Intel 10nm/7nm CPU/GPU Rumor Thread : Cannonlake, Icelake, Tigerlake, Sapphire Rapids, etc.

greyboxer

Limp Gawd
Joined
Oct 23, 2017
Messages
247
even 4% is mostly a stretch according to the GN reviews. alot of times its 1-2%. for $100, huge ripoff. I was expecting 3600XT to compete with a stock 3700x and it just isnt even close. I paid $285 for my 3700x and its light years ahead of the 3600XT even when they get it up to 4.6ghz.
 

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,411
It's also annoying that AMD launched this "refresh" before desktop renoir. The 2400g I bought over 2 years ago is still the best APU they have right now.
At this point, it is pretty much too late and most will wait for RKL or Zen 3 before getting a new APU. Missed opportunity imho.
 

IdiotInCharge

[H]F Junkie
Joined
Jun 13, 2003
Messages
14,492
It's also annoying that AMD launched this "refresh" before desktop renoir. The 2400g I bought over 2 years ago is still the best APU they have right now.
At this point, it is pretty much too late and most will wait for RKL or Zen 3 before getting a new APU. Missed opportunity imho.
The... rumor site that shall not be named... has benchmarks on the new APUs.

Aside from the Vega GPU block (ugh), they look quite nice, especially when overclocked -- but I'd be willing to bet that a more conservative profile alongside some reasonably substantial cooling would provide very quiet operation as well.

More or less what I'd want from a socketed 10nm or smaller Intel part, if one existed...
 

kirbyrj

Fully [H]
Joined
Feb 1, 2005
Messages
26,269
The... rumor site that shall not be named... has benchmarks on the new APUs.

Aside from the Vega GPU block (ugh), they look quite nice, especially when overclocked -- but I'd be willing to bet that a more conservative profile alongside some reasonably substantial cooling would provide very quiet operation as well.

More or less what I'd want from a socketed 10nm or smaller Intel part, if one existed...
They also aren't going to have the huge cache that the regular Zen2 chips have IIRC.
 

kirbyrj

Fully [H]
Joined
Feb 1, 2005
Messages
26,269
I have intel because my ryzen 1700 had some compatibility issues with memory and the software I used at the time. In addition to this, when I upgraded to the 9900k the competing processor was the 2800x which wasn’t as fast.

Ryzen still needs fast memory to work optimally (cause of their infinity fabric) and that is a consideration in price also.

The intel processors, for the average person, aren’t that much worse than the AMD chips, if you don’t look at the 14+++ vs 7nm arguments, or clock for clock comparisons.

Also, you can’t overclock AMD chips worth a damn as they clock themselves, intel chips typically can be overclocked at least 10% on their AVX side, and some on their cache side too. This can make them monsters for single (or few) thread performance.
Memory pricing is coming back down to the point where a decent set of 3600Mhz is only a couple dollars more than a set of 3000/3200Mhz (if any difference at all on sale).
 

IdiotInCharge

[H]F Junkie
Joined
Jun 13, 2003
Messages
14,492
They also aren't going to have the huge cache that the regular Zen2 chips have IIRC.
They don't need it; that cache was / is a hack to overcome the latency penalty incurred by removing the memory controller from the CPU die(s). The APUs are monolithic, which means that they very well could outperform any of AMDs current eight-core parts.
 
  • Like
Reactions: N4CR
like this

kirbyrj

Fully [H]
Joined
Feb 1, 2005
Messages
26,269
They don't need it; that cache was / is a hack to overcome the latency penalty incurred by removing the memory controller from the CPU die(s). The APUs are monolithic, which means that they very well could outperform any of AMDs current eight-core parts.
Except in programs that favor a large CPU cache :p.

But seriously, I understand what you're saying. If the 4.5Ghz clock speeds are to be believed, it could definitely rival the 8 core Intel parts straight up. The cross CCX latency obviously doesn't help Zen2. Removing that performance penalty could be a big boost. Last I saw the pricing was slightly over $300?
 

IdiotInCharge

[H]F Junkie
Joined
Jun 13, 2003
Messages
14,492
Except in programs that favor a large CPU cache :p.
Sure!

I don't really know what those are, though, for a desktop user. Reading the reviews of the Zen 2 XT, it looks like the advantage is mostly restricted to synthetic workloads. Real-world consumer workloads are likely to be too dynamic to benefit.
But seriously, I understand what you're saying. If the 4.5Ghz clock speeds are to be believed, it could definitely rival the 8 core Intel parts straight up. The cross CCX latency obviously doesn't help Zen2. Removing that performance penalty could be a big boost. Last I saw the pricing was slightly over $300?
They're still quoting the same synthetics that boost Zen 2's highlights, so it remains to be seen; the bigger advantage is in terms of simply not really giving anything up from a desktop user's perspective.

I'm not expecting the Vega graphics to be much of a standout, and AMDs media transcoding support (whether due to hardware limitations, their drivers, or just reluctant developers) is abysmal relative to QuickSync and especially NVENC, but supposing that's not needed, AMD finally has a decent GPU-less answer. Or one that can use its GPU to drive additional displays separate from the main GPU.
 

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,411
Some juicy new rumors for RKL:
https://www.techpowerup.com/269667/intel-core-i7-rocket-lake-chips-to-be-8-core-12-thread

It looks like the i7 will be an 8/12 part. This seems necessary as the i9 maxes out at 8/16 and the i5 will be a 6/12 part as to compete with the next gen R5.
What's more is that i3 and below may be CML refresh. That raises a lot of questions on i/o. No leaks on core count, but if I were to wager a bet, the i3 would be a 6/6 part so as to compete with the next gen 4/8 R3. Pentiums and Celerons would most likely be lower clock/cache 4/8 and 4/4 parts.
20200710_153540.jpg
 

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,411
Or that someone made a mistake in the slide deck.
it makes sense from a product segmentation point and is technically achievable. It may balance hot spots on the die as well.

They want to keep their i9 flagship brand and RKL is limited to 8/16. The i5 NEEDS to be 6/12 so this seems logical.
 

IdiotInCharge

[H]F Junkie
Joined
Jun 13, 2003
Messages
14,492
8C/12T? So some cores service 2 threads and some just 1? That seems bananas.
I'm already imagining some adjustment to the physical topology where this might make sense; perhaps some level of cache that is shared between pairs of cores, such that an extra assignable thread between the two could be very well fed.
 

Snowdog

[H]F Junkie
Joined
Apr 22, 2006
Messages
10,749
I'm already imagining some adjustment to the physical topology where this might make sense; perhaps some level of cache that is shared between pairs of cores, such that an extra assignable thread between the two could be very well fed.
It's not that. It's either:

A) Typo
B) Really stupid gimmick disabling HT on half the cores for another segmentation level (reeks of desperation).
 

Chimpee

[H]ard|Gawd
Joined
Jul 6, 2015
Messages
1,548
So any updated guesses on when Intel desktop will actually leave 14nm with the new 7nm delays?
https://www.pcgamer.com/amp/intels-...the-first-casualty-of-the-7nm-node-nightmare/

Alders Lake in 10nm is set to release late 2021/early 2022, but don't be surprised if 7nm delays will cause 10nm shortages in which mobile (Tiger Lake) and server (Ice Lake/Sapphire Rapids) would take priority.
Who knows at this point, just seem like 1 delay after another. It be nice if Intel is on schedule for a change.
 

Denpepe

[H]ard|Gawd
Joined
Oct 26, 2015
Messages
1,596
So any updated guesses on when Intel desktop will actually leave 14nm with the new 7nm delays?
https://www.pcgamer.com/amp/intels-...the-first-casualty-of-the-7nm-node-nightmare/

Alders Lake in 10nm is set to release late 2021/early 2022, but don't be surprised if 7nm delays will cause 10nm shortages in which mobile (Tiger Lake) and server (Ice Lake/Sapphire Rapids) would take priority.
Tiger lake should be out later this year still, now it will be wait and see to see if they will be available in quantity, also supposedly intel ordered some 160k 7nm wafers at TSMC in an attempt to get more chips in the supply chain, but again, wait and see if it is true and what they will be used for.
 

IdiotInCharge

[H]F Junkie
Joined
Jun 13, 2003
Messages
14,492
So what has proven to be better: 9400 six cores or 10300, 10100 4 cores 8 threads? Are two true cores better? Besides locked 7700 vs 8400 and that era.
Still very much workload dependent. Personally I'd take more physical cores (same argument applies to 6C12T vs 8C8T too), but again, having more physical cores isn't always faster. On Intel at least, HT is great for having extra thread resources where light-weight processes that must run can be parked without having the OS interrupt the core for a context switch.
 

Snowdog

[H]F Junkie
Joined
Apr 22, 2006
Messages
10,749
Which is better: 9400 six cores or 10300, 10100 4 cores 8 threads? Are two true cores better? Not for buying now.
Generally HT is worth up to 30% more performance in highly multi-threaded applications. So 1.3 Multiplier.

4C with HT * 1.3 = 5.2

No HT means 1.0 multiplier.

6C without HT *1.0 = 6.

6 Cores without HT, is ahead of 4 Cores with HT.
 

Keljian

[H]ard|Gawd
Joined
Nov 7, 2006
Messages
1,146
Generally HT is worth up to 30% more performance in highly multi-threaded applications. So 1.3 Multiplier.

4C with HT * 1.3 = 5.2

No HT means 1.0 multiplier.

6C without HT *1.0 = 6.

6 Cores without HT, is ahead of 4 Cores with HT.
For the record, this is a good metric for intel, but not for AMD. Ryzen benefits more like 40% per core with SMT, intel is 25% best case.

https://www.agner.org/optimize/blog/read.php?i=838

As long as the code is running from the micro-operations cache, it can execute five instructions per clock, where Intel has only four. [...] floating point code with scalars or vectors [...] will execute on the AMD processor at a maximum rate of four instructions per clock (two additions and two multiplications), while the Intel processor can do only two. [...] The gain in total performance that you get from running two threads per core is much higher in the Ryzen than in Intel processors because of the higher throughput of the AMD core.
 

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,411
Which is better: 9400 six cores or 10300, 10100 4 cores 8 threads? Are two true cores better? Not for buying now.
Even with AMD 6/6 is always better than 4/8 (67% cores). With Intel, you need at least 6/12 to match 8/8 (75% of cores).

The 3300x is able to match the 3500x mostly due to a single ccx and higher clocks. That will most like NOT be the case with Zen 3 if they move to 8 core CCX.

Because of that, I have a feeling that AMD will have a more mainstream 6/6 to fill the gap next gen. Intel 11th gen i3s are rumored to be Skylake-based so it is a good possibility that they will have a 6/6 again as well.
 

Keljian

[H]ard|Gawd
Joined
Nov 7, 2006
Messages
1,146
No, 6/12 will beat 8/8 on intel in productivity apps and general apps most of the time. Will also produce less heat/use less power.
 
Top