Question Raptor Lake - Official Thread

Page 72 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hulk

Diamond Member
Oct 9, 1999
4,525
2,516
136
Since we already have the first Raptor Lake leak I'm thinking it should have it's own thread.
What do we know so far?
From Anandtech's Intel Process Roadmap articles from July:

Built on Intel 7 with upgraded FinFET
10-15% PPW (performance-per-watt)
Last non-tiled consumer CPU as Meteor Lake will be tiled

I'm guessing this will be a minor update to ADL with just a few microarchitecture changes to the cores. The larger change will be the new process refinement allowing 8+16 at the top of the stack.

Will it work with current z690 motherboards? If yes then that could be a major selling point for people to move to ADL rather than wait.
 
  • Like
Reactions: vstar

Just Benching

Banned
Sep 3, 2022
307
156
76
Zen 4 used 6000 in the tests and has the IMC overclocked by 14.5%, but the tested 12900K was also granted the same frequency wich amount to 25% overclocking.

That being said competent reviewers will use stock settings for everything, including RAM of course since the IMC speed is manufacturers guaranted only on thoses conditions.
Yes but the 12900k can actually support 7000+ memory speeds, probably even 8k with the new hynix A dies.
 

biostud

Lifer
Feb 27, 2003
18,700
5,431
136
As someone that will likely do a full platform upgrade by the end of the year with either Zen 4 or Raptor Lake, it won't be a big factor. Pcie 4.0 is plenty fast enough for end users like me, especially the second generation Pcie 4.0 nvme drives like the Samsung 990 pro, which I will probably end up buying in the 2 TB variety.

I can see Pcie 5.0 x4 SSDs being a benefit to prosumers doing workstation like applications that involve heavy amounts of reads and writes to and from storage however.

But for gaming, which is the most system intensive application for consumers by far, it won't make much of a difference. The SSDs in the PS5 and XSX both use Pcie 4.0 so games won't be exceeding that limitation for a long time.

As for Direct Storage, by all indications it already flies on Pcie 4 drives. A Pcie 5 drive won't reduce loading much further due to diminishing returns.

I completely agree.

When I bought my X99 platform in 2014 I wanted to be sure to have one with a PCI3.0 x4 nvme slot so it was future proof (many boards had PCIe 2.0 or x2 slots) 8 years later, and I never bothered to upgrade my sata SSD :p. So when I upgrade next time I will buy the fastest drive I can get my hands on (hoperfully a pcie5 drive), even though it is overkill, because I know I'm going to hold on to it until the computer is ready for a full upgrade. And if the AM5 platform will support three generations then I don't think I will regret getting a PCIe 5 drive, when also going zen4->"zen6".
 
  • Like
Reactions: Carfax83

Just Benching

Banned
Sep 3, 2022
307
156
76
... but that wasn't the point of the post you're answering to, was it now? Trying to have a conversation with you in good faith is a colossal waste of prescious free time.
I dont know what his point was, my point is alderlake supports higher memory frequency than zen 4. Testing with jedec speeds is absolutely useless since noone actually runs their cpu like that.

And since he mentioned clocks, the 12900k supports ddr 3200. That translates to an imc freq of 1600mhz. Running ddr5 even at 6000 mhs is actually a downclock since the imc runs at 1500mhz in that case.
 

uzzi38

Platinum Member
Oct 16, 2019
2,705
6,427
146
I dont know what his point was, my point is alderlake supports higher memory frequency than zen 4. Testing with jedec speeds is absolutely useless since noone actually runs their cpu like that.

And since he mentioned clocks, the 12900k supports ddr 3200. That translates to an imc freq of 1600mhz. Running ddr5 even at 6000 mhs is actually a downclock since the imc runs at 1500mhz in that case.
That's actually factually incorrect.

The IMC (UCLK) on Zen 4 is capable of 3000MHz+ - it runs 1:1 with the actual memory frequency (MEMCLK). It's the FCLK that craps out past 2000MHz. We actually don't know how far Raphael's memory controller can go.

As an example, AMD's test rig for their launch day used DDR5-6000, which means:

- 2000MHz FCLK (IF frequency)

- 3000MHz UCLK (memory controller frequency)

- 3000MHz MEMCLK (memory frequency)
 

Just Benching

Banned
Sep 3, 2022
307
156
76
the first result from google

And what does that supposed to mean? Is the person buying a dell prebuild interested in reviews and benchmarks?

I mean what are you actually saying, that amds zen 4 testing was useless since they used 6000c30 ram?
 

Det0x

Golden Member
Sep 11, 2014
1,299
4,234
136
Should be around 60ns with just xmp enabled. You can drop it down to the 50ns range with just a little bit of tunning, just the primaries and trefi + trfc
You forgot to mention you need to disable e-cores, clock ringbus higher for that to be possible.. As proof above show that you will get around ~65ns @ 6000 MT/s 40-40-40 XMP with all cores enabled and stock clocks otherwise.
 
  • Like
Reactions: lightmanek

Just Benching

Banned
Sep 3, 2022
307
156
76
You forgot to mention you need to disable e-cores, clock ringbus higher for that to be possible.. As proof above show that you will get around ~65ns @ 6000 MT/s 40-40-40 XMP with all cores enabled and stock clocks otherwise.
No, not really, i get 51ns with ecores on and stock cache. Only tuned primaries and trefi + trfc. Stock ram frequency as well
 

Carfax83

Diamond Member
Nov 1, 2010
6,841
1,536
136
That's actually factually incorrect.

The IMC (UCLK) on Zen 4 is capable of 3000MHz+ - it runs 1:1 with the actual memory frequency (MEMCLK). It's the FCLK that craps out past 2000MHz. We actually don't know how far Raphael's memory controller can go.

As an example, AMD's test rig for their launch day used DDR5-6000, which means:

- 2000MHz FCLK (IF frequency)

- 3000MHz UCLK (memory controller frequency)

- 3000MHz MEMCLK (memory frequency)

Intel's memory controller can only do 2:1 ratio and not 1:1 with DDR5, but comparing latency with Zen 4 Alder Lake still comes out on top, let alone Raptor Lake which has a much better IMC.

I'm guessing this is because AMD's chiplet design comes with a big latency penalty vs the monolithic design in Alder/Raptor Lake.

Will be interesting to see how Zen 4 performs in 2:1 mode.
 

uzzi38

Platinum Member
Oct 16, 2019
2,705
6,427
146
Intel's memory controller can only do 2:1 ratio and not 1:1 with DDR5, but comparing latency with Zen 4 Alder Lake still comes out on top, let alone Raptor Lake which has a much better IMC.

I'm guessing this is because AMD's chiplet design comes with a big latency penalty vs the monolithic design in Alder/Raptor Lake.

Will be interesting to see how Zen 4 performs in 2:1 mode.
I think if you try running higher frequency memory you'd just drop from the 3:2:2 ratio to 2:1:1 (or 4:2:2 - same thing) I don't think you'd see a whole lot of difference to latency. I don't want to say too much because we've been doing some testing, but actually it seems like our understanding of how Zen reacts to memory OCs even on Zen 3 is quite off tune. Obviously 1:1:1 is ideal, but other neat ratios seem to perform reasonably well.
 

Det0x

Golden Member
Sep 11, 2014
1,299
4,234
136
I think if you try running higher frequency memory you'd just drop from the 3:2:2 ratio to 2:1:1 (or 4:2:2 - same thing) I don't think you'd see a whole lot of difference to latency. I don't want to say too much because we've been doing some testing, but actually it seems like our understanding of how Zen reacts to memory OCs even on Zen 3 is quite off tune. Obviously 1:1:1 is ideal, but other neat ratios seem to perform reasonably well.
With Zen3 i dont actually think it was the asynced fclk:mclk that was the problem contrary to people beliefs.. It was rather that the uclk (memory controller) dropped to half the memory speed when doing so..
On Zen4 we can run mclk and uclk synced (1:1) even if are unsycned with fclk, hence the auto:1:1* setting is recommended
* = fclk:mclk:uclk
1662563925664.png
 
Last edited:

Rigg

Senior member
May 6, 2020
540
1,273
136
Yes but the 12900k can actually support 7000+ memory speeds, probably even 8k with the new hynix A dies.
It looks impressive in AIDA64, GB3 multi-core memory, and Timespy. Not much else.

 

Rigg

Senior member
May 6, 2020
540
1,273
136
Quite the conclusion for such a small battery of tests. There are many other games that show much larger benefits from using high speed DDR5.
Great news! Can you post links to some charts? I agree its hard to draw any concrete conclusions from such a small sample size. I can't find anything else to work with though. The data I linked seems to indicate that if latency is the same or similar the bandwidth doesn't help much over 6000 MT/s. If the game even benefits from more bandwidth to begin with.
 
  • Like
Reactions: ftt and moinmoin