Ryzen's halved 256bit AVX2 throughput

[itsmydamnation]

TomCruiseLaugh.jpg

And you would be wrong, as illustrated by The Stilt right around the time of demo.

What does that even mean? he compiled from source...... So whats your point exactly?

 

[Genx87]

Paid content creators time is money. The half the cost Ryzen means very little if it takes hours longer to render a final cut. The cost of the CPU is very tiny to the overall cost of the creator(salary), workstation, and software. For avg users this difference probably won't mean much.

 

[Kenmitch]

Paid content creators time is money. The half the cost Ryzen means very little if it takes hours longer to render a final cut. The cost of the CPU is very tiny to the overall cost of the creator(salary), workstation, and software. For avg users this difference probably won't mean much.

Time is money is very true. But for the big boys in the game I doubt they would really be interested in Ryzen....Other than a peek at what Naples will bring to the table. Now if Naples delivers and it's half the price of a comparable Intel rig what would you buy?

 

[parvadomus]

Saying the length of the workload matters is bullshit. Probably the "heavier" tests are having different memory bandwidth requirements (not only memory channels, but cache performance) that is helping intel cpus to pull ahead. But I doubt this happens in all cases.

 

[dacostafilipe]

Those claiming that "GPU" will replace software video encoding clearly have no knowledge of the domain.

Having worked with a big media group in Europe, I completely disagree with you.

GPUs will replace most of the transcoding work in the near future, except for some special cases where it will be encoded on big workstations or on "normal" CPU servers (ex: transcoding to TIF/DPX images sequence for the silver screen)

Post-processing is an other story ...

 

[Genx87]

Time is money is very true. But for the big boys in the game I doubt they would really be interested in Ryzen....Other than a peek at what Naples will bring to the table. Now if Naples delivers and it's half the price of a comparable Intel rig what would you buy?

In my situation CPU power has never been a problem. My bottlenecks are typically network\disc related. The other underlying components are more important to me than processing power. If a Ryzen\Naples box comes with quality nics, HBAs, ram configurations ect then yes, I would consider it.

 

[Atari2600]

Paid content creators time is money. The half the cost Ryzen means very little if it takes hours longer to render a final cut. The cost of the CPU is very tiny to the overall cost of the creator(salary), workstation, and software. For avg users this difference probably won't mean much.

You obviously "do" work. You might even work in a very small business or your own business where those making decisions understand this. Lucky you!

Unfortunately in any kind of medium-large company, the decision makers are completely removed from the actual value-added work and only see X costs $$ and Y costs $$$ - they typically don't have the wit to see which will generate best return.



<rantmode=TRUE>

For instance (and this is a real pet hate of mine)... monitors. F**king monitors.

An engineers time is charged at something like £40-100/hr (safety and nuclear can be far above this). An engineer is paid at something like £25-50/hr.

If your engineer was to be just 5% more productive, then over the course of a 1900 hr year, you generate an extra 95 hrs productivity. Taking the upper and lower bounds of the above that is between £1400-4750 per year in value added.

Now, do you think if you went from using 2x sh!tty 20" Dell at 1280x1024 res to using 2x 40" 4k screens (at £480 a pop), you'd be just 5% more efficient? Would you f**k. You'd be upwards of 20% more efficient - maybe scaling beyond 100% for some tasks. Of course, you could go somewhere in between, maybe UHD 32" monitors - enough to get two full program instances displayed side by side (i.e. excel+word) per monitor.

<rantmode=FALSE>

 

[Carfax83]

I do not know what "rendering" means in this thread, but many DSP and HPC software are already using AVX. The FFTW library is a common building block in all manner of audio and video DSP, and it has supported AVX+FMA for the longest time. Anyone that has run prime95 (FFT) knows how important SIMD width is to dense compute loads.

Reading the reviews for Ryzen made me realize how much more adopted AVX/AVX2 has become, just over the past year or so. I hope this trend continues, as AVX+FMA/AVX2 can provide very large performance gain when done properly.

Those claiming that "GPU" will replace software video encoding clearly have no knowledge of the domain. Real-time streaming has always been run on ASICs (these days, Intel Xeon E3 QuickSync accelerators), but high quality encoding for offline is a strictly CPU workload. This is not going to change either, as video encoding is a highly serial workload, where the results of a DSP block (e.g. DCT) are immediately fed into highly branching mode decision logic.

I am far from an expert in encoding, but from what I've discerned, this paragraph certainly seems to ratify what I've read so far about the virtues of CPU vs GPU encoding.

As for AMD's future plans in Zen 2, they will likely add an AVX-256 unit. I doubt we will see AVX-512 units though, and it may not matter as AVX-512 is mainly desirable for compatibility with Xeon Phi. Intel is unlikely to ever bring AVX-512 to consumer platforms, due to the die area cost.

I think we will see AVX-512 in consumer chips in the HEDT or 'E' variants eventually, since those parts typically have a higher TDP threshold than regular desktop parts. Also with continued die shrinks and possible material enhancements in the future, then I can definitely see this happening.

 

[Carfax83]

Sure. But as 256 bit execution is very power consuming Intel cpu's will throttle to their rated TDP. Without 256 bit ops 140W TDP Intel chips consumes only ~90W and that even with higher voltage and clock with boost states.

Without 256bit ops Intel CPU's could probably have easy +1GHz clock headroom, trading that for 256 bit ops which are almost useless for most workloads is just giveaway for competitors.

It's a tradeoff. Did you see the comparison benchmarks between AVX2 enabled CPUs, and CPUs without AVX2? It's a slaughter. AVX2 and wide SIMDs have their uses in certain workloads that benefit greatly from increased data parallelism. So ultimately, the tradeoff is worth it.

 

[Carfax83]

This is the most recent comparison I could find between CPU and GPU encoding.

Comparing NVENC (which is not GPU powered) against the CPU, you see MUCH faster encoding times, at the expense of both quality and file size. Even at the highest quality setting, NVENC does not match the CPU in terms of image quality, despite having a higher bitrate.

 

[Carfax83]

Having worked with a big media group in Europe, I completely disagree with you.

GPUs will replace most of the transcoding work in the near future, except for some special cases where it will be encoded on big workstations or on "normal" CPU servers (ex: transcoding to TIF/DPX images sequence for the silver screen)

Intel's Quick Sync and NVidia's NVENC are not GPU powered. They both use dedicated media hardware. From what I recall, explicit GPU powered encoding always gave terrible results..

 

[tamz_msc]

In the context of scientific computing, AVX2 is not the one-size-fits-all solution that is the answer to every constraint that limits computational ability.

When it comes to Physics, different people have different requirements:

• HEP guys use packages optimized for AVX2
• CFD/MHD guys use mixed CPU/GPU acceleration
• Numerical Relativity is all about GPU computing(OpenCL/CUDA

So, it's a wrong view that the forward march of AVX2, AVX512 etc. will be of relevance in every application out there. Is it a nice thing to have? Yes. Does it mean that AMD playing catch-up to support these specialized instruction sets is a drawback? Not necessarily.

Why?

Because in the race for copying instructions every time someone proposes a new one has led to unhealthy competition, and could be a headache for software developers in the future.