Discussion Intel current and future Lakes & Rapids thread

[dmens]

There are rumours Intel target a 2x perf/watt improvement.

LOL. I have a bridge to sell you. Prime location.

 

[Joe NYC]

I absolutely agree when it comes to pricing on DDR5 motherboards however. Those are apparently ridiculous. PCIe5 shouldn't affect prices too much as it's only the GPU lanes that carry the signal - primary NVMe and chipset both use Gen 4 lanes.

Is anyone making PCIe5 graphics cards anytime soon?

I am just guessing here, but my guess is that we will not see PCIe5 graphics cards until the next generation graphics cards come out - in late 2022?

 

[scannall]

Is anyone making PCIe5 graphics cards anytime soon?

I am just guessing here, but my guess is that we will not see PCIe5 graphics cards until the next generation graphics cards come out - in late 2022?

It will be more useful for storage.

 

[RTX2080]

This translation below tells us more about ADL power.

In short,

TDP 165w SKUs need peak of 12V2 at 45A/540w, increased from 40A/480w
TDP 125w SKUs need peak of 12V2 at 39A/468w, increased from 34A/408w
TDP 65w SKUs need peak of 12V2 at 38.5A/462w

At last you need extra 50w-100w power budget when build your rig.

I'm not expert of power supply though, but it sounds absurd to me

edit: ouch, already discussed

https://twitter.com/x/status/1423352239817596930

 

[uzzi38]

It will be more useful for storage.

Except ADL doesn't have any Gen 5 lanes for storage unless you use the GPU lanes

 

[JoeRambo]

Except ADL doesn't have any Gen 5 lanes for storage unless you use the GPU lanes

Frankly irrelevant. AMD just released 5XXXG CPUs that do not support PCIE4 at all and noone cares nor should they.
What is important for enthusiasts is that ADL has 4x direct CPU lanes, GEN4 is perfectly fine. And those GEN5 GPU lines will go unused for at least year+.

 

[uzzi38]

Frankly irrelevant. AMD just released 5XXXG CPUs that do not support PCIE4 at all and noone cares nor should they.
What is important for enthusiasts is that ADL has 4x direct CPU lanes, GEN4 is perfectly fine. And those GEN5 GPU lines will go unused for at least year+.

???

I was just pointing out that Gen 5 is going to be a gimmick for ADL at best. The PCIe 5 lanes on the GPU are frankly useless for consumers and the direct storage lanes - the place we should see an actual benefit to Gen5 - are still Gen 4.

 

[mikk]

LOL. I have a bridge to sell you. Prime location.

Almost a Cinebench R20 doubling over RKL-S with a slightly decreased PL2 (251 vs 228W), I know this is not what you would like to hear but you should think about it. Probably not for all SKUs but in some segments 2x perf/watt isn't far-fetched. The prime example might be ADL-P U28 6C+8C versus TGL-U 4C 28W.

 

[lobz]

Almost a Cinebench R20 doubling over RKL-S with a slightly decreased PL2 (251 vs 228W), I know this is not what you would like to hear but you should think about it. Probably not for all SKUs but in some segments 2x perf/watt isn't far-fetched. The prime example might be ADL-P U28 6C+8C versus TGL-U 4C 28W.

Then again, it's like vega vs navi. They have a whole lot of ground to make up, meaning it's cool and utterly necessary after being so inefficient on the desktop for years, but nothing groundbreaking. Anything less would be viewed as an utter failure and with good reason. Being an arrogant winner can be annoying as hell, but being a deluded and really arrogant loser is so repelling that most of the time these days, the tech press don't even give Intel the benefit of the doubt - I'm not happy about that, believe me or not

 

[dmens]

Almost a Cinebench R20 doubling over RKL-S with a slightly decreased PL2 (251 vs 228W), I know this is not what you would like to hear but you should think about it. Probably not for all SKUs but in some segments 2x perf/watt isn't far-fetched. The prime example might be ADL-P U28 6C+8C versus TGL-U 4C 28W.

LOL. You decide the bring up the one workload (AVX enabled R20) where Rocketlake decided to kill the planet and claimed double perf-per-watt. Nice cherry-picking. By the way, you should get your gullibility meter checked on that R20 score claim.

Moreover, you should think about how thermally crippled those 6+8 cores will be at 28W once everything is settled and how much performance they can put out compared to 4 big cores on TGL getting ~5.5 watts apiece.

 

[SAAA]

Almost a Cinebench R20 doubling over RKL-S with a slightly decreased PL2 (251 vs 228W), I know this is not what you would like to hear but you should think about it. Probably not for all SKUs but in some segments 2x perf/watt isn't far-fetched. The prime example might be ADL-P U28 6C+8C versus TGL-U 4C 28W.

I don't have doubts a "7 nm" port of RKL-S would get >50% perf/watt increase either, just look at Rocket vs Tiger on 10SF for some numbers, now count higher IPC plus small cores and it's definitely not out of reality to get 2x perf/watt even running at max stock performance.
Indeed lower PL2 is all that counts for long term power usage and this is going down, a good sign for once. Those 10 ms higher peaks might as well be neutered by good capacitors on MB and PSU, that I would advice for anyway.

As for comparison I think most will be done with the previous -S chips to put them in a better perspective, mobile sort of comes later and will have less noticeable gains, but gains nonetheless.

 

[mikk]

I don't have doubts a "7 nm" port of RKL-S would get >50% perf/watt increase either, just look at Rocket vs Tiger on 10SF for some numbers, now count higher IPC plus small cores and it's definitely not out of reality to get 2x perf/watt even running at max stock performance.
Indeed lower PL2 is all that counts for long term power usage and this is going down, a good sign for once. Those 10 ms higher peaks might as well be neutered by good capacitors on MB and PSU, that I would advice for anyway.

As for comparison I think most will be done with the previous -S chips to put them in a better perspective, mobile sort of comes later and will have less noticeable gains, but gains nonetheless.

I don't even think mobile would have less noticeable gains, it really depends on the exact SKU/segment. TGL-U 4C 28W is highly inefficient because it's high clocked with a low core count, this is really bad in multithread like Cinebench. On ADL-P they can go much much lower with the clock speeds because they have at best 14 cores (6C+8C) instead of 4.

 

[Joe NYC]

It will be more useful for storage.

That makes sense.

 

[Joe NYC]

Except ADL doesn't have any Gen 5 lanes for storage unless you use the GPU lanes

That's a bummer, but you can probably get a $20 adapter card.

But then you would be taking lanes away from GPU. So all around, a suboptimal solution....

 

[Joe NYC]

Recently, someone on reddit by the username of mooreslawisnotdead posted an Intel roadmap all the way through 2025. This roadmap contains some products/code names that haven't been mentioned here before. The roadmap could be completely genuine or a total work of fan fiction. I guess that's for you to decide. The user has since deleted his account, but not before I copied this information. What follows is the roadmap itself, along with commentary by the reddit user that posted it:

---------------------------------------------------------------------------------------------
Alder Lake (Golden Cove/Gracemont) Q4'21 / Q1'22 - predicted to be competitively weak vs AMD/Apple offerings that time.
Raptor Lake (Raptor Cove / Gracemont) Q3'22 / Q4'22 - 10% CPU perf boost and 8/16 configuration puts intel back on par but expect AMD/Apple to refresh their products as well.
Meteor Lake (Redwood Cove / Crestmont) Q2'23 - Intel's first true chiplet or tile based design. Different dies built on TSMC / Intel processes. More of a node shrink with single digit performance improvements. AMD will again extend lead with Zen 4+ / 5.
Arrow Lake (Lion Cove / Skymont) Q4'23 - Will feature an updated compute tile with 8/32 config for the high end enthusiast products. Might achieve parity with AMD offerings at the time but loses out to Apple in power efficiency.
Lunar Lake (Lion Cove / Skymont) Q4'24 - This is the product that will use TSMC 3nm as reported by Nikkei. Big performance jump expected and designed to achieve parity or beat AMD and Apple in both performance and power efficiency.
Nova Lake (Panther Cove [tentative]/ Darkmont) 2025 - This will mark the biggest architectural change in cpu architecture since the Core architecture is introduced in 2006. Intel is working to build an entirely new architecture from the ground up much like Ryzen with up to 50% cpu performance improvement from lunar lake. This is also the reason why Glenn Hinton returned.
---------------------------------------------------------------------------------------------

More mentions, you opened a floodgate:

 

[mikk]

Adore has zero credibility when it comes to Intel. 6C+8C for a Halo product when we already get 6C+8C with ADL-P and by the looks of it a little core increase with Raptor Lake up to 16C? That makes no sense. Furthermore he claims CPU+GPU are made with TSMC 3nm. This is the newest info about TSMC 3nm and Intel:

https://twitter.com/x/status/1424897727469555732

No client CPU, only server and probably dedicated graphics. So this claim is unrealistic as well.

 

[DrMrLordX]

So they're moving Granite Rapids to N3? If that leak is to be believed.

 

[RanFodar]

^ Maybe the introduction of tiles like in Meteor Lake/Ponte Vecchio, with some produced externally?

 

[DrMrLordX]

^ Maybe the introduction of tiles like in Meteor Lake/Ponte Vecchio, with some produced externally?

Possibly. 10kwpm doesn't give them a lot of silicon to play with. But I think they will be at maybe 10-15kwpm on . . . Intel 7nm/Intel 4 so that would be a substantial increase in available wafers at that time.

That being said, full production in July 2022 would be really aggressive since Sapphire Rapids got delayed. That seems really early for Granite Rapids.

 

[jpiniero]

That being said, full production in July 2022 would be really aggressive since Sapphire Rapids got delayed. That seems really early for Granite Rapids.

Remember you have to add 3+ months from when production starts. Could do something like sell to some HPC/Cloud/Facebook.

 

[mikk]

ADL-S mass production starts mid August, in store probably in November or late October in best case. Server has a bigger delay I would guess, not before late 2022 or early 2023.

 

[jpiniero]

Please click the green human to continue - UserBenchmark

Gigabyte laptop with 6 big and 8 small cores spotted. Score sucks so it's likely just an ES for testing.

 

[Doug S]

Adore has zero credibility when it comes to Intel. 6C+8C for a Halo product when we already get 6C+8C with ADL-P and by the looks of it a little core increase with Raptor Lake up to 16C? That makes no sense. Furthermore he claims CPU+GPU are made with TSMC 3nm. This is the newest info about TSMC 3nm and Intel:

https://twitter.com/x/status/1424897727469555732

No client CPU, only server and probably dedicated graphics. So this claim is unrealistic as well.

If Intel is getting only 10K wpm from N3, that is less than 10% of TSMC's claimed mass N3 production numbers - they've stated they would be doing 30k wpm in risk production, and 105k wpm in mass production (maybe not all on day one of mass production, but I'd assume they'd ramp from 30K to 105K relatively quickly)

The rumors that Apple would be using N4 instead of N3 seemed to center on TSMC's statements of "H2 2022" for N3 mass production. July is "H2" but is in plenty of time for Apple, and they are the only feasible customer for the majority of those early N3 wafers with Intel only taking 10k.

So I think the idea that Apple will be using N4 instead of N3 for next year's iPhone can be put to bed if that tweet is at all accurate. They'd have have enough A16s if TSMC was only doing 50K wpm on day one of mass production, once they reach 105k they'll have quite a bit more than Apple needs and already be on other customers.

 

[Hitman928]

If Intel is getting only 10K wpm from N3, that is less than 10% of TSMC's claimed mass N3 production numbers - they've stated they would be doing 30k wpm in risk production, and 105k wpm in mass production (maybe not all on day one of mass production, but I'd assume they'd ramp from 30K to 105K relatively quickly)

The rumors that Apple would be using N4 instead of N3 seemed to center on TSMC's statements of "H2 2022" for N3 mass production. July is "H2" but is in plenty of time for Apple, and they are the only feasible customer for the majority of those early N3 wafers with Intel only taking 10k.

So I think the idea that Apple will be using N4 instead of N3 for next year's iPhone can be put to bed if that tweet is at all accurate. They'd have have enough A16s if TSMC was only doing 50K wpm on day one of mass production, once they reach 105k they'll have quite a bit more than Apple needs and already be on other customers.

Based upon their past nodes, I would expect TSMC to start N3 mass production with ~50k wpm, adding an additional ~30k wpm after the first year, and hitting a total of over 100k wpm in year 3.

 

[Joe NYC]

^ Maybe the introduction of tiles like in Meteor Lake/Ponte Vecchio, with some produced externally?

It could be related to Ponte Vecchio and Aurora, which Intel is having the hardest time to complete.