MP All Models Is Apple ****ing themselves over the ThreadRipper 3 yet?

[koyoot]

https://twitter.com/i/web/status/1241727849431785472

Intel Comet Lake successor Rocket Lake to get PCIe Gen 4.0 Support

So the story goes like this, Comet Lake, the pending to be released Intel desktop processors up-top 10 cores, originally had a PCIe Gen4 design, but with signal validation issues, Intel is sticking to...

www.guru3d.com

I suppose the naysayers will still believe 10 nm is not cancelled, and that we will see large dies, especially that Rocket Lake(Willow Cove) is backport from 10 nm to 14 nm process. I told you about it before, and sources were right, once again.

Good luck arguing otherwise. Charlie once again is right. For the last time. Intel cannot deliver small 10 nm parts in volume and people believe 10 nm Server, large dies will deliver. It ain't happening.

 

[Mago]

Layoff at Intel should begin at their CEOs staff, a shame the most solid silicon manufacturer running in circles due managers irresponsible ambition delaying investment on "Moore law is dead" (actually, not Dead, they loved to have killed it and save on investment for bigger gain share,) however really -AMD and ARM- was about the corner to bring them a hard lesson on how not to manage an corporation.

 

[koyoot]

Layoff at Intel should begin at their CEOs staff, a shame the most solid silicon manufacturer running in circles due managers irresponsible ambition delaying investment on "Moore law is dead" (actually, not Dead, they loved to have killed it and save on investment for bigger gain share,) however really -AMD and ARM- was about the corner to bring them a hard lesson on how not to manage an corporation.

They already did laid off their CEO. Brian Krzanich, reminds you of anything?

Moore's law is not dead. Its just Intel's incompetence and pure arrogance that has killed it for them. Moore's law is still alive at TSMC.

Nobody has ever returned to process node leadership once they have lost it. Intel will never have process and technology leadership, ever again.



Continuing the topic of RKL:

https://twitter.com/i/web/status/1241754120060571648

Adding salt to some naysayers injuries. I was right. You were wrong. About 10 nm server chips everybody who say that they are dead are also right. And you are wrong. Its just a matter of time.

 

[Mago]

Nobody has ever returned to process node leadership once they have lost it.

I'll don't count on that, Intel had a management disturbance, but with regards AMD and every ARM manufacturer, Intel has capital enough to self rebuild or reinvent from total crashes many times.

I mean the problem is not about 10nm or 7nm process mastering, I believe they absolutely control it, it's matter of product diversity, as you shrink the node more time you need to adapt your IP to the new node, even without adding a single new feature, it's know intel maintain a miriad of different CPUs addressing markets where hardly it's justified it's development, these CPU exhaust it's r&d throughput capacity, that's why AMD's chiplets strategy pays, rationalizing r&d, AMD barely develop 2 o 3 different chiplets at 7nm and the interconnect chip at 10-14nm, a huge effort it's saved this way, despite chiplets performance disadvantage, it's the new model, Intel's new 5core CPU for 5G station follows this strategy using the smaller node where it pay and allowing older nodes where its good enough.

Intel has lot of work and investment due but I'm confident as long they allows Jim Keller to rebuild its future.

Said that, I believe intel will lose apple's exclusivity to AMD but they should recover part of it later, maybe as soon as in 2 year, ARM CPU in Mac s after seeing the new iPad pro seems more an unicorn and fanboy click bait than a real option for an better product.

The new iPad pro practically put the shinny new MBA in shame at lower price, later as more apps follow the catalyst model hardly an MacBook will be an option, much less having the same CPU, simple don't worth it's market an ARM mac.

 

[JayKay514]

Not to beat a dead thread but...

I'm going AMD + Windows for my next DAW workstation, purely because Apple doesn't offer anything with a decent price/performance ratio anymore.

I stuck with my classic 5,1 12-core MP for a long time until it was no longer supported.

The new Mac Pro is truly magnificent, but it's really designed for video editors, and at $7,499 CAD for an 8-core starter configuration it's just way out of my price range.

Regular iMacs, Mac Minis and MBPs are underpowered for what I need, and the iMac Pro is not expandable.

DAW work is really one-thread-per-core CPU-bound, though you can offload some things to DSP cards if you buy into particular platforms (UAD, ProTools). So what I need is not gamer-style turbo on one or two cores, but relatively good performance across all cores, and as many cores as I can get.

The other thing is expansion, and for audio recording and playback, that means having lots of PCIe lanes, faster lanes, and much more RAM / memory channels.

I started looking across the aisle and AMD's EPYC platform started to look a lot more attractive.

I'm speccing out a system build now that will have 16 or 24 cores, 128 PCIe 4.0 lanes that feed 7 PCIe x16 slots, dual PCIe M.2 drives to start, using a basic Radeon Pro WX2100 card as I don't need superduper graphics.

This will give me the capability to upgrade to 32 or 64 cores in the future, up to 2 TB of RAM, add up to four UAD-2 Octo DSP cards, more M.2 NVME drives on a PCIe expansion card, and maybe additional USB 3.1 Gen 2 on a PCIe card as well.

Yes, it's missing Thunderbolt, but I think I can live without it. If and when future platforms integrate USB 4 then I'll consider it (if it's available as an AIC that'd be cool).

Calculated out, my starter machine will give me way more power than the entry-level Mac Pro, for about $2000+ less.

Yeah, I'll miss macOS, but I guess I'm at that point where I'm no longer in their target market, and TBH Windows these days works remarkably well, even for things that used to be Mac strengths like AV production.

If Apple offered a Threadripper-based high-end desktop with integrated Thunderbolt - something like the TRX40 platform but better? - for about $2500-3000 starting price then I'd think about it.

 

[mikas]

Well, wouldn't that be a xMac. And unfortunately cost 6,999$ for starters. Not 3000$.

 

[Krevnik]

I'll don't count on that, Intel had a management disturbance, but with regards AMD and every ARM manufacturer, Intel has capital enough to self rebuild or reinvent from total crashes many times.

Yeah, the thing is, AMD isn’t even growing quickly enough to throttle Intel’s money supply yet. They’re a threat now, but there’s market segments they are only starting to go after now they have a competitive foothold in the door on Desktop and up. Problem is that Desktops and Workstations aren’t exactly the strongest part of the PC market right now.

So Intel is in position to have a surprising amount of time to fix things before it really gets gruesome for them. Could we see another Pentium IV to Core Solo/Duo type re-engineering of their product line? Maybe.

ARM CPU in Mac s after seeing the new iPad pro seems more an unicorn and fanboy click bait than a real option for an better product.

I agree with the sentiment, but for different reasons. I think the new iPad Pro demonstrates that Apple’s hitting diminishing returns on the A series at the iPad’s TDP . And that doesn’t necessarily provide a good sign for the scaling up they need to do to compete with Intel/AMD on their home turf.

I think the fact that the SoC is so little changed from 2018 tells me that Apple going AMD is more likely than ARM at the moment.

 

[boyorion]

IMHO. many Apple users may not care what CPU is inside (AMD, Intel, ARM) just as long as stability and performance meet expectations.

If Apple is locked to Intel they might be truly f**ked.

Don't care? Most have no idea.

 

[deconstruct60]

....
Good luck arguing otherwise. Charlie once again is right. For the last time. Intel cannot deliver small 10 nm parts in volume

Chuckle.

Intel’s new ‘Single Customer’ Ice Lake Mobile CPUs: Are These in the Macbook Air?

www.anandtech.com

So what Charlie is 'brilliantly' right about is that the MBA is not shipping in volume. Ordering it in the middle of a pandemic and getting an arrive date in the span of a week is something that isn't in any way representative of a volume product. ( currently the shipping time for a MBA is a couple days shorter than a Mac Pro. ... but yeah ... not a volume product. *cough* ).


Dell XPS 13. ... not a volume product either.

Lenovo C940 .. not a volume product

Some variants of Surface Laptop 3. ... not. volume product
HP Spectre. x360 2 in 1. ... not volume product.

That three of those were on Best Buy show room floors before the pandemic fully hit ... not a indication of volume. That will be > 4 10nm powered solutions on the show room floor when the closed stores reopen for customers again in upcoming weeks. Again supposedly representative that 10nm won't ship in volume at all.


Intel is keeping 14nm in production to fill a larger fraction of their overall orders. That just means the volume of 10nm is lower than 14nm. That doesn't mean that 10nm isn't shipping in volume. That latter is more so just trying to move the goal posts so can sprinkle "I'm infallible" spin on the situation.

and people believe 10 nm Server, large dies will deliver. It ain't happening.

Intel cancelled a 14nm Server part in the highly overlapping same space had a 10nm part ~6 months later . So if had canceled both the 14nm and 10nm part they'd have what? An even older 14nm part?

Intel is keeping the 14nm part in the 4 socket space. That gives a 3x multiple for every 4 socket system for additional 10nm wafers that can be used for single socket systems. Makes up completely incrementally lower yields, perhaps not, but it does cover a gap.

The volume of the MBA + Dell XPS 13 + several other laptops in that price zone is probably in the same ballpark as the number of. server CPUs shipped in the single socket space. Yes the die size is bigger, but the price point is higher too. ( so use more wafers but can pay for more wafers). Will Intel be about to almost print money with super fat profit margins? Probably not. Can Intel fab enough to hit orders? Not so conclusive from these "sky is falling" reports.

The base reality is that Intel does not have to hit PC mainstream , average selling price system's volume levels to service the volume of the core of the low socket count server market. If AMD takes incremental share.. then incrementally easier for Intel to cover what they have. If the global economy goes into recession and folks buy less. Easier still for a supposedly "too limited" volume level to actually work.




P.S. That there are packaging variants of a 10nm product is also yet another indicator of substantive volume. If "nobody" is buying it not too likely would put in extra work for another package variation.

 

[koyoot]

Chuckle.

Intel’s new ‘Single Customer’ Ice Lake Mobile CPUs: Are These in the Macbook Air?

www.anandtech.com

So what Charlie is 'brilliantly' right about is that the MBA is not shipping in volume. Ordering it in the middle of a pandemic and getting an arrive date in the span of a week is something that isn't in any way representative of a volume product. ( currently the shipping time for a MBA is a couple days shorter than a Mac Pro. ... but yeah ... not a volume product. *cough* ).


Dell XPS 13. ... not a volume product either.

Lenovo C940 .. not a volume product

Some variants of Surface Laptop 3. ... not. volume product
HP Spectre. x360 2 in 1. ... not volume product.

That three of those were on Best Buy show room floors before the pandemic fully hit ... not a indication of volume. That will be > 4 10nm powered solutions on the show room floor when the closed stores reopen for customers again in upcoming weeks. Again supposedly representative that 10nm won't ship in volume at all.


Intel is keeping 14nm in production to fill a larger fraction of their overall orders. That just means the volume of 10nm is lower than 14nm. That doesn't mean that 10nm isn't shipping in volume. That latter is more so just trying to move the goal posts so can sprinkle "I'm infallible" spin on the situation.




Intel cancelled a 14nm Server part in the highly overlapping same space had a 10nm part ~6 months later . So if had canceled both the 14nm and 10nm part they'd have what? An even older 14nm part?

Intel is keeping the 14nm part in the 4 socket space. That gives a 3x multiple for every 4 socket system for additional 10nm wafers that can be used for single socket systems. Makes up completely incrementally lower yields, perhaps not, but it does cover a gap.

The volume of the MBA + Dell XPS 13 + several other laptops in that price zone is probably in the same ballpark as the number of. server CPUs shipped in the single socket space. Yes the die size is bigger, but the price point is higher too. ( so use more wafers but can pay for more wafers). Will Intel be about to almost print money with super fat profit margins? Probably not. Can Intel fab enough to hit orders? Not so conclusive from these "sky is falling" reports.

The base reality is that Intel does not have to hit PC mainstream , average selling price system's volume levels to service the volume of the core of the low socket count server market. If AMD takes incremental share.. then incrementally easier for Intel to cover what they have. If the global economy goes into recession and folks buy less. Easier still for a supposedly "too limited" volume level to actually work.




P.S. That there are packaging variants of a 10nm product is also yet another indicator of substantive volume. If "nobody" is buying it not too likely would put in extra work for another package variation.

https://twitter.com/i/web/status/1243956644871991296

Here is breakdown on Rocket Lake info, from very reputable leakers source: Sharkbay.

RKL is a backport, because 10 nm is a turd. Period. It appears that RKL is slightly modified Sunny Cove core on 14 nm process, with absurd power delivery required(well, who expected anything different?). Those 8 core chips will be 300W PL2.

Don't expect and believe in Intel's roadmaps anymore. They were supposed to deliver 14 nm server chips, and haven't and yet you believe they will deliver 10 nm Server parts?

Hah.

 

[toles]

Well, another year is here and for DAW it will be the biggest in a decade. DDR5, PCIe 5.0 and CXL/CCIX will all come bundled together on both AMD and Intel platforms. Regardless of who out performs who, both will have DDR5 which boasts a lofty 2x MT/s. The PCIe 5.0 is notable, but for DAW it's largely overkill due to current NVMe speeds and there will not be RAW video to max out even PCIe 4.0 for a very (very) long time.

The big surprises will come in the CXL/CCIX applications. This specifically is worth reading about elsewhere as the potential parallelism at the rendering level could set new standards for performance similar to how SSE and MMX once did. High noise reduction at 1:1 decode now looks possible with less VRAM and if this possibility becomes a reality, then CPU's directly accessing GPU VRAM now has very serious potential in DAW (this method was introduced with AMD's ZEN 3 and Intel will soon incorporate).

The safe bet for DAW is what was the obvious choice through all of 2020, AMD. However, all eyes are on Intel's Sapphire Rapids as this will decide whether or not Intel is in need of a Band-Aid or stitches. But regardless of how bad Intel is down, one thing is sure to be down, price. If AMD's Genoa meets only 75% of the rumors, Intel can't afford to stand on their prices like they have for the last 3 years, Intel will have to smelter a silver spoon to a silver bullet for targeting AMD's price to performance.

I am poor but, for those of you that can afford Genoa or SR... have fun storming the castle!