Threadripper Pro 5000 CPUs Will Hit The DIY Market H2.

Hi.

Quote: 'AMD said that these Zen 3 processors will become more widely available, expanding to "leading system integrators worldwide." However, PC DIY enthusiasts longing to put together a Threadripper Pro 5000 build will have to wait until sometime later in the year

From:

https://www.tomshardware.com/news/ryzen ... -this-year

And

https://community.amd.com/t5/business/d ... 57479242=1

Regards Carsten.

It's a shame these have been nearly exclusive to Lenovo for some time. I think by the time they hit the consumer market, Zen 4 will have significantly lessened the appeal for most users.

Hopefully 40 series GPU from Nvidia will close the gaps they have today in hardware acceleration.

Today Intel 11/12th gen is the only option (other than Mac M1) for h.265 422. So if you need the extra cores or PCIE of Threadripper, you know have to accept poor decode speed.

Maybe by then Intel GPS's will also be available for decode.

Maybe by then Intel GPS's will also be available for decode.

These could be interesting too. I don't think anyone expect them to actually compete with NV in compute. So you will still need your 3080\90.

Buying a $75 Pcie 1x Intel card to get good decode would be something i could see. A $300 16x card seems a harder sell.

Don't make me waste a 16x slot, or force my motherboard to run in 8x/8x mode just for the decode card.

Also Resolve needs to be smart enough to use the Intel card for Decode, but not try to run NR on it.

I think most of the leaked tests (using alpha or beta drivers) are putting the 770 cards in line with the 3070. But the 770 is supposed to come with 16GB of memory and can share work with Quicksync on Intel systems.

Autumn is when Sapphire lake hedt aka Fishhawk Falls is supposed to launch. Still no word if that will include QS on board.

A while back one of the Resolve releases mentioned multiple Intel GPU support. Wondered if that was CPu QS plus new Intel GPU or more than one discrete Intel GPU . It would be good to have Intel and NVIDIA on board if possible to get hardware decode/encode as well as CUDA. Matching 3070 would be fine for me as it would be a lot more than my 1080TI. Combination may work well.

It was QS from the onboard integrated GPU. I think Intel calls it hyper encode. They put out some videos on YT claiming I think up to a 50% speed boost. But that's likely not an average improvement.

I think Intel calls it hyper encode.

Hi.

Here is a link to Intel CES 2022 Livestream:

From (23:35 to 24:56) it was demoed to be 1.4 times faster in Da Vinchi Resolve with Hyper Encode.

Quote: 'Hyper Encode, Intel’s second feature, is an interesting one. It allows the system to leverage all of the graphics processors in the chassis to handle GPU workloads. If you’re encoding with Intel’s OpenVPL API, which is part of OpenVINO, a software designer can write code that will spread encoding across both the dGPU and iGPU. This is a smart way to make use of iGPU silicon in systems with discrete cards.'

From: https://www.extremetech.com/computing/3 ... =computing

Regards Carsten.

Hi.

Here is more information on the first Intel Arc Graphics cards:



From: https://wccftech.com/exclusive-intel-ar ... sitioning/

As I see it, is the 'Intel Deep Link Hyper Encode' and the 'Intel Deep link Hyper Compute' two different lines. So may be will it be possible for BMD just to use the 'Intel Deep Link Hyper Encode' part, togetter with Graphics cards from nVidea or AMD as others suggested?

I also read that it will support both AV1 Encode and Decode. Let hope both are in Hardware?

Regards Carsten.

AV1 decode seems like it was added to the XE laptop chips this generation. AV1 encode is the new one. Yes in hardware.

The Wiki shows the Arc QS is also slated for Meteor Lake, Arrow Lake

Hi.

Here is more information on the first Intel Arc Graphics cards:

Sucks they are making them 8x PCIE cards. that means they will steal an entire 16 slot when surely they don't need the bandwidth.

And if someone wants to use the cheaper one just for decode, it will force a lot of motherboards into 8x/8x mode stealing potential bandwidth from their good GPU.

Hi.

Here is a new 'AMD Threadripper PRO 5000 WX-Series Content Creation Review'

https://www.pugetsystems.com/labs/artic ... view-2338/

Regards Carsten.

...
And if someone wants to use the cheaper one just for decode, it will force a lot of motherboards into 8x/8x mode stealing potential bandwidth from their good GPU.

Hmmm... I am curious about this, as I am considering an A380 card to supplement an intended RX 7900 XT purchase *if* 10-bit 4:2:2 decode is not supported in the RX 7900 XT.

My question is: will the data rate pushed to the GPU ever exceed the multicam playback data rate? For example, if 10 cameras of 30fps UHD BRAW 3:1 (127MByte/s * 10 = 1.27GByte/s) are playing back, is that 1.27GB the maximum data rate that should be expected to be pushed to the GPU?

BRAW isn't the issue nor do I think it'll be helped by an HEVC decoder . The decoder is important for cameras that output HEVC 4.2.2.

My question is: will the data rate pushed to the GPU ever exceed the multicam playback data rate? For example, if 10 cameras of 30fps UHD BRAW 3:1 (127MByte/s * 10 = 1.27GByte/s) are playing back, is that 1.27GB the maximum data rate that should be expected to be pushed to the GPU?

Depends on braw decoder internal mechanics, whether there is some kind of cpu based decoding step before frame data is pushed to gpu or not. If there isn't, data rate should be more or less similar. If there is, it can only be measured using some gpu profiler.

BRAW isn't the issue nor do I think it'll be helped by an HEVC decoder . The decoder is important for cameras that output HEVC 4.2.2.

Yes, I agree BRAW isn't the issue, and it seems I was conflating things. However, I was trying to ask two things within one post. The Intel A380 possibility is related to my Panasonic GH5S 150Mbps 10-bit 4:2:2, recording option...which seems to be only decoded in hardware by QuickSync on Intel CPUs and within Apple M1/M2 products.

The BRAW question was designed to understand whether or not nodes (like color adjustment and noise reduction) will drive any more bandwidth demand going into the GPU as more complex manipulations are made to frames within the project. If not, then the PCIe lane count at "only" Gen4 x8 is not a concern. It's the same question if H.264 or H.265 camera content is being manipulated.

The Puget Systems A770/ A750 review is up today:
https://www.pugetsystems.com/labs/artic ... view-2376/

I read through it. The article answered many of my questions, and mostly confirmed my expectations. I'm planning to buy *two* A770 cards because I have hope that the Hyper Encoding function will work between two dGPU's, rather than the noted iGPU/dGPU combo in the article.

Right now, the A770's are sold out everywhere I've checked.