Discussion Intel current and future Lakes & Rapids thread

[Exist50]

Indeed, you were also skeptical about a higher core count Meteor Lake die for the desktop in the absence of concrete details.

That scepticism, at least, is plenty justified.

 

[LightningZ71]

I'm in the camp thatthinks that these two LP E cores are simply just to manage the system when its in a moderate sleep state or deep idling. One of Intel's currently biggest areas of uncompetitiveness is in sleep/standby/isle power usage. Using a pair of Atom cores that are heavily optimized for low power (maybe even in-order also?) To handle those very low power states makes a lot of sense to me.

 

[nicalandia]

I did not see this posted before. List of Sapphire Rapids known models so far

https://twitter.com/x/status/1542359388219793409

 

[thigobr]

Going by the language used on the slides (Low-power island CPU offload) it seems they will use these cores while the main compute die is completely turned off...

 

[Markfw]

6414u is the lowest wattage QS @ 250 watt. The 6314u is 32 core. Think this one is 32 core ?

Edit: I see that column. 350 watt for 56 core ? WOW ! Unless they are smoking fast, they have no chance against , what Genoa ?

 

[aigomorla]

I don't like how that looks for saphire rapids....
It shows the end of HEDT maybe true for both AMD and Intel.
Seems like us filthy gamers are seen as not needing many PCI-E lanes, and only care about overclocking with lots of RGB.

 

[nicalandia]

I don't like how that looks for saphire rapids....
It shows the end of HEDT maybe true for both AMD and Intel.
Seems like us filthy gamers are seen as not needing many PCI-E lanes, and only care about overclocking with lots of RGB.

HEDT has been Dead for a while.....

ThreadRipper Pro was recently released but it's Zen3, not bad at all but Zen3 was released nearly 2 years ago. Ice Lake Xeon Workstation was released 2021 but the Ice Lake core was released on 2019... As you can see pretty dated tech so far.


Also..

Arrow lake



Boyd Immersion Boiler Plates for Intel Sapphire Rapids and AMD Genoa

 

[nicalandia]

What are the chances of Arrow lake Arriving before Sapphire Rapids?


Analysis: Meteor impact coming sooner than thought?

If you recall, Meteor Lake was powered on a couple of months back, and if volume production cranks up at the end of this year as promised, we’re looking at a scenario where the 14th-gen processors could potentially be out in Q2 of 2023, or maybe even Q1 – at a real push.

https://www.techradar.com/news/intels-14th-gen-cpus-could-arrive-soon-after-raptor-lake-keeping-pressure-on-amd

 

[dullard]

Seems like us filthy gamers are seen as not needing many PCI-E lanes, and only care about overclocking with lots of RGB.

To me, it seems like PCI-E lanes have been having the biggest improvements recently.

• Comet Lake was only PCI-E 3.0
• Rocket Lake added PCI-E 4.0 support
• Alder Lake added PCI-E 5.0 support and added 8 total PCI-E lanes
• Raptor Lake is adding additional PCI-E 4.0 lanes on the chipset: https://forums.anandtech.com/attachments/1657223993078-png.64173/
• Meteor Lake seems to be more PCI-E 5.0 and less of the slower lanes

Plus, recent increase in use of Intel Wi-Fi over the CNVi link and you save a PCI-E lane that you would have otherwise used.

Note: the Meteor Lake image discussed above is for mobile chips. The desktop chip with its chipset may have different PCI-E support.

 

[aigomorla]

ThreadRipper Pro

I do not consider this HEDT.
Its a Workstation CPU, not a Enthusiast CPU.
By definition Enthusiast means overclockable Enterprise components.
HEDT term was invented by Intel with the introduction of eVGA's Classified SR-2.
Special server CPU's with unlocked multipliers were introduced just for that platform, and that was the birth of "HEDT"

ThreadRipper PRO does not fill that definition in any form with what HEDT stands for.
Its a Slap to us Enthusiast, telling us either we take it or leave it.

This is one thing i completely disagree with AMD on, they could of just left it as a EYPC as it is, and put it on a workstation line, instead they thought they could milk more profits by sneaking it in the HEDT line, which is very shady.
Even Serve the Home states its a Workstation CPU... not HEDT:

AMD Threadripper Pro is a Workstation EPYC or WEPYC

In what is more of a workstation EPYC or "WEPYC" the AMD Threadripper Pro brings all of the features you wanted in Threadripper and EPYC together

www.servethehome.com

So the last HEDT AMD has had was the 3900X series as that was the last TR which offered overclocking.

The last Intel HEDT cpu was the W-3175 which failed horribly..... but this is intel's last HEDT.

https://www.tomshardware.com/news/intel-eols-its-overclockable-28-core-xeon-w-cpu

Its really sad, they want to kill off the HEDT line from what it looks on both sides, and were left with neutered PCI-E lanes, when every last count since the introduction of nVME's.

To me, it seems like PCI-E lanes have been having the biggest improvements recently.

Having faster lanes is not the same as having the quantity.
Sure you can run a 16x gpu in a 8x, but you CAN NOT run a 8x HBA in a 4x, or a 4x nVME in a 2x.
You also can not run a 8x 10gbe card in a 4x slot.

I can think of many people who have at last 2 nVME's if not more in their system, and that its a pci-e count of 8, which is almost 1/3rd of what you get in a consumer gamer platform.

In today's age, i think a minimum of 48x pci-e lanes should be mandatory, which no system ryzen or intel offers until you go up to HEDT or Workstation Enterprise class.

 

[Exist50]

I did not see this posted before. List of Sapphire Rapids known models so far

View attachment 64171

https://twitter.com/x/status/1542359388219793409

The initial die/configuration explains the core count range and TDPs. Probably in a few months we'll start seeing leaks for lower core count MCC/LCC models. That's also where any HEDT chips would likely be derived from.

Also, is this the first rumor/"confirmation" of Sapphire Rapids being QS ready? Because that's a significant milestone, especially in light of the '23 rumor.

If you recall, Meteor Lake was powered on a couple of months back, and if volume production cranks up at the end of this year as promised, we’re looking at a scenario where the 14th-gen processors could potentially be out in Q2 of 2023, or maybe even Q1 – at a real push.

There's not a chance in heck that Meteor Lake is ready for production by the end of the year, even if Intel 4 is. Probably Q2 for volume production, earliest, and Q3 before we can actually buy any.

I'm not sure if Arrow Lake will take a full year after Meteor Lake, but it's definitely going to be a '24 product.

 

[CakeMonster]

https://www.techradar.com/news/intels-14th-gen-cpus-could-arrive-soon-after-raptor-lake-keeping-pressure-on-amd

On balance, we think Q2 2023 now seems the most likely release date for Meteor Lake – maybe with laptop chips coming first (perhaps by some distance), a possibility another rumor pedlar has floated – in terms of giving Raptor Lake enough breathing space.

Ok, that's cool for laptop users, but doesn't really make much difference for most of the audience here I suppose. I'll probably get Z4, but I'm not crying if it gets blown away by surprise MTL performance less than 12 months later.

 

[LightningZ71]

The really sad part is that HEDT could relatively easily be served by just a chipset for most cases. Most of the clamoring for am HEDT platform is looking for more PCIe lanes. With 16 lanes of PCIe 4 or 5 available from the processor, a chipset with an embedded PLX switch could easily turn that into 32 to 64 PCIe 3 lanes. Just expanding the chipset uplink to 8 lanes of PCIe 5, which is not a crazy request for Zen4 processors, could have been enough to give a bunch of lanes from the chipset without too much link congestion.

A 7950x with good DDR5 could easily drive the same throughput as a 3995x and with the right chipset, give 64 lanes of PCIe 3 on an hedt board.

 

[jpiniero]

Also, is this the first rumor/"confirmation" of Sapphire Rapids being QS ready? Because that's a significant milestone, especially in light of the '23 rumor.

Given that Sapphire was (last year) intended to launch in 1H 22, that there are QS shouldn't be surprising. Guessing Intel will announce the new launch timeframe at the next earnings release.

 

[igor_kavinski]

The really sad part is that HEDT could relatively easily be served by just a chipset for most cases.

Or just a dual socket platform would be great too. Give max 16 cores per socket with 8 memory channels and let the enterprise have the higher end stuff with Epyc/Xeon.

 

[Exist50]

Given that Sapphire was (last year) intended to launch in 1H 22, that there are QS shouldn't be surprising. Guessing Intel will announce the new launch timeframe at the next earnings release.

Kuo was claiming H2'23. You don't have QS samples shipping more than a year ahead of launch. QS means almost final.

 

[DrMrLordX]

ThreadRipper PRO does not fill that definition in any form with what HEDT stands for.

While I generally agree with you, it seems that AMD has made a calculation that either

a). The HEDT market isn't worth the trouble and/or
b). people were using "enthusiast" HEDT parts as workstations (potentially at reduced cost vs. single-socket EPYC workstation products)

I seem to remember a thread in the CPU forums about people using Threadripper to do professional rendering for a movie project, for example. That's lost revenue for AMD and integrators when people can "cheap out" with high-end prosumer gear.

 

[nicalandia]

Looking at Intel Meteor Lake diagram info this is what stands out.

P Cores + E Cores and... + LP E Cores..

Are these LP E Cores just down clocked E Cores?



So far both Quad Cluster look identical



@Exist50 @igor_kavinski

 

[Timmah!]

The really sad part is that HEDT could relatively easily be served by just a chipset for most cases. Most of the clamoring for am HEDT platform is looking for more PCIe lanes. With 16 lanes of PCIe 4 or 5 available from the processor, a chipset with an embedded PLX switch could easily turn that into 32 to 64 PCIe 3 lanes. Just expanding the chipset uplink to 8 lanes of PCIe 5, which is not a crazy request for Zen4 processors, could have been enough to give a bunch of lanes from the chipset without too much link congestion.

A 7950x with good DDR5 could easily drive the same throughput as a 3995x and with the right chipset, give 64 lanes of PCIe 3 on an hedt board.

On topic of this, will i be able to run 2 GPUs and 2 nvme drives on that X670 board with 7950x? Cause that is my intention /need. I mean, without compromised performance anywhere.

 

[jpiniero]

Kuo was claiming H2'23. You don't have QS samples shipping more than a year ahead of launch. QS means almost final.

You do when it gets delayed... again.

 

[Saylick]

Looking at Intel Meteor Lake diagram info this is what stands out.

P Cores + E Cores and... + LP E Cores..

Are these LP E Cores just down clocked E Cores?

So far both Quad Cluster look identical

It's been discussed a few pages back...

But the short and skinny is that compute die has the P cores and the E core clusters, and the LP E cores are on the SOC die, which is on the older node (Intel 7). The idea that's been floating around is either the VPU runs off of those cores (not sure if legit), or that the LP cores are there to run the system if the workload does not require turning on the compute die, thus saving power.

 

[jpiniero]

But the short and skinny is that compute die has the P cores and the E core clusters, and the LP E cores are on the SOC die, which is on the older node (Intel 7).

The SoC is rumored to be using N6 actually.

 

[LightningZ71]

On topic of this, will i be able to run 2 GPUs and 2 nvme drives on that X670 board with 7950x? Cause that is my intention /need. I mean, without compromised performance anywhere.

We know that the ability to havw two x8 slots is being carried forward. If both of those GPUs are PCIe 5, and they can't saturate an x8 pcie 5 link, then you're good there. Unless you're splashing enough money to be able to support a workstation platform, its unlikely that you will hit that limitation. As for the NVME, we know that at least one is directly attached to the processor. There will be at least x4 pcie 4 slot available from the chipset, with a similar link speed to the processor, so, aside from some rare instants where you are really pushing a ton of USB traffic or multiple 10Gb ethernet cards, any bottleneck there should be negligible for one drive on the chipset.

 

[Saylick]

The SoC is rumored to be using N6 actually.

Ehh? So Intel is only making the compute die and the base die for Foveros? That's interesting. It's not like they have a shortage of Intel 7 throughput. I wonder why they want to outsource that to TSMC...

 

[JasonLD]

The SoC is rumored to be using N6 actually.

I thought that was N4 or 5.