• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Discussion Intel current and future Lakes & Rapids thread

Page 547 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Doug S

Senior member
Feb 8, 2020
788
1,145
96
Here's an article from Charlie discussing the new DDR5 from an enterprise perspective, note the comments he makes about 50% bandwidth increase with the initial DDR5 and this new RCD set to further increase that. That's who DDR5 is targeted at today.

The only mention of latency here was that it would be "more or less the same" - again, that's from the perspective of an enterprise customer who already accepts higher latency for the capacity of RDIMMs, or further (slightly) increased latency and reduced bandwidth for the power savings of LRDIMMs.

https://www.semiaccurate.com/2021/10/13/rambus-releases-2nd-gen-ddr5-rcds/
 
May 1, 2020
70
81
51
Agree! Can't wait for anandtech for reviews. I don't think Pat (CEO) has anything to do with the development of Alderlake, right? If so, how soon can we find a release with his work?
Him being CEO now, we will never again see a design where he had any kind of influence at all. His work is something else: Get the people, ideas, time and money for innovation together. Maybe make some key decisions based on sophisticated and/or management targeted analysis.
 

uzzi38

Golden Member
Oct 16, 2019
1,817
3,654
116
Needs a more powerful gpu
It shows the framerates the CPU would be able to handle in the test the way SoTR does. Still similar story, in the "simulation" part (first row) the 5950X wins, in the rendering part (second row) the 12900K wins.
 

Ajay

Diamond Member
Jan 8, 2001
9,485
3,951
136
I told you why. Because it isn't being binned to cherry pick the best chips to make DIMMs capable of minimal latency.

I'm not sure what process DRAM makers are using these days, but they might be using a newer process to make DDR5 chips which isn't as mature as the process being using to make most DDR4 chips.

Perhaps the internal ECC calculations might add a bit of latency, so maybe DDR5 never gets quite as fast as DDR4 even when it is being binned within an inch of its life on a mature process like DDR4. If so that would be maybe 5-10% at most though, not 40%.
Samsung Starts Mass Production of Most Advanced 14nm EUV DDR5 DRAM - Samsung US Newsroom

14nm EUV! Didn't look at others, just happened to remember seeing this.
 

Hulk

Diamond Member
Oct 9, 1999
3,116
563
126
ADL-S must be a nightmare for reviewer with all the possibilities. Windows 10, windows 11, DDR4, DDR5, big cores, small cores. I guess most will go with DDR5 and Windows 11. Hopefully some will do a DDR4 versus DDR5 comparison.
I think Will be an opportunity actually for the really good reviewers. The great CPU reviewers, like Ian, will be able to break things down and organize content so that we have a clear understanding of the benefits and drawbacks of each option you mentioned.
 

moinmoin

Platinum Member
Jun 1, 2017
2,772
3,673
136
The enterprise world has always been first to move on new memory and storage standards. First to a new DDRx standard, first to larger capacity hard drives, adopted SSDs years before they appeared in the consumer world, etc.
I'm well aware of all that. The area where gamers are early adopters are rather e.g. graphic cards (not of the crypto and AI variants), displays (especially of the high refresh rate variant), input devices (high polling rate mice), gamer branded stuff (but why?) etc.

I guess I misinterpreted your sentence I originally quoted to exclude gamers in general, not just in the cases you expanded on above?
 

Doug S

Senior member
Feb 8, 2020
788
1,145
96
Samsung Starts Mass Production of Most Advanced 14nm EUV DDR5 DRAM - Samsung US Newsroom

14nm EUV! Didn't look at others, just happened to remember seeing this.
Well remember that node names have not had anything to do with physical dimensions of a transistor for a while for logic processes. Maybe 14nm is some sort of actual dimension of the transistors or capacitors used for making these DDR5 dies, or maybe they've used the names as a placeholder to indicate increases in density the way they have for logic.

Given that they are using EUV for a few layers it seems it would be comparable to Samsung's 7nm logic process as far as drawn line widths.
 

Ajay

Diamond Member
Jan 8, 2001
9,485
3,951
136
I've never really followed DRAM silicon implementation, so IDK. But, yes, line widths and xtor dimensions (if they even use FinFET) should be achievable.
 

Bouowmx

Golden Member
Nov 13, 2016
1,122
519
146
GSKILL-DDR5-Trident-Z5-2.jpg

Based on Samsung, up to 6400 MT/s CL36. Sounds OK for an initial offering, equivalent to DDR4-3200 CL18.

The 8 GB per DIMM capacity appears to be missing from any DDR5 products.
 

geegee83

Junior Member
Jul 5, 2006
19
4
66
Agree! Can't wait for anandtech for reviews. I don't think Pat (CEO) has anything to do with the development of Alderlake, right? If so, how soon can we find a release with his work?
This product and those of the next few years were most probably shaped under Jim Keller’s watch.
 

Det0x

Senior member
Sep 11, 2014
549
772
136
More leaks in 3dmark and gaming

Furthermore, we learn that the system was equipped with Zadak’s DDR4 memory (DDR4 3866 C14-14-14-34 2T Gear1), so it was a DDR4 compatible board.

The creator revealed the scores of Core i9-12900K in UL’s 3DMark Time Spy and Fire Strike benchmarks, which is CPU test to be specific:
i9-12900K CPU:
Time Spy CPU Score: 17915
Time Spy Extreme CPU Score: 9004
Fire Strike CPU Score: 41278
Fire Strike Extreme CPU Score: 41477

My Maxed 5950x:
Time Spy CPU Score: 19062
Time Spy Extreme CPU Score: 11851
Fire Strike CPU Score: 44674
Fire Strike Extreme CPU Score: 44267

Gaming benchmarks also dont look good..
1634204775541.png 1634204793679.png 1634204813450.png
 

mikk

Diamond Member
May 15, 2012
3,253
1,065
136
Your RAM is faster, Time Spy is RAM sensitive and 4950 Mhz 5950x is super high OC for this CPU, this needs to be said.
 

JoeRambo

Golden Member
Jun 13, 2013
1,246
1,183
136
Your RAM is faster, Time Spy is RAM sensitive and 4950 Mhz 5950x is super high OC for this CPU, this needs to be said.
Still, gaming tests show things that can be summarized with this sentence:

Intel's engineers created a chip with IMC that is getting beaten by AMD's IMC on different IOD. Quite an "achievement" for what used to be world class memory controller team.
 

uzzi38

Golden Member
Oct 16, 2019
1,817
3,654
116
Your RAM is faster, Time Spy is RAM sensitive and 4950 Mhz 5950x is super high OC for this CPU, this needs to be said.
If you're looking at the reported frequency then no, it is not. The entire 3D mark suite notes the highest clock recorded at any point during the test for both the GPU and the CPU, and not under any real load either. E.g. mobile Turing Max-Q GPUs often report 1800MHz despite actually clocking at around 1500MHz during the test itself.

4950MHz is actually quite low overall. My own 5950X is also a relatively bad clocker, unable to hold 5GHz on a single core workload (like CB20 1T or GB5) even after tuning PBO with a +75MHz offset, and well:


5GHz reported.

Moat of my tests are done whilst I was testing GPU settings, I'll need to look for one where I actually ran the CPU test. Although my own RAM is far worse, only DDR4-3200 cl14 so not really comparable to the DDR4-3866 cl14 used on the 12900K.
 

lobz

Golden Member
Feb 10, 2017
1,818
2,353
136
If you're looking at the reported frequency then no, it is not. The entire 3D mark suite notes the highest clock recorded at any point during the test for both the GPU and the CPU, and not under any real load either. E.g. mobile Turing Max-Q GPUs often report 1800MHz despite actually clocking at around 1500MHz during the test itself.

4950MHz is actually quite low overall. My own 5950X is also a relatively bad clocker, unable to hold 5GHz on a single core workload (like CB20 1T or GB5) even after tuning PBO with a +75MHz offset, and well:


5GHz reported.

Moat of my tests are done whilst I was testing GPU settings, I'll need to look for one where I actually ran the CPU test. Although my own RAM is far worse, only DDR4-3200 cl14 so not really comparable to the DDR4-3866 cl14 used on the 12900K.
How did the article arrive to a <14000 score for the 5950X?
 

Det0x

Senior member
Sep 11, 2014
549
772
136
How did the article arrive to a <14000 score for the 5950X?
Turning off SMT in Timespy gives you a 2-3k higher cpu score when running a 5950x
(all the others 3dmarks don't punish you for having 32 threads, so SMT can be enabled when on running those)
 
  • Like
Reactions: Tlh97 and Elfear

lobz

Golden Member
Feb 10, 2017
1,818
2,353
136
Turning off SMT in Timespy gives you a 2-3k higher cpu score when running a 5950x
(all the others 3dmarks don't punish you for having 32 threads, so SMT can be enabled when on running those)
Gotcha, thanks. In other words, we still just have to wait. Unless you're Joe Rambo and you have decided what the strengths and what the weaknesses of the new CPU are already.
 

uzzi38

Golden Member
Oct 16, 2019
1,817
3,654
116
How did the article arrive to a score of <14000 for the 5950X?
Mine seems to score around 14500 that makes sense actually provided they tested with similar performance memory.

Power limit makes little difference here, peak power draw is 171.153W as measured by HWiNFO. But limit it to 142W and yeah, you'll probably end up with a score below 14000pts. So I think that makes sense, but it looks like the benchmark is extremely affected by memory performance.
 

uzzi38

Golden Member
Oct 16, 2019
1,817
3,654
116
Turning off SMT in Timespy gives you a 2-3k higher cpu score when running a 5950x
(all the others 3dmarks don't punish you for having 32 threads, so SMT can be enabled when on running those)
I take it back, this is actually a thing.

Huh. The more you know I guess.
 
  • Haha
Reactions: Tlh97 and lobz

ASK THE COMMUNITY