AMDrone voice *NEW*: RENDERMAN DOESNT MATTER

if anything, all the stuff that Intel has made has stuck, and now it's a part of the baseline x64 ISA

meanwhile AMD's "innovations" all die out within a few years. Remember 3DNow?

en.wikipedia.org/wiki/3DNow!

>*NEW* VECTORIZATION DOESNT MATTER!

Yeah? Well, for me, it's ISPC.

>AVX512
>RENDERMAN
THEY sell software or hardware?
Is Nvidia card a update tickek for drivers?

A good architecture is more important than a set of instructions.
The latter is useless without the former.

AMDrone starter pack

>Performance doesn't matter
>BIOS doesn't matter
>CPU VCore doesn't matter
>FPS doesn't matter
>Games don't matter
>Software doesn't matter
>*OLD* Loos and toilets doesn't matter

>*NEW* ISA DOESNT MATTER

>intcels copying ayymd memes
sad

>Remember 3DNow?
I'm pretty sure most of the weird shit AMD added to Bulldozer is also gone now

So basically, AVX512 doesn't matter?

Don't forget
>clock speed doesn't matter
>overclocking doesn't matter
>advertised boost clocks don't matter

Imagine being an intcel lmao that's a BIG yikes from me dawg

The real problem with avx512 is that it is such a fragmented feature set. This isn't like with avx2 where if it was supported, you got all the instructions. Intel only ports a couple of avx512 instructions to desktop and then leaves the rest for server. Probably will prevent it from gaining any real traction in software for a while until it is better supported.

>Intel only ports a couple of avx512 instructions to desktop and then leaves the rest for server. Probably will prevent it from gaining any real traction in software for a while until it is better supported.
False

subsets that were first added in early xeons got improved on and put into new, better super-sets, and since then, it has been a linear implementation

4VNNIW from the server space is now VNNI, 4FMAPS is now IFMA, etc

Icelake is coming out, the first release of consumer-AVX512, and has everything implemented before it and more

VP2INTERSECT coming in TigerLake

linear progression here:
en.wikipedia.org/wiki/AVX-512#Instruction_set

There is also a documented behavior for its implementation, such as how having one subset implies that the other one is there. All subsets require AVX512F, AVX512BITALG implies, AVX512VL, etc.

AVX3.1 formally meaning "F + CD + ER + PF"
AVX512PF and AVX512ER were in server-space demands, but are no longer needed anymore.

AVX3.2 formally meaning "F + CD + BW + DQ + VL"

Attached: chrome_2019-08-03_12-16-14.png (1206x507, 29K)

Does your CPU support avx512?

I've had an X299 setup for a couple of years now and can count the number of non-benchmark AVX-512 loads I've run on no hands.

>p-patented vendor lock in vector instructions are important because muh nonfree software was bribed into using it.

Jesus fucking christ. I'm glad some compiler autismos are working on ways to transform intel specific vector instructions into different ones from different vendors, like altavec.

SIMD has been in x86 since the 90s

Looks like a mess tbqh.

It matters.
Right now IPC and cores are more important.
Not even Intel is able to handle those things simultaneously. Some of its chips have avx512 but none of them have cores.

Quick, name 5 popular programs that utilise AVX-512 and Renderman

Thought so.

>*NEW* price/performance doesn't matter

>run AVX512 workload
>clock speed drops to 1 GHz to avoid thermal meltdown
Thank you based Intel!

>drops to 1ghz
>does 20x the work

AMDrones hate any innovation until AMD has it and AMD is on the lead.

It's not strictly Intel technology, but Nvidia too.

>WHO NEEDS AN I7? I HAVE MY FX AND ITS GOOD ENOUGH!
>HA HA HA INTEL MY 3RD GEN RYZEN BEATS YOUR CPU FROM 2018 AND 2019
>WHO NEEDS A 2080 OR 2080TI? I HAVE MY 5700
>HA HA HA NVIDIA MY 10TH GEN RDNA BEATS YOUR GPU FROM 2020

>AMD of any year
>beating Intel of any year

>does 1.5x the work
ftfy

2 cpus with the same cinebench score

i7 1065G7 = 3.9 ghz
i7 6700k = 4.2 ghz

9% clockspeed difference with the same single core, so intel lied again about their ipc, meanwhile AMD has a 15% ipc gain

Attached: chrome_NPwx5dq1NO.png (608x472, 456K)

AVX good goyim!

Attached: .png (707x1252, 185K)

...that's not what we see on your pic

What do you use AVX 512 for?

Imagine being the retard that thinks cinebench runs the single core test at base clock
kys retard

DELID DIS NOW

Attached: 1527629778452.jpg (679x758, 54K)

parallel acceleration

>have a SINGLE thread using avx512
>whole cpu downclocks to almost half its base clock
nothing personnel, kid...

show the cpu frequency msrs on both chips when running the benchmark then

The PCU has assigned license levels to individual cores since skylake

Bait
The x series has 10+ cores. All with AVX2. Each core with 2 AVX512 units

ah right, then i'm assuming you're using at least linux 5.3, because that has special support to keep avx512-using threads time-pinned to certain cores.
avx512 workloads running on linux prior to 5.3 or windows (any) will migrate to random cores and still slow your cpu much more than they should. enjoy your unusable poz of a vector extension

It's faster than a Ryzen 2700 at 1.5 GHz

They'll neve reach the intard cope list

i9-9980XE has 18 cores.
2990wx has 32 cores. Almost 2x more.
The Zen2 TRs will have even more cores, up to 64 if we're lucky.
There's no point in having avx512 if you don't have cores to run it.

Cinebench runs singlecore on the boost clock, ice lake does 3.9 and the skylake i7 does 4.2, how dumb are you? go back to /v/ faggot

Imagine making an entire instruction set to run one program.

And its 6% slower than a 3700x at 1.5, not that its running at 1.5 because you are a retarded /v/ child that has no idea how cinebench runs single core benchmarks

>more SIMD shit
Remind me why are they making CPUs that pretend being GPUs again?

if the context switches to scalar or low current vector code, the core gets its original license level restored in < 2ms

your scheduler is retarded or your multithreading implementation is terrible if it's migrating threads that often regardless

Gamers continue to think simd is "wtf just do it on the GPU" and not specialized accelerated tasks like aes or bit arithmetic or horizontal addition or variable precious fma

>*NEW* SIMD doesn't matter
AMDrones embarass themselves

>per core through-put through vectorization vs "bro just fucking glue several weaker cores together"

>specialized accelerated tasks
Just do them on a GPU.

repeating yourself and resorting to personal attacks doesn't help your case friend

>multithreading doesn't matter!

Idiot.

If it's not a bottleneck you do it the regular way, if it's a bottleneck you use an accelerator(like a GPU), it's not rocket science.

ok retard
blog.cloudflare.com/on-the-dangers-of-intels-frequency-scaling/
lemire.me/blog/2018/04/19/by-how-much-does-avx-512-slow-down-your-cpu-a-first-experiment/
lemire.me/blog/2018/08/13/the-dangers-of-avx-512-throttling-myth-or-reality/
>If you do not require AVX-512 for some specific high performance tasks, I suggest you disable AVX-512 execution on your server or desktop, to avoid accidental AVX-512 throttling.
you pretty much a separate cpu for avx512-only tasks to make it worthwhile

Amd NIGGAS really in this thread trying to justify their wagelet purchase for their gaming and hentai machine

i9-9900k, the only remotely justified intel cpu, doesn't have avx512, what's your point?

Attached: incel cope.png (426x146, 35K)

>*NEW* SIMD doesn't matter

Gamers really think that GPUs are magical devices and that things like AVX shouldn't exist "be-cuz gee pee yoo"

They really think that someone should just spin up an entire opencl or vulkan or Cuda context and then upload and download work through the pcie lanes over just having it sit optimally in the cpu pipeline

sirs ples buy needful intel cpu thnx

Attached: poointhecpoo1.jpg (700x466, 37K)

AMD fan boys again moving-the-goal-posting whenever they witness intel innovation

AMDrones literally think simd doesn't matter anymore but once amd has it they're gonna suddenly act like AVX512 always mattered and is actually great.

I mean a %13 downclock for a x8 throughput sounds like a win to me

Besides, avx 512 penalties are with using the zmm registers
You can use the new and fast AVX512 features on ymm and xmm registers and not get any of the throttling issues.

You can have roads with various speed limits and various amounts of lanes. A conventional CPU is like a raceway with an ungodly speed limit and few lanes, a GPU is like some 30 lane monstrosity in China where you're lucky to go faster than a bicycle. SIMD is somewhat of a middle ground. Different algorithms get better performance on different approaches.

Zen didn't even have full rate AVX2 until now but it still did very well on rendering and encoding. AVX-512 won't make as much of a difference as 256 bit vs 128 bit on the last generation. Vector extensions are a nice feature for performance but they're hardly the only factor.

Intel only has AVX512 in the product niche where it's annihilated by AMD yet you keep spinning it as a massive win for Intel. "Some of our CPUs have this feature so AMD automatically sucks LMAO" -You

Inteldiots can't meme

Zen had half the throughput because it had a botched avx 2 implementation making intel twice as fast. Video encoding for example is totally faster on Intel due to avx and AVX512.
Avx512 is not just a wider register. It's a total refresh of the X86 SIMD stack that is not trivial to implement like AVX2 was so amd won't be having it any time soon, and icelake is going to burn that hole that much deeper with the new laptops coming out

>I mean a %13 downclock for a x8 throughput sounds like a win to me
on full-core that's closer to 50% and also differs between skus
though your last point i definitely agree with, masking for every operation is a really good feature to have

>so amd won't be having it any time soon
Not before Intel itself implements it consistently lol. How many variants are there at the moment?

that's pajeet code psychotically sprinkled with vector instructions, of course it performs like shit. i'm not really seeing the relevance to thread migration

>"an i7"
>but everything else is generation based
this is why you'll never make it past helpdesk in your career

>copying memes
embarrassing

>VPCLMULQDQ
Just reading these abbreviations makes me vomit, nobody is going to bother with these without explicitly being paid by Intel.

Zen didn't have a botched AVX2 implementation. It had a 128 bit SIMD unit. AVX2 support was only for compatibility. Obviously per core and per clock Skylake was faster than Zen on AVX2 tasks. The difference is Intel has never sold at per core and per clock price parity with AMD. For the majority of the product stack you were able to make up for the lower level of per core floating point throughput with more cores. Of course Zen 2 has full rate AVX2 so now there's almost no case for the Skylake lineup.

...

twitter.com/x86instructions

Wait I didn't browsed Jow Forums in theses past month, what happened to AMDchad and why are intcel rising up?

Summer is ending, all the adults are coming back from vacation so all the kids that got their mom to buy them cheap computer parts are the minority now

No you don't /v/fag

Lol, all these butthurt fanboys in this thread.
Can you be any more pathetic? Specially the coping Intel ones, it's sad.

Attached: file.png (1672x710, 221K)

Why does AVX2 exist when SSE4.2 was already a 256bit SIMD? Just useless segmentation?

What the fuck else is SIMD for you dumb ass

Does the 9900k support AVX512?

Nope,
en.wikichip.org/wiki/intel/core_i9/i9-9900k
Its the x series one that does
en.wikichip.org/wiki/intel/core_i9/i9-9900x

Here's all the AVX512 devices en.wikichip.org/wiki/Special:SearchByProperty/:has-20advanced-20vector-20extensions-20512/true

No it doesn't it's limited to 15W, the 3.9 is probably just for a few seconds.

rendering is faster on a 5700xt than a 2080ti

objectively false

>parallel acceleration
and I thought you don't need more cores than what intel offers since 2008

Attached: bongson.jpg (1000x563, 51K)

Isn't GPU rendering 1000x faster than CPU rendering?

cpu X good, cpu Y bad
when will those threads fucking die