Discussion Intel current and future Lakes & Rapids thread

Page 432 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Feb 17, 2020
100
245
116
Keep in mind that clock scaling, all else equal, implies voltage scaling. For that reason, even a quite bad 1:2 IPC:Cdyn ratio will still provide power savings at the same performance level (assuming leakage isn't crazy). But yes, without process scaling to keep things in check, you end up with significantly higher power with both at max performance.

Yeah, I'm definitely oversimplifying. Leakage, area, and supply droop are huge factors as well.
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,684
136
I believe Intel did have a design ethos at one time which was something along the lines of: "For every 1% increase in power, a microarchitecture feature should deliver at least 2% IPC improvement." Not sure what happened to this approach but if Sunny/Willow Cove are anything to go by I think we can assume this philosophy was dropped a while ago.
I think this was related to the original goals of the Atom team. All their IPC jumps had to be met with a corresponding increase in (energy) efficiency as well, they couldn't rely on node efficiency jumps to carry them through.

Actually based on you mentioning this 2:1 ratio I was finally able to find a reference to this info:
Atom's architects, similar to those who worked on Nehalem, had the same 2:1 mandate: every new feature added to the processor's design had to deliver at least a 2% increase in performance for every 1% increase in power consumption.
 

eek2121

Platinum Member
Aug 2, 2005
2,904
3,906
136
In order to "lose performance" with a 20% IPC increase, clockspeed would have show a greater than 20% decrease. Are you seriously trying to say Golden Cove will only clock slightly above 4ghz? I believe GC will also be on a further enhanced process so my estimation is that it will clock at least a high or maybe higher.

We have seen a ADL-S ES hit 4.6 Ghz. It is likely that Intel will hit 5+ Ghz at release.
 

mikk

Diamond Member
May 15, 2012
4,112
2,106
136
Genoa is not a year and a half away. It's a year and a bit, but not coming right at the end of 2022. There is also a high possibility it is the first Zen 4 part, contrary to prior generations.


Don't be surprised if Geona won't come in 2022 and don't be surprised if the server part will come after the mainstream version.
 

rainy

Senior member
Jul 17, 2013
505
424
136
Don't be surprised if Geona won't come in 2022 and don't be surprised if the server part will come after the mainstream version.
Are you sure, you didn't confuse AMD with Intel - the latter delayed or even cancelled many projects in the last few years.

Btw, looks like you are really afraid of Zen 4, isn't?
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Dell's Ice Lake Xeons are now available and you can configure it with all the way to the 40 core. Now the 40 core pricing seemed so high that you would be crazy to buy it over Milan, but you could in theory buy a rack and have it delivered in a week or so.
 
  • Like
Reactions: lightmanek

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Hotchips conference has been announced, The Intel stuff is:

Sapphire Rapids
Alder Lake
Ponte Vecchio
SmartNic/DPU

any many other fun topics (but unrelated to this thread). Should be fun.

Great, another Intel hype season coming.

Intel shouldn't even be invited to Hot Chips anymore. Last year with Tigerlake, Intel at Hot Chips just regurgitated the information they revealed a month or two before. Literally there was nothing new, or techical that would justify being on Hot Chips. Whether they have good products or bad ones aside, their attitude just plain sucks.

It used to be Intel actually revealing something technical at Hot Chips.

It's like them shouting "I'm still relevant!" and "Please pay attention to me!"

Are you sure, you didn't confuse AMD with Intel - the latter delayed or even cancelled many projects in the last few years.

Btw, looks like you are really afraid of Zen 4, isn't?

Subtle attacks are still attacks.

You know it was at the height of Intel's technical superiority that they faltered right? Same can plague anybody. Actually it was exactly the same with AMD when after Athlon 64 they basically stopped.
 
Last edited:

itsmydamnation

Platinum Member
Feb 6, 2011
2,743
3,074
136
Great, another Intel hype season coming.

Intel shouldn't even be invited to Hot Chips anymore. Last year with Tigerlake, Intel at Hot Chips just regurgitated the information they revealed a month or two before. Literally there was nothing new, or techical that would justify being on Hot Chips. Whether they have good products or bad ones aside, their attitude just plain sucks.

It used to be Intel actually revealing something technical at Hot Chips.

It's like them shouting "I'm still relevant!" and "Please pay attention to me!"
it at least has a chance to be better then AMD's. we already have deep dives into Zne2 and RDNA2 .... sigh......
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,572
136
Great, another Intel hype season coming.

Intel shouldn't even be invited to Hot Chips anymore. Last year with Tigerlake, Intel at Hot Chips just regurgitated the information they revealed a month or two before. Literally there was nothing new, or techical that would justify being on Hot Chips. Whether they have good products or bad ones aside, their attitude just plain sucks.

It used to be Intel actually revealing something technical at Hot Chips.

It's like them shouting "I'm still relevant!" and "Please pay attention to me!"

To be fair to Intel, AMDs deep dives (after the famous Zen 1 presentation) have also only included last-year products. So even if Intel were AMD the best you could hope would be detailed analysis of Tiger Lake level tech.
 

mikk

Diamond Member
May 15, 2012
4,112
2,106
136
Great, another Intel hype season coming.

Intel shouldn't even be invited to Hot Chips anymore.


Fortunately you are not relevant, you really turned into a big moaner and whiner lately. I'm glad we will (should) get more infos about Alder Lake-S from Intel late August and every tech geek should be happy. It's clear your are not fine that will see something new from Intel, I guess you preferred the numerous Skylake refresh releases every year, unfortunately for you this is over.
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,572
136
@mikk, @eek2121

BTW to reiterate where I stand on Alder Lake. My estimates are:
  • It will be faster (in all likelyhood 10-20% faster) than Zen 3 in ST or lightly threaded productivity apps
  • A top-of-the-line 8 + 8 Alder Lake will give Ryzen 5900X a run for it's money in MT performance, not sure how much faster/slower but being all-around in the same range. This means:
    • 8 + 8 will be in very similar balpark while 8 + 0 conf will certainly win in heavily parallel AVX-512 enabled workloads, but in no AVX2 or below workloads that utilize more than 8 threads.
    • It will not outperform a 5950X on average in MT loads. At least not actual reviewer benchmark suites (say Tom's hardware, Anandtech etc)
  • In Gaming performance it should eek out Rocket Lake and Zen3 slightly but the difference will not be large (up to 5% on average) as gaming is mostly sensitive to cache and Memory Latency and Cache size and only to a lesser extent IPC. The titles where Zen 3 is doing better than Rocket lake (due to cache) will see the biggest gain going from 16MB -> 24MB of L3. All of This will only happen if all of the issues listed below are solved:
    • There the gain can be considerably larger than what's stated above. They will need to fix the cache-latency issues that plague Rocket Lake for that to be the case, but that's most likely not a problem as the latter was hampered by a backport to 14nm.
    • Windows Scheduler will also need updates to not raise any issues in games that scale to > 8 threads. This can be fixed by running in 8 + 0 config.
    • For clear gaming wins either DDR4 based MOBOs (for best OC latency in the beginning) or very good early DDR5 XMP module availability is requried.

Let's revisit this after reviews to see see how terribly AMD biased Intel hater I end up being in the end.

I still remember people being angry at me for claiming that Rocket Lake might not significantly outperform Zen 3 in gaming (due to aforementioned cache and memory limitations, which certainly ended up being true)
 
  • Like
Reactions: Tlh97

B-Riz

Golden Member
Feb 15, 2011
1,482
612
136
@mikk, @eek2121

BTW to reiterate where I stand on Alder Lake. My estimates are:
  • It will be faster (in all likelyhood 10-20% faster) than Zen 3 in ST or lightly threaded productivity apps
  • A top-of-the-line 8 + 8 Alder Lake will give Ryzen 5900X a run for it's money in MT performance, not sure how much faster/slower but being all-around in the same range. This means:
    • 8 + 8 will be in very similar balpark while 8 + 0 conf will certainly win in heavily parallel AVX-512 enabled workloads, but in no AVX2 or below workloads that utilize more than 8 threads.
    • It will not outperform a 5950X on average in MT loads. At least not actual reviewer benchmark suites (say Tom's hardware, Anandtech etc)
  • In Gaming performance it should eek out Rocket Lake and Zen3 slightly but the difference will not be large (up to 5% on average) as gaming is mostly sensitive to cache and Memory Latency and Cache size and only to a lesser extent IPC. The titles where Zen 3 is doing better than Rocket lake (due to cache) will see the biggest gain going from 16MB -> 24MB of L3. All of This will only happen if all of the issues listed below are solved:
    • There the gain can be considerably larger than what's stated above. They will need to fix the cache-latency issues that plague Rocket Lake for that to be the case, but that's most likely not a problem as the latter was hampered by a backport to 14nm.
    • Windows Scheduler will also need updates to not raise any issues in games that scale to > 8 threads. This can be fixed by running in 8 + 0 config.
    • For clear gaming wins either DDR4 based MOBOs (for best OC latency in the beginning) or very good early DDR5 XMP module availability is requried.

Let's revisit this after reviews to see see how terribly AMD biased Intel hater I end up being in the end.

I still remember people being angry at me for claiming that Rocket Lake might not significantly outperform Zen 3 in gaming (due to aforementioned cache and memory limitations, which certainly ended up being true)

Alder Lake looks like what Intel might have supplied to Apple had Intel not eff'd up so hard on tick-tock. Seems very similar in Big Little to the M1.

Rumors have already said it will be slightly faster than Zen3 in ST, and should match the 5900X in MT, but then, there is still the 5950X MT monster that has been out and a possible 6K series to fight Alder Lake, so, Intel will be matched / beat again.
 
  • Like
Reactions: Tlh97 and podspi

Timmah!

Golden Member
Jul 24, 2010
1,395
602
136
Greetings

i kind of stopped following things, given nothing interesting for me was really happening on Intel front, nor was supposed to happen for quite a while, but recently randomly saw the Saphire Rapids die shot, apparently 56-core/ 4x14core CPU, which apparently will come out for HEDT market as well, and i have questions:

- is it known, when this will happen? This year? Next year?
- apparently it has 64GB HBM as part of the chip, what does that mean? Does it mean each 14core tile has there 16GB of built-in RAM or will it be somehow layered upon those 4 tiles, i mean the foveros stacking thing?
- is that HBM meant to serve as some kind of L4 cache or is it supposed to serve as actual system RAM? 64GB system RAM for servers seems fairly low...and pricing-wise i cant see so much RAM to be part of consumer grade product... so will the HEDT part not have it? Anyway, it would surely provide massive performance boost by itself, would it not?

I am quite curious about it. I an still on by now oldish 7940x, so Alder Lake with its 8 cores does not really excite me (nor would it with 16, really), neither do AMDs current 16core mainstream CPUs, which feel like a minor upgrade (14 to 16cores with 30percent better IPC) for plenty of money,...could get Threadripper, but decided to hold-on another year or 2 to see whats next and then decide...
 

dullard

Elite Member
May 21, 2001
24,998
3,326
126
- is it known, when this will happen? This year? Next year?
- apparently it has 64GB HBM as part of the chip, what does that mean? Does it mean each 14core tile has there 16GB of built-in RAM or will it be somehow layered upon those 4 tiles, i mean the foveros stacking thing?
- is that HBM meant to serve as some kind of L4 cache or is it supposed to serve as actual system RAM? 64GB system RAM for servers seems fairly low...and pricing-wise i cant see so much RAM to be part of consumer grade product... so will the HEDT part not have it? Anyway, it would surely provide massive performance boost by itself, would it not?
All rumors so far, of course, but it looks like late 2021. The rumors say "up to" 64 GB of HBM2e memory. So, I would imagine that the cheaper chips would have less of it. The memory is supposedly flexible. If you can fit your software all in the HBM2e memory, then great, you'll have high bandwidth memory with low latency. If you need additional memory, then I think the HBM2e can act either as L4 cache or just as more system memory.
 
  • Like
Reactions: Tlh97 and Timmah!

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
- is that HBM meant to serve as some kind of L4 cache or is it supposed to serve as actual system RAM? 64GB system RAM for servers seems fairly low...and pricing-wise i cant see so much RAM to be part of consumer grade product... so will the HEDT part not have it? Anyway, it would surely provide massive performance boost by itself, would it not?

I think it will offer all 3 possible modes: L4 cache, part of linear mem and separate pool. 64GB is actually a huge working set to be had with 1TB/s of memory bw.

Linear mem mode is kinda hard to use, need to ensure that workload gets moved around to correct physical addresses that are backed by HBM. OS would need to do such page migration, not impossible for Linux with some hints from HW ( like counter of memory BW usage for some memory region)
L4 cache is automagic, but would probably come at some extra costs for tags of what is in L4 and so on, 64GB is huge amount to track and "the win" between DDR5 and HBMe is not even an order of magnitude.
Separate "hybrid" pool is kinda best for custom app, where you use API to allocate memory in said pool and use it directly, knowing that nothing will touch it. Kinda like hugepages of old times, but with much more support nowadays. Very interesting for HPC and ML boyz and girlz.
 
  • Like
Reactions: Timmah!

andermans

Member
Sep 11, 2020
151
153
76
L4 cache is automagic, but would probably come at some extra costs for tags of what is in L4 and so on, 64GB is huge amount to track and "the win" between DDR5 and HBMe is not even an order of magnitude.

Those tags are what make me sceptical about using it as L4. Just keeping the physical address of each cacheline around is about 8 GiB for 64 GiB of cache. Now you can probably reduce that by removing the unnecessary parts of the physical address but you are still stuck with 4 GiB+ of tags. That means either they have some really huge tag memory or the tags are in HBM too. Either way I think the tag comparisons are going to be too slow (latency) to really be useful, as AFAIU HBM2e doesn't have that much lower latency than DDR5?
 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
Those tags are what make me sceptical about using it as L4. Just keeping the physical address of each cacheline around is about 8 GiB for 64 GiB of cache. Now you can probably reduce that by removing the unnecessary parts of the physical address but you are still stuck with 4 GiB+ of tags. That means either they have some really huge tag memory or the tags are in HBM too. Either way I think the tag comparisons are going to be too slow (latency) to really be useful, as AFAIU HBM2e doesn't have that much lower latency than DDR5?

There is no/little chance of itbeing tracked as cache lines for L4. I think last time Intel did such cache was Broadwell with eDRAM and those tags ate 1/4th of L3 for just 128MB. Does not make much sense to have a line cached in L4 if tags access has to hit memory each time.

But it might be tracked in hw at say page granularity. 4KB pages would result in a truckload of "pages", 2MB not so. Adding some additional bits in hw TLB to mark some pages as being in HBMe could work. By definition hot and large working set needs coverage by TLB and you also "cache" said pages in HBMe ?
 

Timmah!

Golden Member
Jul 24, 2010
1,395
602
136
All rumors so far, of course, but it looks like late 2021. The rumors say "up to" 64 GB of HBM2e memory. So, I would imagine that the cheaper chips would have less of it. The memory is supposedly flexible. If you can fit your software all in the HBM2e memory, then great, you'll have high bandwidth memory with low latency. If you need additional memory, then I think the HBM2e can act either as L4 cache or just as more system memory.


Well, thats given its "up to" and i wont certainly be looking at the 56-core model, cause that would surely beyond my budget, but if Intel had sense and wanted to compete with Threadrippers, i could see myself having a look at potential 2-tile version of this with 28cores - that i would consider acceptable upgrade over what i currently have. If it was in the 32core TR price-range, or even better 24-core, or somewhere in between, and performed similarly, which i expect to happen, i would give it a thought. And if full 56-core version is supposed to have 64GB of that HBM, i think its safe to assume 28core part might have 32GB of it. Which is indeed still awesome for a home workstation. I wonder if its possible for it to cost those 1500~2000 EUROs like Threadrippers cost.

BTW, once again, regardless of what precisely the HBM is, out of those options, would it not provide the most significant performance improvement, i mean more so than the new core architecture or whatever AVX stuff? HBM is way faster than DDR5, right? And it will sit literally next to the cores, so the latency will be way lower, as you said.
 

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
Great, another Intel hype season coming.

Intel shouldn't even be invited to Hot Chips anymore. Last year with Tigerlake, Intel at Hot Chips just regurgitated the information they revealed a month or two before. Literally there was nothing new, or techical that would justify being on Hot Chips. Whether they have good products or bad ones aside, their attitude just plain sucks.

It used to be Intel actually revealing something technical at Hot Chips.

It's like them shouting "I'm still relevant!" and "Please pay attention to me!"
God your messages are awfully depressing now. I get it Intel has been Mis-executing and not living to expectations doesn’t mean they ‘shouldn’t be invited’. Intel hotchips program looks to be exciting, it’s their latest a greatest, you want them to show Tigerlake again? 🤷‍♂️
 
  • Like
Reactions: pcp7

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Well, thats given its "up to" and i wont certainly be looking at the 56-core model, cause that would surely beyond my budget, but if Intel had sense and wanted to compete with Threadrippers, i could see myself having a look at potential 2-tile version of this with 28cores - that i would consider acceptable upgrade over what i currently have. If it was in the 32core TR price-range, or even better 24-core, or somewhere in between, and performed similarly, which i expect to happen, i would give it a thought. And if full 56-core version is supposed to have 64GB of that HBM, i think its safe to assume 28core part might have 32GB of it. Which is indeed still awesome for a home workstation. I wonder if its possible for it to cost those 1500~2000 EUROs like Threadrippers cost.

BTW, once again, regardless of what precisely the HBM is, out of those options, would it not provide the most significant performance improvement, i mean more so than the new core architecture or whatever AVX stuff? HBM is way faster than DDR5, right? And it will sit literally next to the cores, so the latency will be way lower, as you said.

Assume at this point the Saphhire-X is going to be 2 chiplets only. And the HBM products are exclusive to HPC customers and won't be publically available... and if they are it'll be with eye watering prices.
 

Timmah!

Golden Member
Jul 24, 2010
1,395
602
136
Assume at this point the Saphhire-X is going to be 2 chiplets only. And the HBM products are exclusive to HPC customers and won't be publically available... and if they are it'll be with eye watering prices.

You think they wont try to compete with 64core TR, when they can? I mean, if only to have the halo-product, to be able to claim they have the fastest CPU and whatnot. Surely they wont be able to say that about 2-tile version.
Regarding HBM, i guess you are right. But do you assume then it will be stacked on top of those tiles and its not part of it?