Discussion Intel current and future Lakes & Rapids thread

[Exist50]

According to Moore's Law Is Dead. Intel has cancelled the Discrete ARC GPUs

I highly doubt that. And even if they do cancel a consumer Alchemist release (still doubt), there's no chance they cut the program entirely, so it doesn't fit the statement.

 

[nicalandia]

I highly doubt that. And even if they do cancel a consumer Alchemist release (still doubt), there's no chance they cut the program entirely, so it doesn't fit the statement.

MLID cites a source at the executive level who claims, rather brutally, that Arc is “finished”, and another source who says simply that Intel is ending its Arc discrete efforts.

 

[Exist50]

MLID cites a source at the executive level who claims, rather brutally, that Arc is “finished”, and another source who says simply that Intel is ending its Arc discrete efforts.

My toilet paper has more insight than most of MLID's "sources".

 

[IntelUser2000]

Intel doesn't seem to think their server market share will improve till '25 or even '26. Has some interesting implications for GNR...

Also, they expect to cut more businesses. Not sure what's left at this point, tbh. Losing networking or wireless could be a disaster.

AMD also seems to be getting mindshare. If you take his words, 2024 is competitive but 2025 is leadership. So if GNR is slightly behind or at best equal, there's no reason for sustained marketshare gains. It certainly won't be like now whether there's no hope of catching up in general purpose workloads.

Let's see. FPGA, NUC, server racks, 5G modems for laptops?, eASICs.

So a guy that claimed to have worked with Gelsinger before said he's very good at cutting to keep businesses and focused at core competencies. He said himself they want to focus on logic*.

Are dGPUs a distraction? Perhaps with him at the helm "dGPUs" will be things like Falcon Shores and for client big, big tile GPUs?

*Regarding Optane he said "Intel keeps exiting memory. I will close the freaking door on it."

 

[FangBLade]

I doubt Intel will shut down Arc, yes first generation is fail, they spent billions into it while they could spent on cpu division, but because they invested so much money and resources, and also because there is so much space for third player, they will continue with it.

 

[Exist50]

AMD also seems to be getting mindshare. If you take his words, 2024 is competitive but 2025 is leadership. So if GNR is slightly behind or at best equal, there's no reason for sustained marketshare gains. It certainly won't be like now whether there's no hope of catching up in general purpose workloads.

The problem is that GNR will, at least nominally, have a process advantage. vs Turin And yet Intel still doesn't think that's enough to gain marketshare? I had been assuming that Intel switched GNR to Lion Cove, but now I'm starting to wonder...

Let's see. FPGA, NUC, server racks, 5G modems for laptops?, eASICs.

NUCs and server racks, I could see. Small things that don't really matter. I guess eASICs too, but they seem at least a little more fundamental. 5G modems, they barely seem involved in any more.

FPGA would be problematic, I think, especially in light of AMD acquiring Xilinx. Really, the problem as far as I see it is that Optane was losing a ton of money, and NAND kind of alternating between profitability and not. But most of the rest is both smaller, and/or actually profitable. So I do have concerns about how far Gelsinger plans to take things...

So a guy that claimed to have worked with Gelsinger before said he's very good at cutting to keep businesses and focused at core competencies.

Do you have a source for that quote, perchance?

Are dGPUs a distraction? Perhaps with him at the helm "dGPUs" will be things like Falcon Shores and for client big, big tile GPUs?

I really don't see a future for Intel in the performance GPU space if they abandon Arc. A large tile GPU would basically be the same thing, so there's no relief there, and without the scale and ubiquity of the consumer market, what would make Intel stand out in the datacenter?

 

[IntelUser2000]

I doubt Intel will shut down Arc, yes first generation is fail, they spent billions into it while they could spent on cpu division, but because they invested so much money and resources, and also because there is so much space for third player, they will continue with it.

Spending only gets you so far. Rest is hard work, ingenuity and time.

@Exist50 I don't know. Unless he was really talking about those small businesses. There is merit to people saying Gelsinger has to see ARC succeed since he got fired working on Larrabbee. But is that logical and/or will he keep it?

It was from a hardware forum. Don't know if you still want to read it.

As for GNR again who knows? I remember someone that put out a "roadmap" of how the lineup will be for Intel in the future and Granite Rapids was listed as having Granite Cove cores. I thought that was interesting but in light of what was said recently perhaps it's a boosted Redwood Cove over Lion Cove that will go in Arrowlake?

 

[Exist50]

@Exist50 I don't know. Unless he was really talking about those small businesses. There is merit to people saying Gelsinger has to see ARC succeed since he got fired working on Larrabbee. But is that logical and/or will he keep it?

If nothing else, I don't think he'll give up on it so soon. It's too big a bet to be so skittish about. If Intel does give up on discrete GPUs, I suspect it will be years down the line.

As for GNR again who knows? I remember someone that put out a "roadmap" of how the lineup will be for Intel in the future and Granite Rapids was listed as having Granite Cove cores. I thought that was interesting but in light of what was said recently perhaps it's a boosted Redwood Cove over Lion Cove that will go in Arrowlake?

"Granite Cove" certainly isn't a thing. But if it does use Redwood Cove, even an improved version on Intel 3, that will bode very, very poorly.

 

[DrMrLordX]

The problem is that GNR will, at least nominally, have a process advantage. vs Turin And yet Intel still doesn't think that's enough to gain marketshare?

That assumes it reaches market in a timely fashion. And as @Exist50 points out, Redwood Cove would be a bit long in the tooth by that time. Let's hope that's not the case.

 

[RTX]

Is the 14900K's IGP going to be at least 1650 Mhz? The 13900K's IGP is listed as running at 1650 Mhz

https://www.tomshardware.com/news/raptor-lake-igpu-catches-up-to-amd-vega-10-in-opengl-benchmark

 

[SteinFG]

Is the 14900K's IGP going to be at least 1650 Mhz? The 13900K's IGP is listed as running at 1650 Mhz

https://www.tomshardware.com/news/raptor-lake-igpu-catches-up-to-amd-vega-10-in-opengl-benchmark

nobody knows, and honestly, I don't think anone cares for iGPU perf in i9K

 

[DrMrLordX]

Is the 14900K's IGP going to be at least 1650 Mhz? The 13900K's IGP is listed as running at 1650 Mhz

What even will the 14900k be? Arrow Lake?

 

[jpiniero]

The problem is that GNR will, at least nominally, have a process advantage. vs Turin And yet Intel still doesn't think that's enough to gain marketshare? I had been assuming that Intel switched GNR to Lion Cove, but now I'm starting to wonder...

You should take this as a hint that even they aren't expecting to be able to do much with 7 nm, despite all the cheerleading.

 

[witeken]

"Granite Cove" certainly isn't a thing. But if it does use Redwood Cove, even an improved version on Intel 3, that will bode very, very poorly.

Granite Rapids uses Lion Cove and will have 112-120 cores.

 

[mikk]

Is the 14900K's IGP going to be at least 1650 Mhz? The 13900K's IGP is listed as running at 1650 Mhz

https://www.tomshardware.com/news/raptor-lake-igpu-catches-up-to-amd-vega-10-in-opengl-benchmark

I'm expecting higher clocks which are more in-line to current RDNA2 iGPU clock speeds. Xe HPG has a 50% higher clock speed scaling on TSMC 6nm and TSMC 5nm clocks even higher. Even the 1650 Mhz are quite low compared to AMD iGPUs.

 

[Exist50]

You should take this as a hint that even they aren't expecting to be able to do much with 7 nm, despite all the cheerleading.

Nah, if they couldn't ship it in large volumes, they'd continue to lose a lot of market share. I think this points to some performance issue with GNR more than anything else.

 

[Exist50]

Granite Rapids uses Lion Cove and will have 112-120 cores.

Originally, I would have agreed. Now I'm not so sure.

 

[uzzi38]

The problem is that GNR will, at least nominally, have a process advantage. vs Turin And yet Intel still doesn't think that's enough to gain marketshare? I had been assuming that Intel switched GNR to Lion Cove, but now I'm starting to wonder...

I just think GNR is a little bit later for mass availability than first assumed personally.

 

[uzzi38]

I'm expecting higher clocks which are more in-line to current RDNA2 iGPU clock speeds. Xe HPG has a 50% higher clock speed scaling on TSMC 6nm and TSMC 5nm clocks even higher. Even the 1650 Mhz are quite low compared to AMD iGPUs.

I don't know about you but I'd personally want to make assumptions on this once someone does some more in-depth testing on power at varying clock speeds. I'd do it myself like I did for RDNA2 but I haven't seen an A380 on sale in the UK yet...

 

[igor_kavinski]

My prediction is, Intel may try to do something "desperate", like win back the enthusiasts it lost to AMD, by releasing cheaper CPUs with more cores than AMD. That's the only thing they have left in their warchest. Everything else is too far away for them and they can't afford to bleed money for two straight years. It's gonna rain Intel soon, folks. Mark my words (feel free to PM me offering donations if you agree ).

 

[Markfw]

https://www.tomshardware.com/news/intel-expects-continued-market-share-loss-throughout-2023-will-likely-exit-more-businesses

Intel doesn't seem to think their server market share will improve till '25 or even '26. Has some interesting implications for GNR...

Also, they expect to cut more businesses. Not sure what's left at this point, tbh. Losing networking or wireless could be a disaster.

This quote below, I just don't get, tell me what I am missing. AMD has it all over SR, Milan does and Genoa is even better yet.

"Intel's recently delayed its Sapphire Rapids launch again, with the chips now slated to arrive in 2023. Gelsinger remarked that while the new chips are 'better than the AMD alternatives" in power and performance and will win in some benchmarks, the advantages aren't dramatic enough to slow AMD's advance. As a result, Intel's data center business won't grow at the same rate as the market, meaning the company will continue to lose market share. "

 

[igor_kavinski]

Gelsinger remarked that while the new chips are 'better than the AMD alternatives" in power and performance and will win in some benchmarks,

AVX-512 and AMX are two things that will help Intel win some benchmarks and draw less power doing so compared to AMD. Remember, there are Intel optimized workloads where AMD can't do anything yet. Also, their HBM CPU SKUs will also be good at some workloads and may make sense for some customers more than AMD alternatives. AMD is focusing on brute performance whereas Intel, like Nvidia, is trying to capitalize on its proprietary technologies.

 

[uzzi38]

You should take this as a hint that even they aren't expecting to be able to do much with 7 nm, despite all the cheerleading.

I still expect GNR to be competitive against Turin with regards to raw performance. I'm just worried about timelines. It slipping to 2025 isn't exactly ideal for significant availability, if that's what's going on.

 

[uzzi38]

Interestingly this review shows Alderlake-U 9W having great battery life. Greater than 14.5 hours using 51WHr battery is fantastic.

While the reviewer criticizes it having low performance, the MT performance rivals 15W Tigerlake despite only having 10W sustained. Also unlike Amberlake and previous low TDP attempts, this chip does not suffer from noticeably lower single thread results. Single thread performance is only 10% behind top Tigerlake which is an excellent result. This is twice the MT performance of Y chips, and those chips had MT performance at a standstill since the Broadwell Core M.

The reviewer conclusion is as always with the modern press, mediocre at the best. I don't remember them complaining when Tigerlake had 2/3rd MT performance of Renoir. Why the change in attitude now?

https://www.notebookcheck.net/Dell-XPS-13-9315-laptop-in-review-Low-performance-incredible-battery-life.640173.0.html

This is Amberlake class efficiency and what I hoped to see with Lakefield. This device achieves better battery life per WHr than Lakefield! I wouldn't have minded getting this device at all. This is a great progress and I would even call it Hybrid Unleashed.

I wonder if the problems aren't fundamental but higher level such as system and firmware settings? Otherwise this difference should not be possible. While this laptop benefits having 15W PL2, it shouldn't matter this much. Perhaps the discrepancy lies in the greater than normal binning differences between the 9/15/-P chips?

In general Dell seem to have done something or another on the firmware side, all of their ADL laptops have been much more impressive than their competitors on battery life. Not sure what it is they've done, but they've clearly done _something_ right.

 

[igor_kavinski]

In general Dell seem to have done something or another on the firmware side, all of their ADL laptops have been much more impressive than their competitors on battery life. Not sure what it is they've done, but they've clearly done _something_ right.

If you check their UEFI BIOS, they have some adaptive battery life "secret" sauce.