Discussion Intel current and future Lakes & Rapids thread

[IntelUser2000]

I did use the picture without the orange outline yesterday and got 45mm².

And Icelake is 41.1mm2 according to wikichip. One thing we can both agree on is that its a big advancement.

Something that I hope they can do it again with 7nm variants with the space afforded and the increased perf/watt.

Intel doesn't show die data for Icelake on their datasheet. I get it. Publicly they shy away from telling press like Anandtech because people start grilling and arguing about it. But why not on datasheets? I know when I post die data on forums the visibility is way, way less than if Intel told it directly.

All we have since Skylake are measured data, which inherently contain errors.

 

[A///]

Intel doesn't show die data for Icelake on their datasheet. I get it. Publicly they shy away from telling press like Anandtech because people start grilling and arguing about it. But why not on datasheets? I know when I post die data on forums the visibility is way, way less than if Intel told it directly.

Didn't Ian debate an Intel rep over this and called him out on his BS or am I thinking of someone else?

 

[IntelUser2000]

Didn't Ian debate an Intel rep over this and called him out on his BS or am I thinking of someone else?

No clue. That does sound like something a company trying to hide things will do.

 

[A///]

No clue. That does sound like something a company trying to hide things will do.

Don't take my word on it. It might have been about another company. He wasn't pleased then if he's who I'm thinking of and he wasn't pleased with some asshat who threw some lame credentials around. I want to say it was around the Cascade Lake launch time but like I said, I don't remember very well.

 

[Ajay]

Making matters worse, 64C Zen 3 will be out before anything more than 28C from Intel. This is slowly becoming ridiculous.

And will be ridiculous for the next several years, IMHO. And that's if Intel can make meaningful changes to their culture.

Pretty sad, but given what we are learning about the inner workings of Intel's management hierarchy and things like the balkanization of the design teams and the manufacturing team.
Silly slogans like "no transistor left behind" is just useless clatter - maybe some young engineers will think it matters, till they get more experience. And what is this constant obsession with Moore's law, who cares?
Making great products that meet or exceed customer expectations is the only thing that matters.

 

[IntelUser2000]

Making great products that meet or exceed customer expectations is the only thing that matters.

Moore's Law is still a big part of that. There's no other thing that can bring such big gains in perf/watt and perf/$. If you scrap that you are left with conventional improvements - think Automobiles and Airplanes and other technology improvements. Compared to MPUs they are glacial slow.

Eventually I think that's what will happen in the compute world too. It'll slow to slightly better than conventional improvements.

 

[A///]

I've heard some interesting rumors about potential Genoa specifications. Nothing worth sharing because they're fairly outlandish, but if Intel is worried now, then they'll run around like chicken with their heads cut off if what I've heard is even a little bit true. Even "insert painter name" for Zen 4-5 is going to test Intel's engineering skills. I'm half excited but mostly worried if Intel don't step up with a solid product pipeline for their datacenter hardware division.

 

[IntelUser2000]

I've heard some interesting rumors about potential Genoa specifications. Nothing worth sharing because they're fairly outlandish,

Not super hard what direction they can potentially take. Moving to 96 cores for example. I think I saw 128 cores somewhere but they aren't going to do it on 5nm being that its pretty much like a half node(like 28 to 20, and 14 to 10).

 

[A///]

Not super hard what direction they can potentially take. Moving to 96 cores for example. I think I saw 128 cores somewhere but they aren't going to do it on 5nm being that its pretty much like a half node(like 28 to 20, and 14 to 10).

Yeah that's been floated a long time now. 128 would/may be later on several generations in. Desktop ryzen will supposedly see an increase in cores as well. There's other stuff but it's so wild I'm putting it in a salt mine. If they can provide the performance each generation alongside core count increases, then I can't complain. If it's just core increases with negligible performance gains then it just becomes a contest who can cram more on a die. I'll probably pick up an Alderlake myself. A midrange chip or low end just to play around with. If it's bad, well, so be it it'll become an htpc.

 

[IntelUser2000]

If its too wild, then its probably not true. 15-20% additional perf/clock improvements which is not at all negligible, maybe support HBM2 in certain versions, 10-channel memory being floated.

The core count being behind in servers is an indication of being behind in process technology. Just forward it by one year, and we'd have had Icelake 38 cores for few months now and be expecting Sapphire Rapids.

4.3GHz is maximum for Cooper Lake. Why do we need a circuit designed for 5GHz+ operation? Even for Tigerlake. They said they had an option for better uarch, higher clocks, or a combo of both. Why not aim for 4.5GHz with better uarch? The pursuit for unreasonable clocks on desktop hurts rest of the product line.

That or they need a third core.
-Core for 5GHz+ operation on desktops and H
-Core for 4.3GHz operation on servers and laptops with smaller core size
-Atom-based

Alternatively,
-Super-Core with 4.5GHz+ operation having 20% over Core for hybrids and laptops
-Core for servers with <4.3GHz clock target and half the size of Super-Core
-Atom-based

 

[LightningZ71]

From a purely mathematical point of view that looks at the current physical packages, on desktop Ryzen, a switch to N5+ compute Dies and a smaller node for the I/O die (though, it doesn't have to be N7), would give AMD enough room to put three compute dies in the AM4/AM5 package. If they stick to 8 cores per CCD on N5+, that's still 24 cores. DDR5 should keep that well fed with data.

On TR/EPYC, a switch to a smaller process I/O die and N5+ compute dies should give them enough space for 12-14 compute dies (instead of pairs in the corners, it would be three per corner, and possibly one at either end of the I/O die, or, if the N5+ compute dies are small enough, it could be as many as three groups of three on each side of the I/O die and maybe one at either end, for a grant total of 20 compute dies... which is 160 cores, a crazy high, near outlandish number, and likely near impossible to wire to the I/O die successfully with even tech from two years down the road), which in a 2P system, is 24-28 compute dies, or 192-224 cores. With modest clocks, that should be doable within a reasonable power envelope. Just a number to think about, with 12 compute dies alone, and with no increase in L3 cache for N5+, that's still 384MB of L3 cache in each processor. If they elected to do a pair of HBM2E stacks in the package, one at either end of the I/O die, that could also include 8GB of L4 cache as well.

Those numbers sound unreal in the lens of 2016. Now, it's mathematically possible to fit it all in he package and not out of the realm of possibility.

 

[Dave2150]

And will be ridiculous for the next several years, IMHO. And that's if Intel can make meaningful changes to their culture.

Pretty sad, but given what we are learning about the inner workings of Intel's management hierarchy and things like the balkanization of the design teams and the manufacturing team.
Silly slogans like "no transistor left behind" is just useless clatter - maybe some young engineers will think it matters, till they get more experience. And what is this constant obsession with Moore's law, who cares?
Making great products that meet or exceed customer expectations is the only thing that matters.

It's pretty ridiculous that AMD still can't beat Intel at gaming, despite them having a brand new architecture on a brand new process. Intel has a 5 year old architecture and even older process, yet still outperform AMD at gaming.

 

[Thala]

It's pretty ridiculous that AMD still can't beat Intel at gaming, despite them having a brand new architecture on a brand new process. Intel has a 5 year old architecture and even older process, yet still outperform AMD at gaming.

AMD wins the MT benchmarks mostly due to higher core counts and not because the Ryzen core microarchitecture is particularly impressive - infact it is pretty average.

 

[A///]

It's pretty ridiculous that AMD still can't beat Intel at gaming, despite them having a brand new architecture on a brand new process. Intel has a 5 year old architecture and even older process, yet still outperform AMD at gaming.

Yeah the i3-10320 performs similar to the 7700K with the new processor sometimes having a 15-30 fps lead in some games (I couldn't find any 10 game comparisons between it and the 6700K), but Intel has also managed to improve on Skylake in all that time, so it's not straight up Skylake. Intel has also improved frequency by a lot. The original 6700K had a base of 4 Ghz with a boost clock of 4.2 Ghz. The i3-10320 boosts from 3.8 to 4.4 Ghz with a super boost of 4.6 Ghz. The 10900K boosts from 3.8 Ghz base to an all core of 4.8 Ghz, turbo boosts to 5.1 Ghz and turbo maxes at 5.2 Ghz.

AMD's boost algorithm isn't great on Zen2, and should hopefully improve with Zen3, fingers crossed. There's also the issue of the CCX and cache latency which penalizes AMD in gaming. Gaming is pretty latency sensitive. You also forget that Ryzen's only been around for 3 almost 4 years now. There's not a lot of games optimized for it if any that I can think of. Major game developers and even software companies tend to optimize for the dominant CPU vendor. For almost 10 years, Intel remained on top of the hill. They were the king, so it was a no brainer for companies to optimize their software or games for Intel processors.

 

[A///]

AMD wins the MT benchmarks mostly due to higher core counts and not because the Ryzen core microarchitecture is particularly impressive - infact it is pretty average.

Ok, but the benchmarks I've seen (GN) showed the 3700X either very close to matching or just trailing a 10900K at stock clocks for both processors from Blender, Premier, Chromium compile times, etc. They either trade blows or one edges out the other. 3700X is a 8/16 3.6/4.4 Ghz processor. No idea how well or bad it boosts. Don't own Ryzen, haven't owned an AMD system in a couple decades. The 10900K with its extra 2 cores and higher clocks should put a big gap in your example. Not picking on your post but it wasn't too well thought out and made it seem as if AMD just slapped together some stuff. Personally I'd still not buy an AMD system and I'm on edge of buying a Zen 3 processor unless it addresses the main issues people complain about.

Intel Core i9-10900K CPU Review: Gaming, Overclocking, & Benchmarks vs. AMD Ryzen

Intel’s continuing to bring the heat, literally and figuratively, and is now leveling its new 10C/20T part squarely at AMD. We have a separate in-depth review coming up on the 10600K, which has interesting implications for the R5 3600 and the realm of gaming, but for launch, we need to start...

www.gamersnexus.net

I think gaming is a wash on AMD at the moment, and I honestly wouldn't recommend AMD to any gamer, small or big time simply because of the issues present. What's interesting in the last graph, unless I misread it, at load curing a Cinebench test the 3990X uses around 20 watts more than a 10900K at stock, while having 54 more cores. Impressive, but I guess that's a wash since there's a 3,500 price difference and serve two very different purposes.

 

[jpiniero]

AMD's boost algorithm isn't great on Zen2, and should hopefully improve with Zen3, fingers crossed. There's also the issue of the CCX and cache latency which penalizes AMD in gaming. Gaming is pretty latency sensitive. You also forget that Ryzen's only been around for 3 almost 4 years now. There's not a lot of games optimized for it if any that I can think of. Major game developers and even software companies tend to optimize for the dominant CPU vendor. For almost 10 years, Intel remained on top of the hill. They were the king, so it was a no brainer for companies to optimize their software or games for Intel processors.

You mean optimize for Jaguar.

 

[A///]

You mean optimize for Jaguar.

I don't know much about older consoles. Haven't owned one since the PS2. I hardly ever play games. Maybe an hour two every few months.

 

[lobz]

It's pretty ridiculous that AMD still can't beat Intel at gaming, despite them having a brand new architecture on a brand new process. Intel has a 5 year old architecture and even older process, yet still outperform AMD at gaming.

This kneejerk reaction was pretty funny, considering that we were talking about server processors. It's like someone stepped on your toe. It's not AMD's fault that even more than 1,5 years late the best ICL Xeon Intel can come out at the end of 2020 is a 28C chip.

Wanna hear something cool? Intel would LOVE IT if it were the other way around, so AMD leading at gaming while having abysmal value in the server space. With that in mind, I'm not really sure it's that ridiculous that AMD hasn't sacrificed all its advantages when designing Zen 2, just to be 5-10% better in gaming.

 

[itsmydamnation]

AMD wins the MT benchmarks mostly due to higher core counts and not because the Ryzen core microarchitecture is particularly impressive - infact it is pretty average.

load of BS....

Its just the trade offs AMD made to hit what other vendors do with ~5 different dies with in effect one die, its not the core its the package.
if you look at Renoir memory scaling you will see its gaming minimums beat matisse despite 1/4 the cache. If you then look with comparable memory you can see the effects of that cache as matisse beats Renoir.

If AMD chose to do both of those together we wouldn't be talking about this right now would we. And now we can see how just changing the package can have a big over all impact to perception.

 

[tamz_msc]

load of BS....

Its just the trade offs AMD made to hit what other vendors do with ~5 different dies with in effect one die, its not the core its the package.
if you look at Renoir memory scaling you will see its gaming minimums beat matisse despite 1/4 the cache. If you then look with comparable memory you can see the effects of that cache as matisse beats Renoir.

If AMD chose to do both of those together we wouldn't be talking about this right now would we. And now we can see how just changing the package can have a big over all impact to perception.

Well, he's talking about the Zen cores, not AMD's innovations in packaging. In that regard he's right. For example compared to Sunny Cove, Zen 2 lags behind in terms of branch prediction, even though the latter is an improvement over Zen 1. In branch-misprediction sensitive benchmarks Sunny Cove is around 50-100% ahead of Zen 2 and Skylake. From AT's Ice Lake deep dive:

Amongst SPECint2006, the one benchmark that really stands out beyond all the rest is the 473.astar. Here the new Sunny Cove core is showcasing some exceptional IPC gains, nearly doubling the performance over the 8550U even though it’s clocked 100MHz lower. The benchmark is extremely branch misprediction sensitive, and the only conclusion we can get to rationalise this increase is that the new branch predictors on Sunny Cove are doing an outstanding job and represent a massive improvement over Skylake.

 

[A///]

No Sunny Cove on desktop, though. We won't see any Cove cores for another year or two. I'm presuming Intel's 11th gen desktop will launch in January or February. Cove cores were redesigned so it's no surprise they outperform Intel processors and AMD, though the AMD is a bit odd since AMD bungled the CCX and cache with their unique take.

Edit: Thought I recognized your SN. Lol

 

[itsmydamnation]

Well, he's talking about the Zen cores, not AMD's innovations in packaging. In that regard he's right. For example compared to Sunny Cove, Zen 2 lags behind in terms of branch prediction, even though the latter is an improvement over Zen 1. In branch-misprediction sensitive benchmarks Sunny Cove is around 50-100% ahead of Zen 2 and Skylake. From AT's Ice Lake deep dive:

Because one core is stronger in one area does not make a core overall "average".
Especially when you look at what he was quoting with said reply. In the particular instance ( gaming performance ) misses to memory has a far bigger impact then many other workloads, a large amount of those other workloads AMD leads both per clock and overall per core. Now with the exact same core the gaming performance scenario can have a sizeable impact by simply having a 16mb ccx monolithic die.

Thus my point , the AMD gaming performance levels are design choices made outside of the core and without gaming/desktop being the primary driver.

We will have to wait and see what Zen3 desktop memory latency looks like, hopefully we see advances over Matisse.

 

[coercitiv]

Well, he's talking about the Zen cores, not AMD's innovations in packaging.

The cores themselves are the result of a strategy which included that higher core count target. Calling them average in isolation is bizarre. Imagine telling someone a team is composed solely of average players, it just so happens these players are exceptional at playing together. The "exceptional" attribute may be exclusive to the resulting system, but such a result still requires special calibration of the individual parts, and that calibration is an attribute of the part, not the system.

Remember when someone on this forum told us even a monkey can scale a world-class core into a world-class chip? I wonder what kind of monkey scales an average core into a world-class chip.

Going back to Tiger Lake, I like what I'm seeing with Lenovo's Yoga line, the Tiger Lake design wins are quite appealing. However, availability starts with November 2020. Has anyone else announced models with Sep/Oct availability?

 

[tamz_msc]

The cores themselves are the result of a strategy which included that higher core count target. Calling them average in isolation is bizarre.

There is nothing wrong in comparing uarchs in isolation - just because Ice Lake struggles to clock beyond 4GHz doesn't mean that Sunny Cove isn't a good design. People liked to compare Bulldozer with Zen 1 and be impressed about the +52% IPC uplift - when there was nothing exceptional about it since one was a speed demon and the other was a brainiac design.

 

[tamz_msc]

Ice Lake AVX-512 Downclocking

Examining the extent of AVX related downclocking on Intel’s Ice Lake CPU

travisdowns.github.io

Looks like Ice Lake has minimal-to-nonexistent downclocking when running AVX-512 provided you do not run into power and thermal limits. Lets hope this holds true for Tiger Lake as well though I don't expect it to because 4.7GHz is pretty high for a ST turbo frequency.