Even AMD knows it

Even AMD knows it.
amd.com/en-us/innovations/software-technologies/zen-cpu

Other urls found in this thread:

archive.is/RcHob
archive.is/vA3EZ
twitter.com/SFWRedditVideos

Zen is a power play for becoming the CPU of datacenters with more cores in less space using less electricity. If anyone's hoping it will beat Jewtel in per-core performance they're going to be disappointed.

Knows what? If they come out and say it's faster than Intel, shitty review sites will just run Cinebench R11, SYSMARK, and other bullshit and ICC compiled programs and talk about how awful Zen is because it's slower in Cinebench R11 and some benchmarks Intel shit up so bad everyone else quit development on them.

You sound like you're part of the $300M Intel Diversity Technology® initiative :^)

archive.is or don't post links

There's nothing here that says anything worth "even they know"

did you expect AMD to rebase Intel in a single generation?

It took Intel many years to regain ground from the K7/K8 AMD masterpiece.

Not like I'm going to buy backdoored CPU just because they are AMD, anyway.

sage because not risc-v

Does this help: archive.is/RcHob

So, eh, any significant news apart from the old “40% better”? Like, what's single tread performance? How's the dev team going?

We should already have seen Zen in real tests, according to AMD's old timelines.

Let AMD make their own compiler if they're so concerned about ICC fairness, but I'm going to use the fastest compiler before I pay for a more expensive processor so I want ICC benchmarks. I don't give a shit about gcc/vc++ for perf.

1/10

...

but what if it does

Single-thread performance got kicked up a notch in priority.

They'd have been shouting it from the rooftops for months if that were true. Instead, every presentation is about value. Doesn't take a genius, user.

k

I would like to see test results too. If zen live up to the memes then 40% better will make it basically haswell but with more cores.

If zen gives me basically Haswell but with 8 cores at the price of newest intel with 4 cores, I choose zen. Would be very comfy for some heavily parallel coding I would like to try.

Don't forget idling at 5 watts. That's basically in line with a scaled-up Atom.

Thrusting bechmarks
Talking about benchmarks

x86 is ugly like shit.
I think sse2 are okay.
AVX is for morons.
You don't need multiple operations on
at the same time when you don't have multiple memory controllers.
You can't do anything useful with 16 fp operations at time when you can't fetch
16 floating point from unaligned locations at the same time.
Long life to APUs.

APUs with HSA*

Why would your data be unaligned?

I'm impressed how wrong you managed to make a single post.

This. The FUD is already starting.
archive.is/vA3EZ


They seem to be complaining that both CPUs are clocked identically and AMD wins, though this is exactly opposite what they were complaining about when AMD was winning with higher clocks.

AMD have multiple compilers. Probably the most interesting is HCC, a sepples compiler that can automatically unroll loops into GPU accelerated code.

COMFY
O
M
F
Y

They're complaining that AMD underclocked the competition to avoid comparing performance, making their test pretty useless. No one cares about IPC, that's AMD marketting their chip on /potential/. And you only market potential when you've got nothing today. It's shaping up to be a spectacular flop.

Why don't they pay the benchmark companies to make a build with it, if ICC is really so unfair? If they knew they were better I'd think that would be a priority.

Look, Satan, I get that you think every company ever just has trillions of dollars to throw around, but have you ever considered that Intel can just pay higher?

The reason you'd want SIMD operations, particularly with x86, is because of how much power their huge instruction decode unit is.

So in that situation, only decoding 1 instruction for 16 ops is a win.

If they're getting destroyed in the marketplace because they believe there are no fair benchmarks or compilers and they always look slow and garbage because of it, making fair benchmarks and compilers would be some mission critical priority shit, don't you think? Cloning a CPU benchmark tool would only take a couple million.
But they don't. Why is that? Maybe they know they are actually slow and garbage and would rather be able to claim the tests are unfair than prove they were fair?

I'm not saying that. AMD's issues stem from Intel paying off OEMs in the mid 2000s coupled with gambling on a rise of multithreaded software and continued transistor scaling.

The datacenter loves lots of cores, but they buy Xeons because AMD has shit performance.

Read what happened with SYSMARK and BAPCO. Basically, Intel, Nvidia, AMD, ARM, Via, and a few other companies (IIRC) got together to make a fair and balanced benchmark suite called SYSMARK.

Intel managed to completely shit up the benchmark so bad, everyone else besides Intel quit BAPCO. Intel somehow managed to pull shit like to make the GPU score not matter in the overall system score. As you can guess, Nvidia wasn't very pleased. AMD left because Intel managed to make AMD CPUs look worse than they are. ARM and Via left for the same reasons.

It's basically impossible to create a balanced benchmark, which is exactly why AMD used Blender and probably a somewhat neutral compiler like GCC or MSVC.

As for running Zen at 3ghz, I have a feeling AMD is sandbagging final clocks. They have no reason to tell Intel what they will release Zen at if it's so competitive with Broadwell-E. Intel would just increase clocks a little bit and then win if they knew final clocks. AMD is going to let Intel release Broadwell-E and then try and get Zen to run a little bit higher frequency so they can win more.


The caveat is some software has per core licensing, so many weak cores costs you more for the software.

Dear Reddit users who have come to Holla Forums, please remember that "sage" is not a downvote and shouldn't be used as such nor should "sage" be in any field except "Email".

Thank you.