 graphics).  I got a pretty good deal on it, but it only comes with 8gb of DDR5-4400 RAM and a 256gb M.2 2280 PCIe 4.0x4 SSD.  Should I consider upgrading either from the get-go?

 

[bcm00re 18]

On 8/8/2023 at 3:14 AM, bcm00re said:

So I settled on a new Lenovo desktop with the i5-12500 (which has the UHD 770 graphics).  

 

Well my order got cancelled just now.  I ordered via eBay from the official Lenovo store.  They claim there was an issue with my shipping address, but I've had the same one on file with eBay for 10+ years so I know it's a bogus excuse.  And they still have the item available on eBay too, but they greatly increased the price.  I would suggest people steer clear of Lenovo after this unscrupulous business practice!  

 

[RanmaCanada 345]

Lenovo has been having serious stock issues.  I've had several friends have orders backdated, and then cancelled from orders that were originally in stock.  Trying to order a server from them right now for a friend's business and the earliest date is January 2024! and we can't upgrade anything at all as there are no options.  Everyone is still having these problems.

[bcm00re 18]

@RanmaCanada

Thanks for the info.  I think this is different though because they are still selling the same model on their website -- but for a considerably higher price.  They shouldn't be allowed to back out of the deal days after I accept their terms and make payment simply because they changed their mind and now want more for it. 

[bcm00re 18]

Noticed today the eBay listing that I bought is back with stock (but still at the much higher price).  I contacted Lenovo via eBay -- where they have responded before -- and asked them to honor my purchase.  If they do not then it should clearly show they are playing games.

 

[bcm00re 18]

This is my last post on the tangent...Lenovo's eBay store is continuing to refuse to honor my purchase despite both eBay and Lenovo's website showing availability of the product (at a new much higher price).  

[bcm00re 18]

So I bought a desktop with an i5-12500 (with 16gb RAM and 256gb SSD) as well as a notebook with an i7-1255U (with 12gb RAM and 512gb SSD).  The desktop has the UHD 770 while the notebook has the Iris Xe GPU.  I plan to put Handbrake on both to do some encoding testing using their queue/batch feature.  Does anyone have other ideas or suggestions on testing to perform to compare the two CPUs/GPUs?

[Luke 37095]

In all of my research they seem to produce very similar results.

[RanmaCanada 345]

They should perform the same as per the fact that the ASICS are the same.  There are some filtering and stuff that could be used differently, but encoding wise, they will produce the same results with the same command lines.

[bcm00re 18]

Well that isn't the case in the real world!  I did some testing tonight, and the i5-12500 is notably more capable than the i7-2255U when is comes to QSV encoding.  I will post more details about my results in the next few days.

[bcm00re 18]

First the PCs that I tested:
A - new/refurbished Lenovo M80t Gen 3 desktop with i5-12500, UHD 770 iGPU, 16GB RAM (single stick), and 256GB SSD running Win10
B - new/refurbished Lenovo IdeadPad 5i 7 notebook with i7-1255U, Iris Xe iGPU, 12GB RAM (soldered), and 512GB SSD running Win11
C - my old HP ProDesk 600 SFF desktop with i7-6700, HD 530 iGPU, 16GB RAM (two sticks), and 256GB SSD running Win10

I freshly installed Handbrake 1.6 on both the new PCs and upgraded the old one to the same version.  All computers had the source files on their SSD and were writing the output to their respective SSDs.  For all of them, I unchecked "Enable Low Power QuickSync Hardware" in Handbrake preferences.  I had the notebook power plugged in and set its power mode to "Best Performance".  In all cases, I started with the Roku 1080p preset then selected H264 QSV.  I created the same batch file for each that consisted of an episode of X-Files (480i), an episode of Lego Masters (720p), Guardian of the Galaxy (480i) movie, and an episode of Saturday Night Live (1080i).  Both the X-Files and GotG were ripped from DVDs (into MKVs) but left in their unaltered mpeg2 formats.  The Lego and SNL episodes were recorded from OTA (to TS files) using a HDHomerun tuner and left in their native mpeg2 formats.

Here are the results showing the encoding time as well as the average frame rate:

		A			B			C
X-Files 	 1:38	692.92		 2:05	554.25		 3:02	372.96
Lego	 	 5:52	644.95		10:04	375.91		15:31	243.52
GotG	 	 3:43	797.81		 5:35	535.05		 7:23	409.39
SNL		18:24	159.75		33:35	 87.53		47:44	 61.56

So even though both the new PCs are using Generation 12 Intel processors and iGPUs, there are very notable differences in performance results.

 

Edited August 24, 2023 by bcm00re

[RanmaCanada 345]

You should run the same test using HEVC files and staxrip as you can then use rigaya's quicksync encoder as it's better than the Handbrake crap.  720p 1080p and 4k.

It will also make it easier for you to post your actual command line so that, as bad as this sounds, we can start compounding results and take a leap forward for quicksync encoding.

I am sure softworkz would love more empirical data to view.

[bcm00re 18]

I didn't realize Handbrake was crap?!?  I might look into that other stuff, but I do have my Handbrake log files if anyone thinks they might be useful.  Last night, I also noticed some differences in what Task Manager was reporting for GPU usage -- see attached images.  They were all taken while doing the SNL encode.  Notice the UHD770 was reporting fairly heavy use on 3D but nowhere else, while the Iris Xe was reporting lesser usage on 3D but also had some usage on Video Decode.  I do notice now that the Iris Xe drivers were over a year old -- they have since been updated to drivers with a July 2023 date.  Maybe I will do the same testing again this weekend but encode using h265 10-bit QSV instead.

[RanmaCanada 345]

I do know that they did update drivers to increase performance.  And what you could be seeing is what softworkz mentioned about different processes using different parts of the chip.  But hard to make any real conclusions as different drivers were used.

[bcm00re 18]

I updated drivers on the notebook and re-ran the same test, but the results didn't change appreciably:

X-Files 	 2:40	415.67
Lego	 	 10:06	374.61
GotG	 	 5:47	510.93
SNL		31.19	 93.83	

It appears that the i5-12500 is still notably better/faster using QuickSync than the i7-1255U -- so the UHD770 trumps the Iris Xe.

 

I did notice the updated driver changed what I saw in the Task Manager window (which again was taken during the SNL encode):

Like the i5-12500 pic, the i7-1255U now only shows activity in the 3D window.

 

Edited August 28, 2023 by bcm00re

[RanmaCanada 345]

hrmm interesting.

[bcm00re 18]

I did a h265 QuickSync test a couple weeks ago but forgot about posting the results until now!  My setup was the same as last time except I started with the Roku 1080p preset then selected H265 QSV -- and I also changed Handbrake's audio setting to only do pass-thru audio (whereas with the h264 tests it was doing the pass-thru and also creating a stereo AAC track).

Here is what I saw with my new desktop:

		time	FPS
xiles		 2:58	371.85
lego		 8:27	447.63
gotg		 7:10	409.36
snl		26:09	112.38

 

And here is what I got with my new notebook:

		time	FPS
xfiles		 3:24	323.17
logo		10:36	356.4
gotg		 8:31	344.3
snl		33:45	87.03

 

So the i5-12500 & UHD 770 combo in the desktop is still faster than the i7-1255U & Iris Xe combo, but it's not near as noteworthy as when doing the h264 encoding.