Single-core performance of Intel's Sunny Cove chips Surface - Shows Big IPC Increase

I love your mindshare. Vaporware counters unreleased product.
One likely does not exist and you know very little, borderline nothing about other.

And then people are surprised that AMD does not play way they want them to.

 

Any shrink provides efficiency and performance gains. Intel has achieved a lot with their 14nm process, and I'm certain that 10nm and 7nm will provide great increases.

However, the main question here is - WHEN and at what price.

 

Intel must show something to slow down the zen2 hype. Release date and pricing unknown. Hah, most of us will already rock a ryzen cpus until these see the light of day.

 

When was the last time Intels pre-release performance didn't match up with the final product? They have a far better record than AMD in this department, yet I can't recall you ever adding a disclaimer to news about their performance numbers. Come to think of it, i also don't recall you making product announcement articles from nV or Intel pinned at the top of the forum either. At least try and show a little objectivity

 

While Hilbert already pointed out the need for salt it's probably important to drive home just how much is needed;

• CPU-Z isn't a reliable benchmark. Even if these numbers ended up being correct it may not amount to anything.
• At face value this puts Sunny Cove at a 44% higher IPC, compared to Coffe Lake refresh. That's not just similar to AMDs uplift going from Bulldozer to Zen, it's directly contradicting Intel's own (likely optimistic) claims.

I have no doubt Intel will have an answer to Zen 2 but I doubt it'll be as early as next year, and frankly I'd be very surprised to see another 40+% IPC uplift over a single generation on either x86 architecture. Aside from very narrow applications obviously, like another doubling of AVX throughput or dedicated instructions for en/decoding etc.

 

Last time their figures matched was last time they used term IPC as instructions per clock/cycle and not as instructions per core.

Since then you have to somewhat convert their "IPC" figures to real world ones. That's if you know at what clock they tested. Should be noted that their comparison with high 18% "IPC" gain was against skylake chip (where even 6700K had boost of 4.2GHz only) likely equipped with as slow memory (2133MHz) as they officially support for given chip.

Memory difference easily explains those up to 40% higher results too.

And it should be noted that this skylake comparison was made for very specific purpose since skylake is officially "Discontinued".

 

Will believe numbers with real reviews, not until. AMD will be cheaper, and therefore more likely to get my money.

 

Let's wait and see real benchmarks, until that time everything is just talk.

 

Know or presume/believe. Make your mind. If you lie to yourself, and spread it, then you lie to us!
I do not like liars. But I like fools even less. Because liars do that at least for purpose.

 

I say let the turing refresh come. Let the amd release navi 20. Let the intel compete with zen2.

That will all benefit us so lets be happy instead arguing.

 

I remember, maybe I'm wrong, that 40% improvement will happen on AI related workloads.

18% it's the best score it will achieve in usual workloads, but still impressive performance uplift.

It will came first to mobile chips, close to Zen 2 APU, and it will be great, because we'll have better CPUs and integrated GPUs around, if vendors offer good quality AMD laptops and Intel don't play dirt, as it does whenever is on trouble.

Then, SunnyCove will come glued to servers, and later to desktop, I think

 

Objectivity works with official release notes and confirmed numbers. Being sceptical about something some guy puts together a few lines with some CPU names and a few numbers, posting it in some forum on the nets does not make it something to even be considered worth noting, and doubting this is in no way against being objective. The "grain of salt" here is that it's not an official chanel, it's not an official rep, and so on. Read again before complaining:

Doubting this is more logical than blind belief everything in that "table" (which it isn't even or the names and numbers would be lined up better).
It would also be logical and sensible to doubt any of this until there's true benchmarks comparing the products. When that stuff someday lands in test areas outside of any marketing department and any NDAs.

 

Intel trolling us again.

 

Link, please

 

Although I'm not convinced about these leaks, healthy competition is always good for consumers.
Intel has been lazy for too long, they will have to spit something decent eventually.

 

You are wrong right there. RDNA's performance per transistor per clock is better than Turing's.
This means AMD is capable to make GPU of same size as nVidia which will perform better at same clock. Sadly, AMD does what nVidia did in past.
They make smaller GPU, clock it higher. This means that while performance is competitive and AMD pays less for each chip, they eat more than needed.

If AMD took that RX 5700 XT and clocked it to around 1500~1600 MHz only. Then price it $400, everyone would go crazy upon 1st reviews showing that they clock to 1900MHz and more.
But realistically AMD must have considered that OC headroom is just too high. And luckily for us it looks like RDNA boosts in similar way as Zen.

On other hand, we do not have 1st real review. And we do not know if launch drivers will be same as those used last week.
What I expect is that I'll no longer need to use monitor's built-in edge detect sharpening which is poor to begin with. Since AMD promised ability to run post-process shader similar to good old smartshader effects. Which will be fun.

I can imagine streaming community where pronounced details prior encoding on AMD's platform will result in higher IQ stream.
I have such thing done in OBS via doubling resolution, running edge detect sharpening and then downsampling back. That increases quality of streamed material a lot. But computational power needed is far from small.
Few times I had to explain to people that their nVidia card does not have washed out details in same game and show them what I have set up in OBS as proof.
But once those drivers are out...

 

The same CPU-Z that changed its benchmark implementation because Zen was to good at it? no thank you, I do not care about it. The only synthetic benchmarks I care are AIDA64 tests since they optimize then as much as possible for every uArch.

 

I still don't agree with this because it's not an apple's to apples comparison. At minimum it's a 25% die space savings by ripping out Tensor (which is largely unnecessary) and if you're not then you need to find a way to average in INT4/8 & RT workloads into the "performance" part of that statement.

Some of us will, including me, but it's going to take a long time to dethrone Intel and Intel knows that. They have a ton of new tech in the works and it's all going to start hitting over the next few years. Intel's R&D per quarter is more than AMD's entire revenue for a year. The silicon phontonics, foveros, decoupling their architecture from manufacturing process - the entire new engineering team they built in the last year. It's all going to come together and it's going to be up to AMD to figure out a way to compete with it. I think AMD's only option is to continue doing what it's been doing and just take massive risks - which they seem to acknowledge because they basically said it's their only option at the tech conference last week.

 

I'm not trying to bash AMD here but what do you refer to mate? Which workload? Benchmarks? Link? I'd like to read up.