Intel Announces Sunny Cove Brings 18% IPC Improvement

Price is irrelevant. Xeons performance was crippled by AMD in its 'demo', not strange that AMD tried to hide that they played dirty and didn't give us configurations of its 'demo'.

Intel favored game? Do you really mean that Intel is better for gaming?

No it's price point orientated, AMD is throwing 64c at Intel's 28 price point and consumers are loving it.

You failed to answer the question, where is the GPU limit, PUBG is not a GPU limited game. While most of us remain cautious to real world performance, at least we can draw our own logical conclusions.

You seem willfully ignorant to how slowly but surely the market is changing, and that is a good thing, except for Intel.
 
If AMD hid it you don't know what it was ten you come to a conclusion that is not based on facts but rather on speculation ...

It is a fact that Xeons did run 30% slower in their demo.

Juanrga lacks the logic to think like that. He is a timing fanboy, and it is fun to watch him spin the same damage control over and over.

What is his proof that AMD gimped the test? They just enabled security checks that Intel has off due to their core integrity problems. Intel's test is more unrealistic so I believe a live demo over Intel running an internal bench.

Damage control is what someone I know did when Zen was very different to what his 'sources' leaked him. I remember him invented the crazy theory that AMD wasn't selling the real Zen, but "engineering samples rebranded" :D

The difference was that Intel enabled AVX512 for the demo.

No it's price point orientated, AMD is throwing 64c at Intel's 28 price point and consumers are loving it.

You failed to answer the question, where is the GPU limit, PUBG is not a GPU limited game. While most of us remain cautious to real world performance, at least we can draw our own logical conclusions.

You seem willfully ignorant to how slowly but surely the market is changing, and that is a good thing, except for Intel.

I remark again that pricing is irrelevant to AMD crippling performance of competitors. I didn't fail to answer any question. I did a remark and you misinterpreted completely what I said.

I remember perfectly the exagerated claims made about how Naples was going to change everything, and Intel was in trouble, reality was that Intel not only keep the datacenter business, but even increased it.

Zen only got momentum because Intel 10nm fiasco, but Sunny Cove is already here and not only increases the IPC gap with AMD, but brings us a true die shrink.
 
Last edited:
Zen only got momentum because Intel 10nm fiasco, but Sunny Cove is already here and not only increases the IPC gap with AMD, but brings us a true die shrink.

And if Sunny Cove is the messiah you and others want people to believe it is it would be all over the product stack instead of in the low power and low clocked segment. There's a reason it's only in that one part of the segment, it's the only place they can get it and the 10nm process to work.

By the way, Intel's own tests state that not all mitigation patches may have been applied which calls into doubt the real IPC increase.

I'm also wondering why you chose the words "a true die shrink". That statement is quite confusing since it seems to refer to AMD but that would make no sense. What "true die shrink" are you talking about and what is fake about the die shrink you're complaining about?
 
as said above, Skylake to coffe lake are the same architecture. another fact, broadwell due the huge 128mb eDRAM L4 cache pool allow better IPC than the newer Skylake but that made broadwell way more expensive.. if you think there's an architectural difference between 6700k and 9900k then you are wrong, they are the same but with more cores.

Another reason I still run my 6700K at 4.2 and call it a day. When we see some major progress I'll buy back in.
 
It is a fact that Xeons did run 30% slower in their demo.



Damage control is what someone I know did when Zen was very different to what his 'sources' leaked him. I remember him invented the crazy theory that AMD wasn't selling the real Zen, but "engineering samples rebranded" :D

The difference was that Intel enabled AVX512 for the demo.



I remark again that pricing is irrelevant to AMD crippling performance of competitors. I didn't fail to answer any question. I did a remark and you misinterpreted completely what I said.

I remember perfectly the exagerated claims made about how Naples was going to change everything, and Intel was in trouble, reality was that Intel not only keep the datacenter business, but even increased it.

Zen only got momentum because Intel 10nm fiasco, but Sunny Cove is already here and not only increases the IPC gap with AMD, but brings us a true die shrink.


AVX is a gimp, it is not a universal standard, Intels own slides state no support for non intel parts beyond 128 so it is really just intel running their own favourable conditions and then putting up a disclaimer that says real world will be different. It was as juvinile as getting an unknown "tech company" to run benches then claim the 9900K was 2x faster than the 2700 when they gimped the 2700 with slow Memory and stock cooling. When the final reviews came out, it was obviously not true then they just threw that company under the bus.

Intels refute was also not live it was slides, again begging the question as to why.

The datacenter market has shown progress on AMD's side, intel are not increasing market they are losing it, given the relative size between the two companies and Intels debt loss even at 2-3% per year will end up substantial losses.

Investor confidence also suggests that the world is open to AMD, if nobody wanted AMD they stocks would not be at over 30 dollars.
 
And if Sunny Cove is the messiah you and others want people to believe it is it would be all over the product stack instead of in the low power and low clocked segment. There's a reason it's only in that one part of the segment, it's the only place they can get it and the 10nm process to work.

Intel 10nm can't do 5GHz+ like it's mature 14nm, but that doesn't mean it would only work in low power.

It's just that 14nm is better suited to desktop where ultimate performance matters most, and 10nm is best for mobile where perf/watt matters most.

That is just logical usage of technology available to Intel. Even TSMCs process with the apparent ability to do 4.6GHz would not be an adequate replacement for Intels 5GHz+ process on the desktop.

4.6 GHz is an improvement for AMD, it would be a step backwards for Intel.
 
Really, it's an outright miracle for TSMC, and at Intel it's just Thursday :D

It is substantial, current gen stock and loaded have clocks around 3.7/3.8ghz, say the newgen does 500+ on that, it would be substantial gains in say gaming while less valuable in highly parallel production.

Whether you accept it or not,the competition is real and AMD is excelling in good business strategy.
 
Let me stop you right there- you can make the same argument for every extension Intel or AMD have added to x86 since x87. And you'd be wrong, given that AMD is building AVX into their hardware and software developers are using it.

In Intel's AVX slide it states support for non genuine Intel is limited to AVX, yup it is a gimp, which is why AVX is not universally accepted.
 
And if Sunny Cove is the messiah you and others want people to believe it is it would be all over the product stack instead of in the low power and low clocked segment. There's a reason it's only in that one part of the segment, it's the only place they can get it and the 10nm process to work.

By the way, Intel's own tests state that not all mitigation patches may have been applied which calls into doubt the real IPC increase.

I'm also wondering why you chose the words "a true die shrink". That statement is quite confusing since it seems to refer to AMD but that would make no sense. What "true die shrink" are you talking about and what is fake about the die shrink you're complaining about?

Messiah is AMD parlance. For everyone else Sunny Cove is only a new muarch from Intel with lots of improvements. Intel always introduce new products in function of economic and technologic parameters. Recall that not only mobile is much more important than desktop, but 10 and 10+ haven't been designed to compete with 14++ in raw performance. We will have to wait to 10++ for >5GHz desktops

D7xSleqXkAA4Pvq?format=jpg.jpg


We have to doubt both Intel and AMD IPC claims and wait to reviews.

Intel 10nm is a true die-shrink of 14nm. We cannot say the same about other foundries, e.g. TSMC 10nm isn't a true shrink of 16nm, but a ~12nm node with a marketing name.
 
Last edited:
Sunny Cove is the ultrabook part with Ice Lake cores...

It is just the opposite. Sunny Cove is the name of the core. Sunny Cove is the sucessor to Palm cove cores.

Sunny Cove cores are used in several products such as Icelake (client), Icelake (server), Lakefield, and Nervana NPP-I series.
 
It is substantial, current gen stock and loaded have clocks around 3.7/3.8ghz, say the newgen does 500+ on that, it would be substantial gains in say gaming while less valuable in highly parallel production.

I'd agree that it is substantial, and that it's also still substandard.

Whether you accept it or not,the competition is real and AMD is excelling in good business strategy.

AMD does what AMD does, mostly to survive, and they occasionally flirt with being competitive.

In Intel's AVX slide it states support for non genuine Intel is limited to AVX, yup it is a gimp, which is why AVX is not universally accepted.

AVX is universally accepted. It's the right tool for the job, and due to that AMD is implementing it at their usual lethargic pace.
 
AVX is a gimp, it is not a universal standard, Intels own slides state no support for non intel parts beyond 128 so it is really just intel running their own favourable conditions and then putting up a disclaimer that says real world will be different. It was as juvinile as getting an unknown "tech company" to run benches then claim the 9900K was 2x faster than the 2700 when they gimped the 2700 with slow Memory and stock cooling. When the final reviews came out, it was obviously not true then they just threw that company under the bus.

AMD supports 256bit wide AVX since Excavator and 512bit wide AVX is a standard in both HPC and datacenter. Sunny Cove brings 512bit support to mainstream client products.
 
I'd agree that it is substantial, and that it's also still substandard.



AMD does what AMD does, mostly to survive, and they occasionally flirt with being competitive.



AVX is universally accepted. It's the right tool for the job, and due to that AMD is implementing it at their usual lethargic pace.


This is cringy level of WCCFtechism generic response fanboism.

I said it last week, I will say it again, you couldn't sell me land in waterworld.
 
Messiah is AMD parlance. For everyone else Sunny Cove is only a new muarch from Intel with lots of improvements. Intel always introduce new products in function of economic and technologic parameters. Recall that not only mobile is much more important than desktop, but 10 and 10+ haven't been designed to compete with 14++ in raw performance. We will have to wait to 10++ for >5GHz desktops

View attachment 166444

We have to doubt both Intel and AMD IPC claims and wait to reviews.

Intel 10nm is a true die-shrink of 14nm. We cannot say the same about other foundries, e.g. TSMC 10nm isn't a true shrink of 16nm, but a ~12nm node with a marketing name.

Intel are undisputed leaders in slideshow technology for sure, that part I don't deny.
 
Messiah is AMD parlance. For everyone else Sunny Cove is only a new muarch from Intel with lots of improvements. Intel always introduce new products in function of economic and technologic parameters. Recall that not only mobile is much more important than desktop, but 10 and 10+ haven't been designed to compete with 14++ in raw performance. We will have to wait to 10++ for >5GHz desktops

View attachment 166444

We have to doubt both Intel and AMD IPC claims and wait to reviews.

Intel 10nm is a true die-shrink of 14nm. We cannot say the same about other foundries, e.g. TSMC 10nm isn't a true shrink of 16nm, but a ~12nm node with a marketing name.

Go ahead, keep trying to spin for Intel since Intel isn't even trying to spin 10nm anymore. Have you forgotten about the release Intel put out not that long ago that said they were putting resources towards 7nm at this point. Translation: 10nm is a failure and you won't see any high end parts out of it because it's a failure. We're going to eek out what little we can from it until 7nm comes around.
 
Go ahead, keep trying to spin for Intel since Intel isn't even trying to spin 10nm anymore. Have you forgotten about the release Intel put out not that long ago that said they were putting resources towards 7nm at this point. Translation: 10nm is a failure and you won't see any high end parts out of it because it's a failure. We're going to eek out what little we can from it until 7nm comes around.

From May Investor Meeting

intel-2019-investor-meeting-10-nm.png


"We are delivering, without compromise, on the original performance and scaling targets".
 
From May Investor Meeting

View attachment 166493

"We are delivering, without compromise, on the original performance and scaling targets".

What does that even mean and where do you even get all those slides from lol !?
Please show us a roadmap slide that formerly list a high performance desktop part on 10nm, 9700K and up for example (naming is irrelevant, I meant performance tier wise)

Edit: [H] isn't an investor meeting where you can throw nice graph with numbers interpreted in a skewed way for show. Everything in those slides are aimed at drawing blurry pictures of the market while "still" be accurate to ensure shareholders are motivated. Drawing conclusion and using those for making a point is low, you can do better.
 
It’s a statement that can mean anything to anyone, it’s how politicians answer their questions.
 
What does that even mean and where do you even get all those slides from lol !?
Please show us a roadmap slide that formerly list a high performance desktop part on 10nm, 9700K and up for example (naming is irrelevant, I meant performance tier wise)

Edit: [H] isn't an investor meeting where you can throw nice graph with numbers interpreted in a skewed way for show. Everything in those slides are aimed at drawing blurry pictures of the market while "still" be accurate to ensure shareholders are motivated. Drawing conclusion and using those for making a point is low, you can do better.

Intel announced multiple 10nm processors for 2019 and 2020, including the 10+ server-class, but didn't give a detailed roadmap with models and specs.
 
Say what you will, but 18% IPC is fucking incredible for one generation.
 
What does that even mean and where do you even get all those slides from lol !?
Please show us a roadmap slide that formerly list a high performance desktop part on 10nm, 9700K and up for example (naming is irrelevant, I meant performance tier wise)

Edit: [H] isn't an investor meeting where you can throw nice graph with numbers interpreted in a skewed way for show. Everything in those slides are aimed at drawing blurry pictures of the market while "still" be accurate to ensure shareholders are motivated. Drawing conclusion and using those for making a point is low, you can do better.

They get to blow smoke up their investors rear end.

To Juan it is Intel so it must be real
 
  • Like
Reactions: N4CR
like this
Say what you will, but 18% IPC is fucking incredible for one generation.

Over Skylake.

Skylake, kabylake, coffee lake, so 3 generations give or take. It may not be much better than the 9th Feb. Intel selectively used older generation to tout to investors.
 
Over Skylake.

Skylake, kabylake, coffee lake, so 3 generations give or take. It may not be much better than the 9th Feb. Intel selectively used older generation to tout to investors.

Skylake (client), Kabylake, and CoffeLake use the same core architecture. You confound core with family product.
 
Over Skylake.

Skylake, kabylake, coffee lake, so 3 generations give or take. It may not be much better than the 9th Feb. Intel selectively used older generation to tout to investors.

Round and round we go... Tell me what was the IPC change between Sky Lake, Kaby Lake and Coffee Lake?? Last I checked they are essentially identical CPU cores that everyone complains had no IPC change.

Now after many years of complaining there was no IPC change, many of the same complainers, are now implying there was an IPC change on the *Lake cores.

This whole thread is a cesspool of double standards and hypocrisy.

Intel's IPC claim for Sunny Cove, given the very significant core changes, makes some sense, just like the work AMD did on Zen 2, makes sense of their IPC claims.

Both are claiming similar changes and similar IPC improvements.

Both will ultimately need to be verified on wide selection of benchmarks by reliable third parties.

Both give reasons some optimism tempered with a bit of skepticism until the results are verified.
 
Round and round we go... Tell me what was the IPC change between Sky Lake, Kaby Lake and Coffee Lake?? Last I checked they are essentially identical CPU cores that everyone complains had no IPC change.

Now after many years of complaining there was no IPC change, many of the same complainers, are now implying there was an IPC change on the *Lake cores.

This whole thread is a cesspool of double standards and hypocrisy.

Intel's IPC claim for Sunny Cove, given the very significant core changes, makes some sense, just like the work AMD did on Zen 2, makes sense of their IPC claims.

Both are claiming similar changes and similar IPC improvements.

Both will ultimately need to be verified on wide selection of benchmarks by reliable third parties.

Both give reasons some optimism tempered with a bit of skepticism until the results are verified.

Is it the same? that would be like saying that Intel have made no IPC gains since first Gen Core I series as it is the same uArch, while the generation is the same the uArch's have changed and every generation has brought about some IPC gains, the problem with Intel is that from Haswell to Skylake/Kaby the gains were minimal, since then Intel has used higher frequency to increase those gains.
 
Total performance = ClockSpeed x IPC.

Intel has had lackluster IPC gains for a few years. If they really have 18% with a new core architecture, that will be amazing.

It might just be that they have implemented hardware fixes for Spectre/Meltdown vulnerabilities, and regained the lost IPC the prior generations felt from applying the mitigation's (microcode + whatever else). Even if this is the case, it is good news and likely will put Intel back on top for IPC performance or increase the lead.

People have said they "cheated" to gain performance advantage in the way side-channels operate. More likely no one (at Intel) thought anyone would ever figure out a way to siphon out information they way they have. Hardware fixes for the portions of these newer class of vulnerabilities were likely not all that difficult to implement, for those vulnerabilities that can be fixed. And for those vulnerabilities that apply to the overall architecture and can't be "fixed", affect AMD as well. Fixes for those will probably take longer and require an overhaul of the design. This may or may not be what Sunny Cove is. Would need an engineer to answer that.
 
More likely no one (at Intel) thought anyone would ever figure out a way to siphon out information they way they have.

Given the type of access needed to exploit these vulnerabilities- which exist in different forms across x86 out-of-order CPUs- I can see why both Intel and AMD made the choices they did. Intel also made them ~6 years ago, and had they been able to follow their own timelines then (and not essentially been lying to themselves about 10nm), they'd already be working on the release of their second uarch after Skylake.

Meaning that many of the servers where these vulnerabilities are most likely to be attacked with potential exploits would already be retiring.

Since that isn't the case... yeah. This affects datacenters significantly as they need to weight their attack surface with respect to each vulnerability, however, from the consumer / gaming perspective the impact is quite minimal.
 
Total performance = ClockSpeed x IPC.

Intel has had lackluster IPC gains for a few years. If they really have 18% with a new core architecture, that will be amazing.

It might just be that they have implemented hardware fixes for Spectre/Meltdown vulnerabilities, and regained the lost IPC the prior generations felt from applying the mitigation's (microcode + whatever else). Even if this is the case, it is good news and likely will put Intel back on top for IPC performance or increase the lead.

The average impact of Spectre/Meltdown vulnerabilities is almost zero in real software. The increase in IPC of Sunny Cove core is a consequence of a new improved microarchitecture. Compared to cores on Skylake the changes are:
  • Front-end
    • 1.5x larger µOP cache (2.25k entries, up from 1536)
  • Back-end
    • Wider allocation (5-way, up from 4-way)
    • 1.6x larger ROB (352, up from 224 entries)
    • Scheduler
      • Larger scheduler (up from 97 entries)
      • Larger dispatch (10-way, up from 8-way)
      • Execution ports rebalanced
      • New store data port
      • New store AGU port
  • Memory subsystem
    • LSU
      • 1.8x more inflight loads (128, up from 72 entries)
      • 1.7x more inflight stores (72, up from 42 entries)
    • Larger L1 data cache (48 KiB, up from 32 KiB)
    • Larger L2 cache (512 KiB, up from 256 KiB)
      • Larger STLBs
        • Larger 1G table (1024-entry, up from 16)
        • Larger 4k table (2048 entries, up from 1536)
        • New 1,024-entry 2M/4M table
    • 5-Level Paging
      • Large virtual address (57 bits, up from 48 bits)
      • Significantly large virtual address space (128 PiB, up from 256 TiB)
 
wow its kinda fresh to see juanrga defend intel like this.

but we know the test is purposely compare to skylake due to avx512 reason, which boost IPC of sunny cove. imo with the list of things intel changed is no different from broadwell -> skylake. at best without avx512 and security patches, we'll only see 3% improvement.

juanrga you should get a zen 2 CPU, 32 cores 7nm TR will only use as much power as 16 cores 2950x, which is more power efficient than intel's 16 cores.
 
wow its kinda fresh to see juanrga defend intel like this.

but we know the test is purposely compare to skylake due to avx512 reason, which boost IPC of sunny cove. imo with the list of things intel changed is no different from broadwell -> skylake. at best without avx512 and security patches, we'll only see 3% improvement.

juanrga you should get a zen 2 CPU, 32 cores 7nm TR will only use as much power as 16 cores 2950x, which is more power efficient than intel's 16 cores.
Seems a bit optimistic, would expect a little more, or maybe even at a lower clock. If it is equal I will be very impressed.
 
The average impact of Spectre/Meltdown vulnerabilities is almost zero in real software. The increase in IPC of Sunny Cove core is a consequence of a new improved microarchitecture. Compared to cores on Skylake the changes are:
  • Front-end
    • 1.5x larger µOP cache (2.25k entries, up from 1536)
  • Back-end
    • Wider allocation (5-way, up from 4-way)
    • 1.6x larger ROB (352, up from 224 entries)
    • Scheduler
      • Larger scheduler (up from 97 entries)
      • Larger dispatch (10-way, up from 8-way)
      • Execution ports rebalanced
      • New store data port
      • New store AGU port
  • Memory subsystem
    • LSU
      • 1.8x more inflight loads (128, up from 72 entries)
      • 1.7x more inflight stores (72, up from 42 entries)
    • Larger L1 data cache (48 KiB, up from 32 KiB)
    • Larger L2 cache (512 KiB, up from 256 KiB)
      • Larger STLBs
        • Larger 1G table (1024-entry, up from 16)
        • Larger 4k table (2048 entries, up from 1536)
        • New 1,024-entry 2M/4M table
    • 5-Level Paging
      • Large virtual address (57 bits, up from 48 bits)
      • Significantly large virtual address space (128 PiB, up from 256 TiB)

Aren't all generational upgrades a result of subtle changes to the existing uArch. Would AMD improvements not be because of same? Or is that mystical unicorns at silicon level farting twinkle dust?
 
Back
Top