First professional review of Intel's Rocket Lake (11700K) is out.. and it's a disaster

[pythonmegapixel]

Well, RIP Intel.

 

Soon there will be literally no reason for most users to choose them over AMD.

[leadeater]

21 minutes ago, FakeKGB said:

The 10900K is also 8C/16T, and from my understanding, will just be a cherry-picked 10700K.

10900K? I know Intel naming is bad but releasing new products with the exact same name as current products, wow that is bad.

 

Spoiler

I think you both mean 11700K/11900K 

 

[Spindel]

On 3/5/2021 at 11:21 PM, TrigrH said:

No further comment necessary. 

Apple Silicon just entered the chat

[LAwLz]

3 hours ago, illegalwater said:

I'm not even going to mention performance regressions in this analogy because apparently that ticks people off because "CpUs ArE CoMpLeX" even though there have been countless releases throughout history that were across the board better in real world tasks than their predecessors.

Well since you did mention it, and since you did mock me I'll respond by quoting the article you apparently didn't even read before posting:

 

 

Quote

All workloads at their core, even when browsing the web or word processing, can be split into integer (whole numbers, most workloads) and floating point (numbers with decimal places, workloads with math). In our testing, we saw the following:

• Single thread floating point: +19.0%
• Multi-thread floating point: +19.5%

Sounds great, right?

• Single thread integer: +13.0%
• Multi-thread integer: +7.3%

Oh. While Intel’s claim of +19% is technically correct, it only seems to apply to math-heavy workloads. The benefits of non math-based throughput are still better than average, 7-13%, but vary rarely do Intel’s big claims come with an easily identifiable asterisk.

When we look at our real-world data, in almost every benchmark the 11700K either matches or beats the 10700K, and showcases the IPC gain in tests like Dolphin, Blender, POV-Ray, Agisoft, Handbrake, web tests, and obviously SPECfp. It scores a big win in our 3DPM AVX test, because it has AVX-512 and none of the other CPUs do.

 

Let me repeat that.

in almost every benchmark the 11700K either matches or beats the 10700K, and showcases the IPC gain in tests like Dolphin, Blender, POV-Ray, Agisoft, Handbrake, web tests, and obviously SPECfp

 

Can you please stop pretending like there is a performance regression when there isn't? It is slightly slower in some benchmarks, mostly games, where it it doesn't even matter because it's like 200 FPS vs 190 FPS, or within margin of error.

Meanwhile you completely ignore most benchmarks where there is a ~10% performance integer performance increase, and about a 20% fp performance increase.

[illegalwater]

42 minutes ago, LAwLz said:

Well since you did mention it, and since you did mock me I'll respond by quoting the article you apparently didn't even read before posting:

 

 

 

Let me repeat that.

in almost every benchmark the 11700K either matches or beats the 10700K, and showcases the IPC gain in tests like Dolphin, Blender, POV-Ray, Agisoft, Handbrake, web tests, and obviously SPECfp

 

Can you please stop pretending like there is a performance regression when there isn't? It is slightly slower in some benchmarks, mostly games, where it it doesn't even matter because it's like 200 FPS vs 190 FPS, or within margin of error.

Meanwhile you completely ignore most benchmarks where there is a ~10% performance integer performance increase, and about a 20% fp performance increase.

You say there isn't performance regression then immediately proceed to say it's slightly slower in some benchmarks. Sounds like regression to me, but okay?

 

Oh and FYI Rocket Lake caps out at 8 cores, so compared to Comet Lake it's literally going to be a regression in performance for any workload that scales with cores, the IPC gains just aren't enough to make up a 2 core deficit.

[LAwLz]

1 hour ago, illegalwater said:

You say there isn't performance regression then immediately proceed to say it's slightly slower in some benchmarks. Sounds like regression to me, but okay?

You: The new Intel CPUs are slower than the old ones!

Me: No they aren't. On average they are 10-20% faster. They are slower in some meaningless tests and in those it's an unnoticeable difference.

You: So you agree with me! I am right!

 

You are correct in the same sense that I would be correct in saying Zen 3 performs worse than Zen 3. It is technically correct in some specific scenarios, but it is incorrect if we are talking about performance in most scenarios, which is what matters.

 

 

1 hour ago, illegalwater said:

Oh and FYI Rocket Lake caps out at 8 cores, so compared to Comet Lake it's literally going to be a regression in performance for any workload that scales with cores, the IPC gains just aren't enough to make up a 2 core deficit.

Yes, and what is your point?

Again, you are talking about very specific scenarios (programs that fully utilize more than 8 cores), and comparing CPUs in two different price brackets (the 8 core Rocket Lake chip will most likely be significantly cheaper than the 10 core Skylake processor) and trying to make it out as if that's the overall chip.

 

Why are you not satisfied with giving an accurate depiction of these new chips? 10-20% faster than the old ones on average, but not enough to catch up with AMD's offerings.

Intel is still slower and uses more power. There is no need to exaggerate the power draws by looking at AVX512 numbers. There is no reason to cherry pick benchmarks to try and paint them in worse light than necessary. 

[D13H4RD]

8 hours ago, illegalwater said:

You say there isn't performance regression then immediately proceed to say it's slightly slower in some benchmarks. Sounds like regression to me, but okay?

 

Oh and FYI Rocket Lake caps out at 8 cores, so compared to Comet Lake it's literally going to be a regression in performance for any workload that scales with cores, the IPC gains just aren't enough to make up a 2 core deficit.

They’ve got a point though. There were numerous productivity benchmarks that showed the increased IPC. However, the dip in overall latency was apparent in the gaming benchmarks though they were also mostly apparent at the 1080p low benchmarks, especially with a 2080 Ti. Hardware Unboxed’s Steven Walton also made a point that some games may not be nearly as latency dependent, so there may be some titles where the IPC gain is apparent.

 

For a lot of CPUs, it’s kind of a give-and-take situation. Not everything is going to be universally better in a new generation and some things will have to give. It’s mostly a question of whether what is gained is significant enough to offset what’s being taken away. 

 

With Rocket Lake, the goal of it is very clearly to extract as much IPC out of it as possible. The move to Cypress Cove cores, which are backported 10nm Sunny Cove cores on 14nm mixed in with Xe execution units hints at hard limitations on extracting more out of the now 5+ year old Skylake micro-arch. Whilst they certainly hit that goal, we’d be remiss if we also didn’t look at what they had to sacrifice. Is it noticeable in the real-world? Honestly, not really, but it’s also hugely disappointing in the sense that it’s not really any better than Comet Lake in a manner that makes it significant.

 

The big question mark is the price. With the performance we saw, and assuming these hold up at launch, then I personally think the best they do is suck it up, swallow their pride and undercut Zen 3. While there will be reasons to buy Intel, in a pure numbers and performance sense, AMD has the superior product. This is especially true for the Core i9 11900K, where it most likely has a significant disadvantage in both gaming and productivity. The disappointment will grow if Intel decides to charge a premium over Zen 3 for what looks like an inferior product. Remember that AMD already had a significant premium on their Zen 3 parts over Zen 2. Even more expensive CPUs will be an even tougher pill to swallow.

[Derangel]

12 hours ago, illegalwater said:

I guess I'm just baffled they even bothered releasing this thing. They would've been better off just taking the L for a year while Alder Lake finishes, especially since the gap between the two is seemingly going to be less than six full months. All Rocket Lake does is further the perception that Intel has become a joke in the desktop market, it's also negatively impacted the last thing Intel had going for them (gaming performance).

 

I mean could you imagine if Nvidia went five years just releasing refreshes of Pascal on 14nm, and then they finally come out with a new architecture.. and the best thing you could say about it is that its just marginally better in some areas? I'm not even going to mention performance regressions in this analogy because apparently that ticks people off because "CpUs ArE CoMpLeX" even though there have been countless releases throughout history that were across the board better in real world tasks than their predecessors.

 

Basically, Intel has to. They need to keep investors happy and the only way to do that is to keep releasing product. They also need to appease system integrators and retail parents. Not to mention that simply not releasing Rocket Lake would mean tens of millions of dollars just flushed down the toilet from a completed design never being used. Plus, these things will sell. With AMD’s stock still being rough and the Intel name, Rocket Lake will push units. It will give them time to, hopefully, get something competitive to market. Leaving the market bare for an entire cycle would be disastrous.

[illegalwater]

31 minutes ago, Derangel said:

Basically, Intel has to. They need to keep investors happy and the only way to do that is to keep releasing product. They also need to appease system integrators and retail parents.

I'll give you that.

 

Although I don't feel like it would've been that offensive to investors and system integrators if they had asked for an extra six months on Comet Lake. They could've sold it as a marketing thing by advertising Alder Lake as having a colossal performance bump over Comet Lake. Then again maybe Intel will just ignore Rocket Lake in the marketing anyway and plaster "~40% performance boost over Skylake!" everywhere.

36 minutes ago, Derangel said:

Not to mention that simply not releasing Rocket Lake would mean tens of millions of dollars just flushed down the toilet from a completed design never being used.

That's pocket change to Intel, companies their size write off losses like that all the time.

36 minutes ago, Derangel said:

Plus, these things will sell. With AMD’s stock still being rough and the Intel name, Rocket Lake will push units. It will give them time to, hopefully, get something competitive to market.

AMD's stock situation is getting better though, the 5600X and 5800X are coming in stock more frequently and for longer, Intel can't bank on that for much longer.

41 minutes ago, Derangel said:

Leaving the market bare for an entire cycle would be disastrous.

They wouldn't be leaving it bare though, Comet Lake is still competitive in the low end to mid range markets, especially for gaming. Plus AMD has nothing new coming this year unless you believe that sketchy rumor of a Zen 3 refresh in Q4, which wouldn't be relevant anyway because that'd be competing with Alder Lake, not Rocket Lake.

[Derangel]

17 minutes ago, illegalwater said:

I'll give you that.

 

Although I don't feel like it would've been that offensive to investors and system integrators if they had asked for an extra six months on Comet Lake. They could've sold it as a marketing thing by advertising Alder Lake as having a colossal performance bump over Comet Lake. Then again maybe Intel will just ignore Rocket Lake in the marketing anyway and plaster "~40% performance boost over Skylake!" everywhere.

 

Six months? Alder Lake isn't scheduled to go into volume production until the second half of the year, assuming there are no delays. If it even manages to launch this year it won't be until Q4, maybe super late Q3. 2022 is entirely possible for Alder Lake at this point.

[illegalwater]

55 minutes ago, Derangel said:

 

Six months? Alder Lake isn't scheduled to go into volume production until the second half of the year, assuming there are no delays. If it even manages to launch this year it won't be until Q4, maybe super late Q3. 2022 is entirely possible for Alder Lake at this point.

Comet Lake-S release date: April 30th 2020

Comet Lake-S one year anniversary plus six months: October 30th 2021

 

Literally in the middle of Q4..

 

Why is Intel so opposed to longer product cycles anyway? They seem to have worked pretty well for AMD. They work just fine in the GPU market too.

[Kisai]

4 hours ago, illegalwater said:

Comet Lake-S release date: April 30th 2020

Comet Lake-S one year anniversary plus six months: October 30th 2021

 

Literally in the middle of Q4..

 

Why is Intel so opposed to longer product cycles anyway? They seem to have worked pretty well for AMD. They work just fine in the GPU market too.

I'm going to say the Spectre/Meltdown had something to do with all these revisions of the Skylake 14nm

 

First of all, there wasn't enough testing done on Skylake, thus all these exploits wound up in every version of the chip on 14nm up to the 10th gen cpu. Second of all, they aren't (apparently) using the same dies on the mobile parts, and Intel has been prioritizing mobile parts, despite how utterly terrible the performance is.

 

See Apple's ARM CPU. There is something wrong with everything you're doing when an ARM part can run circles around it. So because Apple has decided to release ARM based laptops (not just tablets) now, that lights a fire under HP, Dell, Lenovo, Samsung, etc's butts to release an ARM laptop as well, and that requires relying on Microsoft to release an ARM OS, and we've all seen what's happened before with Windows RT.

 

If you like Mac's in general, the ARM based laptop kicks the Intel laptop's behind in every way except RAM. Release a 32GB RAM model laptop, and you'll likely have most mid-tier professionals willing to replace their existing laptop in a heartbeat.  High-end laptops (and desktops) however still aren't going to be at parity with an ARM part for probably 2 years. So Intel still has some time to catch up, but AMD has already beaten Intel to that punch. Unless Intel has been working on 3nm all this time is is going to put out chips on 3nm before AMD gets to, they're going to get lapped.

 

Intel kept putting stuff out on 14nm because that's their stable platform, and desktops/servers aren't quite as sensitive to TDP, unlike laptops. But I would not buy a 14nm cpu unless it was the only CPU available.

[suicidalfranco]

feels like i've been reading of a new intel release every 3 months now

[GDRRiley]

On 3/6/2021 at 4:07 AM, porina said:

There is one test result in that review that really doesn't make sense to me, and it is the one you're looking at. That perf increase seems too much but we don't have enough context into understanding it.

its his PHD code optimized by intel and AMD for AVX512

[GDRRiley]

On 3/6/2021 at 7:40 AM, illegalwater said:

Skylake is over five years old, so I'm sorry but there's no excuse for Rocket Lake performing worse anywhere, it should be better in every scenario. It's disappointing that this is the best Intel can offer after half a decade of stagnation on architecture.

skylake is really 6-7 years old (they were done designing at least 12 months beforehand)

and this is already 2-3 years old.
14nm has been at its limits for 2-3 gens now, its got nothing left to give and the backport clearly didn't help them gain more than a few % in workloads

[GDRRiley]

On 3/6/2021 at 10:23 PM, MightyMishka said:

I feel bad for Intel, AMD is crushing them right now, even with the silicon shortages and stock problems.

don't feel bad, last time intel was like this they paid to keep their spot as best they could.
 

[pythonmegapixel]

It still amazes me how common Intel-based laptops are.

 

It doesn't feel like that long ago they Ryzen ones were rare and only came from a few OEMs. That was when Intel was still clearly ahead. I'm frankly surprised that laptop manufacturers don't now have AMD making up the same proportion of their product line as Intel used to; I can only assume that some of their agreements with Intel were long-term and/or had exclusivity clauses

[GDRRiley]

Just now, pythonmegapixel said:

It still amazes me how common Intel-based laptops are.

 

It doesn't feel like that long ago they Ryzen ones were rare and only came from a few OEMs. That was when Intel was still clearly ahead. I'm frankly surprised that laptop manufacturers don't now have AMD making up the same proportion of their product line as Intel used to; I can only assume that some of their agreements with Intel were long-term and/or had exclusivity clauses

AMD doesn't have enough 7nm supply, they needed to push mobile chips onto 6/5nm this gen to even have a hope of getting enough supply.
They've got at least 40 million laptop APUs not shipping because of the new PS5/Xbox X/S (rough math they can fit 2x more laptop APUs)

[cj09beira]

1 minute ago, pythonmegapixel said:

It still amazes me how common Intel-based laptops are.

 

It doesn't feel like that long ago they Ryzen ones were rare and only came from a few OEMs. That was when Intel was still clearly ahead. I'm frankly surprised that laptop manufacturers don't now have AMD making up the same proportion of their product line as Intel used to; I can only assume that some of their agreements with Intel were long-term and/or had exclusivity clauses

the answer is complicated but it boils down to:

amd not wanting to byte more than they can shew so they are increasing volume production targets slowly.

amd being quite limited in how much volume they can make overall now that global foundries is almost completely irrelevant (they are still making the io dies there) intel's factories production volume is no joke.

oems not trusting amd as much.

oems not wanting to loose that sweet sweet multi million dollar "advertising" budget, aka the dont sell amd bribe.

those 4 are probably the main ones

[GDRRiley]

1 minute ago, cj09beira said:

oems not trusting amd as much.

AMD did the whole hand off drivers to the OEMs, that didn't go well. my 3500U would get a broken driver every other month from a windows update, so now linux

[pythonmegapixel]

1 minute ago, cj09beira said:

oems not wanting to loose that sweet sweet multi million dollar "advertising" budget, aka the dont sell amd bribe.

That's the kind of thing I expected, though there can't be an actual blanket "don't sell AMD bribe" because almost every major OEM has some AMD models; also that would probably lead to a massive antitrust case.

 

It's probably done significantly more subtly than that.

[GDRRiley]

4 minutes ago, pythonmegapixel said:

That's the kind of thing I expected, though there can't be an actual blanket "don't sell AMD bribe" because almost every major OEM has some AMD models; also that would probably lead to a massive antitrust case.

 

It's probably done significantly more subtly than that.

Intel has done it before and I doubt the fine made them lose money overall, so they could be consider it again 
Intel has cards other than advertising like training for big chains sales people 

[D13H4RD]

1 hour ago, pythonmegapixel said:

It still amazes me how common Intel-based laptops are.

 

It doesn't feel like that long ago they Ryzen ones were rare and only came from a few OEMs. That was when Intel was still clearly ahead. I'm frankly surprised that laptop manufacturers don't now have AMD making up the same proportion of their product line as Intel used to; I can only assume that some of their agreements with Intel were long-term and/or had exclusivity clauses

Much of that relates to AMD's history with laptops. The last time I saw an AMD laptop available for sale in a store where I can take one home is inside a crappy cheap Samsung laptop with performance that is noticeably worse versus the Intel-based model sitting beside it. They simply never had a long history of making great laptop processors. Even Ryzen-based APUs, whilst a massive improvement over their last one, still weren't fantastic up until the Zen 2-based models in the R4000 lineup in 2020, which was the first time in at least an absolute eternity where AMD had not only made a mobile CPU package that could properly compete with Intel's best but also best them in several key areas, notably efficiency.

 

However, a good product is only part of the puzzle. That history of AMD laptops being historically kind of "not-very-good" does also extend to which OEMs are willing to go all-in at this point in time. The fact that we're seeing more designs, including some high-end flagship-level products being offered with Ryzen APUs in 2021 showcase improved laptop vendor confidence in AMD's product stack, though it also needs to be said that due to AMD's history with laptops, they don't exactly have a big validation team, which does hurt adoption, especially combined with 7nm supply constraints, which can explain why some laptops like the Dell XPS lineup continue to be Intel-based. Also worth noting that there have been some reports of minor instability on some Ryzen-based laptops, which have been steadily improving.

 

As for the whole "Intel bribed OEMs to keep Ryzen out", whilst that is certainly a possibility, given Intel's known history of anti-competitiveness, I also think it's little more than AMD fanboy babble at this point. Especially now, there are increasing numbers of compelling laptops which incorporate Ryzen APUs, notably from ASUS and Lenovo, with some others like Acer starting to incorporate them in more designs. Demand for AMD-based laptops has apparently been pretty strong. I really think most of the reasons for the limited number of upper-end laptop offerings with Ryzen APUs stems more from the supply chain and validation side of things, rather than malicious intent from their chief rival. It's not an impossibility; Intel's marketing team has usually come up with some significant brands like Centrino, Ultrabook and most recently, Evo, but I don't think it's a huge factor.

[porina]

2 hours ago, GDRRiley said:

its his PHD code optimized by intel and AMD for AVX512

Thanks for the timestamp. I love Ian but I can't stand HUB so wasn't going to watch that.

 

He had stated before that Intel optimised code for him, but it was new to me about data type casting in AVX-512. I had only looked at it from an execution perspective, where I'd only expect 2x FP64 performance relative to AVX2. So for such a big difference, gains had to be from elsewhere. Note AVX-512 is actually a family of instructions, but FP64 part is the basic one supported by all implementations. I did also wonder if other parts were usable.

 

BTW I got a copy of 3DPM from Ian in the past, although it is an older unoptimised version. I'm not aware of him distributing the newer versions openly. What got my interest was that this was the first software I saw that showed over 50% performance boost with HT/SMT enabled compared to without.

[leadeater]

2 hours ago, GDRRiley said:

its his PHD code optimized by intel and AMD for AVX512

Sweet, thanks for this. Really interesting info, AVX2 also being function/feature deficient compared to AVX-512 in terms of acceleration path support. Honestly with improvements like that even single unit AVX-512 implementations make sense as even if the theoretical performance compared to AVX2 for this is equal it's these detailed specific differences that could give significant performance improvements. This with AVX-512 power offsets to lower peak power and you'll have generally the same power as AVX2 with the technical benefits of AVX-512.