Wednesday, August 11th 2021

Tokyo Olympics 8K Broadcast Was Powered by Intel Xeon Platinum 8380H Servers

Intel has recently unveiled that their technology powered the 8K 60 FPS Tokyo Olympics broadcast which was available to select customers in Japan. The games were recorded in 8K at 60 FPS with 4:2:2 Chroma subsampling and 10-bit HDR which resulted in a bitrate of 48 Gbps. This uncompressed stream was then encoded on servers each featuring quad 28 core Xeon 8380H processors and 384 GB of ram into two HEVC distribution streams at 250 Mbps and 50-100 Mbps bitrates. These streams were then distributed to users over the internet where the 8K stream could be decoded and displayed on an 8K TV over HDMI 2.1. Intel used a workstation PC with an 18-core Xeon W-2295 and 64 GB of ram to decode and play the video stream on the TV. This 8K service was only made available to select NHK subscribers in Japan while most international broadcasts offered a maximum resolution of 4K.
Add your own comment

53 Comments on Tokyo Olympics 8K Broadcast Was Powered by Intel Xeon Platinum 8380H Servers

#1
Richards
Intel server cpu's are reliable that's why they dominate the server market.. and they used optene ssd's
Posted on Reply
#2
Steevo
RichardsIntel server cpu's are reliable that's why they dominate the server market.. and they used optene ssd's
Intel pays off companies to use their hardware, that’s why.
Posted on Reply
#3
R0H1T
Is that the reason (8380H) why there was a sudden spike in earth's temperatures the last few weeks :laugh:
Posted on Reply
#4
eidairaman1
The Exiled Airman
RichardsIntel server cpu's are reliable that's why they dominate the server market.. and they used optene ssd's
Blind much
SteevoIntel pays off companies to use their hardware, that’s why.
This
Posted on Reply
#5
Crackong
RichardsIntel server cpu's are reliable that's why they dominate the server market.. and they used optene ssd's
1. Optane SSDs isn't platform specific.
2. "Reliable" is a relative term in these large scale installments since they run custom OS and custom software tuned specifically to its application.
Posted on Reply
#6
Cheese_On_tsaot
Well those CPU's might be a bit faster than a Tualatin right?
Posted on Reply
#7
quanash
They are using INTEL XEON CPUs, not because INTEL is paying them nor anything close to it; but because the encoding software they are using demanding AVX512 VNNI ( AVX512 Vector Neural Network Instructions ) instruction set that only can be found on INTEL XEON CPUs.

I don't know why people are just making comments without reading the fine prints ( the software used in the pipeline mentioned in the slide ) and make a little research ( looking at the systems specs for that specific software ). Not every software is AMD EPYC CPU compatible, sorry.

They are using Spin Digital enc live V1.0 software and this software's platform support is;

OS Red Hat Linux 7/8, Ubuntu Linux 18.04/20.04

CPU X86_64. SIMD instructions: SSE 4.1, AVX2, AVX512, VNNI

SDI AJA Kona 5 (12G), AJA Corvid 44 (12G), AJA Corvid 88 (3G)

Source
Posted on Reply
#8
Pepamami
quanashThey are using INTEL XEON CPUs, not because INTEL is paying them nor anything close to it; but because the encoding software they are using demanding AVX512 VNNI ( AVX512 Vector Neural Network Instructions ) instruction set that only can be found on INTEL XEON CPUs.

I don't know why people are just making comments without reading the fine prints ( the software used in the pipeline mentioned in the slide ) and make a little research ( looking at the systems specs for that specific software ). Not every software is AMD EPYC CPU compatible, sorry.

They are using Spin Digital enc live V1.0 software and this software's platform support is;

OS Red Hat Linux 7/8, Ubuntu Linux 18.04/20.04

CPU X86_64. SIMD instructions: SSE 4.1, AVX2, AVX512, VNNI

SDI AJA Kona 5 (12G), AJA Corvid 44 (12G), AJA Corvid 88 (3G)

Source
Aha! Intel payed them for using that specific software!
**joke**
Posted on Reply
#9
Cheese_On_tsaot
quanashThey are using INTEL XEON CPUs, not because INTEL is paying them nor anything close to it; but because the encoding software they are using demanding AVX512 VNNI ( AVX512 Vector Neural Network Instructions ) instruction set that only can be found on INTEL XEON CPUs.

I don't know why people are just making comments without reading the fine prints ( the software used in the pipeline mentioned in the slide ) and make a little research ( looking at the systems specs for that specific software ). Not every software is AMD EPYC CPU compatible, sorry.

They are using Spin Digital enc live V1.0 software and this software's platform support is;

OS Red Hat Linux 7/8, Ubuntu Linux 18.04/20.04

CPU X86_64. SIMD instructions: SSE 4.1, AVX2, AVX512, VNNI

SDI AJA Kona 5 (12G), AJA Corvid 44 (12G), AJA Corvid 88 (3G)

Source
Because Intel bad, AMD good even though intel have the best value CPU for casuals.

(Please don't tell team red I said that the supporters are rabbid)
Posted on Reply
#10
R0H1T
quanashnot because INTEL is paying them nor anything close to it
You don't know that for sure, there could well be (significant) money involved. And it's not as if that's the only encoder which works for 8k streams, though admittedly it's likely pretty efficient at what it's doing.
Posted on Reply
#11
Vya Domus
quanashbecause the encoding software they are using demanding AVX512 VNNI ( AVX512 Vector Neural Network Instructions ) instruction set that only can be found on INTEL XEON CPUs.
And that's an instruction set available on what, <0.1 % of CPUs that are in use in servers ? And how many people are using devices than can play and decode 8K60 HDR ? Why not write an encoder for a GPU ?

I don't know if they paid anyone but you have to be pretty obtuse to not realize this is a publicity stunt for something that is otherwise impractical and irrelevant.
Posted on Reply
#12
Cheese_On_tsaot
Vya DomusAnd that's an instruction set available on what, <0.1 % of CPUs that are in use in servers ? And how many people are using devices than can play and decode 8K60 HDR ? Why not write an encoder for a GPU ?

I don't know if they paid anyone but you have to be pretty obtuse to not realize this is publicity stunt for something that is otherwise impractical and irrelevant.
Logically this is hypocritical as we are all pretty obtuse, impractical and irrelevant in this topic as it is all guess work and conspiracy.
Posted on Reply
#13
Tsukiyomi91
guess it's the many reasons why they chose that particular processor and a particular piece of software that's tuned to take advantage of the processor to encode the recordings efficiently.
Posted on Reply
#14
ZoneDymo
RichardsIntel server cpu's are reliable that's why they dominate the server market.. and they used optene ssd's
No Richard....just no...
Posted on Reply
#15
Vya Domus
Cheese_On_tsaotguess work and conspiracy.
It's not a guess or conspiracy that 8K60 HDR streams aren't of any real use for most people right now, or that such an encoder would be a better fit for a GPU.
Posted on Reply
#16
TumbleGeorge
I bet $0 that same work easily will be made with one cheap mainstream CPU with ZEN5 architecture.
Posted on Reply
#17
Tigger
I'm the only one
quanashThey are using INTEL XEON CPUs, not because INTEL is paying them nor anything close to it; but because the encoding software they are using demanding AVX512 VNNI ( AVX512 Vector Neural Network Instructions ) instruction set that only can be found on INTEL XEON CPUs.

I don't know why people are just making comments without reading the fine prints ( the software used in the pipeline mentioned in the slide ) and make a little research ( looking at the systems specs for that specific software ). Not every software is AMD EPYC CPU compatible, sorry.

They are using Spin Digital enc live V1.0 software and this software's platform support is;

OS Red Hat Linux 7/8, Ubuntu Linux 18.04/20.04

CPU X86_64. SIMD instructions: SSE 4.1, AVX2, AVX512, VNNI

SDI AJA Kona 5 (12G), AJA Corvid 44 (12G), AJA Corvid 88 (3G)

Source
Probably the most intelligent post in the thread.
Posted on Reply
#18
Richards
quanashThey are using INTEL XEON CPUs, not because INTEL is paying them nor anything close to it; but because the encoding software they are using demanding AVX512 VNNI ( AVX512 Vector Neural Network Instructions ) instruction set that only can be found on INTEL XEON CPUs.

I don't know why people are just making comments without reading the fine prints ( the software used in the pipeline mentioned in the slide ) and make a little research ( looking at the systems specs for that specific software ). Not every software is AMD EPYC CPU compatible, sorry.

They are using Spin Digital enc live V1.0 software and this software's platform support is;

OS Red Hat Linux 7/8, Ubuntu Linux 18.04/20.04

CPU X86_64. SIMD instructions: SSE 4.1, AVX2, AVX512, VNNI

SDI AJA Kona 5 (12G), AJA Corvid 44 (12G), AJA Corvid 88 (3G)

Source
Tell them.. well explained
Cheese_On_tsaotBecause Intel bad, AMD good even though intel have the best value CPU for casuals.

(Please don't tell team red I said that the supporters are rabbid)
Its so many of them here
Posted on Reply
#19
ratirt
Gruffalo.SoldierProbably the most intelligent post in the thread.
I'm really puzzled why people constantly dismiss the idea of marketing. Guess most people don't realize what power and opportunities lie in the marketing.
And this is a huge marketing point for Intel. Software usage has nothing to do with it. A GPU would have done the 8k encoding/decoding way more efficient and faster than any given CPU.
Besides, there is no mention the AVX512 Decoder/Encoder has been used.
Posted on Reply
#21
Vayra86
ratirtI'm really puzzled why people constantly dismiss the idea of marketing. Guess most people don't realize what power and opportunities lie in the marketing.
And this is a huge marketing point for Intel. Software usage has nothing to do with it. A GPU would have done the 8k encoding/decoding way more efficient and faster than any given CPU.
Besides, there is no mention the AVX512 Decoder/Encoder has been used.
Marketing! Exactly, that one thing AMD historically has a very, very bad relationship with. Market dominance is another one like that. Part of their decision to axe GF and go fabless. We can applaud them producing on TSMC 7nm but let's face it... the volume isn't there nor the control over it.

This blade cuts both ways I hope you can see that. Intel's better marketing is AMD's lack of it. This is about spotting and capturing opportunities.
Posted on Reply
#22
ratirt
Vayra86Marketing! Exactly, that one thing AMD historically has a very, very bad relationship with. Market dominance is another one like that. Part of their decision to axe GF and go fabless. We can applaud them producing on TSMC 7nm but let's face it... the volume isn't there nor the control over it.

This blade cuts both ways I hope you can see that. Intel's better marketing is AMD's lack of it. This is about spotting and capturing opportunities.
Yes. the blade cuts both ways. But fan-faring Intel's about being better in encoding and decoding is like saying the earth is flat. It is just marketing nothing more to me. AMD's marketing has been suffering and everyone knows that but it doesn't change anything in my eyes.
Posted on Reply
#23
95Viper
Stay on topic.
Stop the political comments.
This is a tech site, not your political sounding board.
Posted on Reply
#24
Chrispy_
Is it just me or does anything think that 8K with 4:2:2 Chroma subsampling is total garbage and inferior to 4K at 4:4:4

Get the image quality right first, increase resolution only if you can do so without degrading the image.
Posted on Reply
#25
Vya Domus
Chrispy_Is it just me or does anything think that 8K with 4:2:2 Chroma subsampling is total garbage and inferior to 4K at 4:4:4
Yep, this is done just to make the whole 8K thing feasible. For a full RGB 8K60 signal you would need to encode 8 GB/s worth of data, there is no way to do that in real time, especially on a CPU.
Posted on Reply
Add your own comment