Threadripper Interest Poll

[Sacriledgy]

1. I didn't buy Thread Ripper because its too expensive and i don't need it anyway.

2. 16C/32T Thread Ripper is set to be priced at $850

3. Epyc is enterprise grade hardware, equivalent to Xeon's. Who on earth would buy 32C/64T for day to day task? ?

[porina]

I have a 1700 and 1600, and that's enough AMD cores for now. With vague assumptions, I'd hope low clock 16c TR would be comparable to 2x 1700, however that in itself isn't enough to make me buy it. Once the numbers get to that sort of ball park, I need a CPU that will cope with everything and that still means Intel with much higher AVX2 performance, and AVX512 for potentially even more once support builds up. Note Google cloud service already offers Skylake Xeons and people have been testing AVX512 on it. Early days yet...

[done12many2]

10 minutes ago, porina said:

I have a 1700 and 1600, and that's enough AMD cores for now. With vague assumptions, I'd hope low clock 16c TR would be comparable to 2x 1700, however that in itself isn't enough to make me buy it. Once the numbers get to that sort of ball park, I need a CPU that will cope with everything and that still means Intel with much higher AVX2 performance, and AVX512 for potentially even more once support builds up. Note Google cloud service already offers Skylake Xeons and people have been testing AVX512 on it. Early days yet...

 

I'm fairly confident that your 4 core 6700k will encoded H.265 as fast if not faster then your 8 core 1700 simply because of the AVX deficit in Ryzen.  Unfortunately, there's no BIOS update or memory improvement that can fix that kind of issue.  

[PCGuy_5960]

1 minute ago, done12many2 said:

I'm fairly confident that your 4 core 6700k will encoded H.265 as fast if not faster then your 8 core 1700 simply because of the AVX deficit in Ryzen.  Unfortunately, there's no BIOS update or memory improvement that can fix that kind of issue.  

AFAIK, this is an issue with the Zen architecture, so it is present in every CPU, even in the EPYC CPUs

[porina]

3 minutes ago, done12many2 said:

I'm fairly confident that your 4 core 6700k will encoded H.265 as fast if not faster then your 8 core 1700 simply because of the AVX deficit in Ryzen.  Unfortunately, there's no BIOS update or memory improvement that can fix that kind of issue.  

I work on the assumption that in this type of use case I would need double the Ryzen cores than I would Skylake cores for a comparable performance, thus 6700k and 1700 are good comparisons due to pricing. Then Intel would still win from higher clocks. Note HT/SMT doesn't help in this scenario, so even my OC 6600k would beat the OC 1700. Of course, for everything else, Ryzen is fast which is why I have two of them.

[porina]

3 minutes ago, PCGuy_5960 said:

AFAIK, this is an issue with the Zen architecture, so it is present in every CPU, even in the EPYC CPUs

It is not an issue, it was a design decision made by AMD. They decided most users didn't need it and they gain a saving in die size, manufacturing cost and headline TDP. So "most users" are happy with this decision.

[done12many2]

8 minutes ago, porina said:

I work on the assumption that in this type of use case I would need double the Ryzen cores than I would Skylake cores for a comparable performance, thus 6700k and 1700 are good comparisons due to pricing. Then Intel would still win from higher clocks. Note HT/SMT doesn't help in this scenario, so even my OC 6600k would beat the OC 1700. Of course, for everything else, Ryzen is fast which is why I have two of them.

 

My plan was to use the Skylake X rig as my daily rig and the Threadripper as my backup / background tasks rig.  One of those tasks includes H.265 via Handbrake.  It bugs me to think that if I chose H.265 as an encoder (best quality / file size) that it will take more then twice as long on Threadripper as it would have on an Intel chip with the same amount of cores.  A chip marketed as a work horse should not be limited in this capacity.  This in turn would force me to either accept the delay, run it on my daily rig, or skip Threadripper altogether.  

 

6 minutes ago, porina said:

It is not an issue, it was a design decision made by AMD. They decided most users didn't need it and they gain a saving in die size, manufacturing cost and headline TDP. So "most users" are happy with this decision.

 

Agreed.  Definitely a calculated decision, but one that impacts you more in HEDT then it does in mainstream. 

 

[PCGuy_5960]

2 minutes ago, done12many2 said:

Agreed.  Definitely a calculated decision, but one that impacts you more in HEDT then it does in mainstream. 

Especially in server..... I am not sure if people looking into buying EPYC CPUs will be happy with this decision

[DrMikeNZ]

6 hours ago, done12many2 said:

One of those tasks includes H.265 via Handbrake.  It bugs me to think that if I chose H.265 as an encoder (best quality / file size) that it will take more then twice as long on Threadripper as it would have on an Intel chip with the same amount of cores.

This will depend on a large number of factors, like the resolution of the file being encoded, video content, what settings you use, particularly denoise filtering etc.

At 1080p the 6900K was typically 100% faster than my 1700 and at 4K the 6900K was only ~50% faster with NLMeans denoise filtering, without filters there was only a ~30% difference at 1080p and ~15% difference at 4K.

Large differences in relative performance between the processors were observed depending on the video content (eg. complex gritty action vs slow paced animated films), for example at 1080p w/ NLMeans, some files were consistently only 60% faster on the 6900K over the 1700 while others were up to 110% faster using the same settings.

 

It would be nice to have an easy answer, but there are too many factors with h.265 encoding to draw conclusions based on a few tests to cover everyone's requirements. As I mostly deal with 1080p files with NLMeans, it unfortunately falls right into Ryzen's weakest performance based on my tests. My 6900K system cost twice as much as my 1700 system, so I am not as concerned about half the performance.

Depending on what types of files you are working with and the settings you need to use, your mileage may vary quite significantly though.

 

Do you have a preset and test file(s) that would represent your workload that you want me to queue up and test on my 6900K and 1700 for you?

[Simon771]

under 500$ ? Nope

Best case scenario it will be around 800$.

[Dash Lambda]

I'm honestly quite surprised this many people are saying they plan to buy it. I've gotta wonder if they're exaggerating, if it's actually that attractive, or if my sample group is just skewed.

[bimmerman]

I would probably buy Ryzen over threadripper, though there's a part of me that wants to support HEDT alternatives rather than Intel.

 

Then again, I don't have a compelling reason to upgrade from my 6c/12t X58 Xeon other than Shiny Object Syndrome. I don't feel there is a lack of performance with what I have. It's fairly pathetic that my ancient platform is still good enough for enjoyable gaming.