No one is making iGPU/GPU for client on 2nm soon except Phone Vendors
Well, as Raven Ridge/Picasso clearly proved, relatively powerful iGPUs don't do much for the ASPs you can ask, so they're mostly just a cost adder and it makes sense to prioritise cost over performance for them.
Out of curiosity, is it mostly a cost/yield issue (not mature/cheap enough for desktop GPUs by 2027) or a volume issue (booked out for NVL, Zen6, HPC GPUs and phone chips)?
But yeah, apart from AMD with Vega20 + RDNA1 on N7 and NV with Ada on "4N" (semi-custom N5 afaik), both GPU IHVs have been rather conservative with their process choices over the past 10 years, so this isn't a new phenomenon.
Cost definitely. Even the N3 Zen 6 products are going to be pricey, can't wait to see what any theoretical N2 ones will cost.
Josh128
Golden Member
- Oct 14, 2022
- 1,318
- 1,983
- 106
lol look at this silliness. The 'kek already picked up on our discussion this morning and has an article about it!! They must have a bot scraping "Kepler_L2" instances in these forums....
wccftech.com
AMD's Next-Gen Ryzen "Zen 6" CPUs To Feature TSMC 2nm "N2P" CCD & 3nm "N3P" IOD
AMD's next-gen Ryzen CPUs based on the Zen 6 core architecture are reportedly going to use TSMC's 2nm & 3nm process for CCD & IOD.
wccftech.com
Raven Ridge/Picasso had the mountain of poor previous products and their accompanying reputation to compete with. They were good products, but, look at what they were selling against. Follow on products dealt with limited availability due to factors including OEM recalcitrance, AMD not wanting to over produce, and improving competing products as Tiger Lake improved a lot over previous Intel iGPUs, making that not as much of a competitive selling point in the market.
The other issue AMD continues to face across much of the market is that those processors are continually placed either in bottom end laptops that don't do them any favors with power/thermals, or, if they are in anything better, they are paired with dGPUs that take the place of the iGPUs. While they did gut the bottom end of Nvidia's mobile stack of cheaper dGPUs, leaving almost nothing on the market below the xx50 series, Nvidia responded by making the xx50 series cheap enough for laptop vendors to sell in the "in between" market that exists above barely above junk level volume retail models and the lower end gaming and professional market devices.
The market never really got to speak because there was barely an opportunity for it TO speak. I can count on one hand the number of laptops that were marketed without a dGPU, but had other useful gaming oriented features, decent power profiles for the CPUs, decent cooling, and weren't priced higher than other laptops that came with xx50 series dGPUs.
The other issue AMD continues to face across much of the market is that those processors are continually placed either in bottom end laptops that don't do them any favors with power/thermals, or, if they are in anything better, they are paired with dGPUs that take the place of the iGPUs. While they did gut the bottom end of Nvidia's mobile stack of cheaper dGPUs, leaving almost nothing on the market below the xx50 series, Nvidia responded by making the xx50 series cheap enough for laptop vendors to sell in the "in between" market that exists above barely above junk level volume retail models and the lower end gaming and professional market devices.
The market never really got to speak because there was barely an opportunity for it TO speak. I can count on one hand the number of laptops that were marketed without a dGPU, but had other useful gaming oriented features, decent power profiles for the CPUs, decent cooling, and weren't priced higher than other laptops that came with xx50 series dGPUs.
It helps that even the garbage stuff is a lot faster than the IGPs. Even Strix Halo likely only made it to market because of AI Hype.
Thunder 57
Diamond Member
- Aug 19, 2007
- 4,025
- 6,740
- 136
I always found that annoying. There would be the low power chips for thin and lights, fine. Then the high power chips that could've been good enough for light/older gaming but everyone just HAD to put a stupid dGPU in them.
Thunder 57
Diamond Member
- Aug 19, 2007
- 4,025
- 6,740
- 136
35W RTX 3050 is a crime in laptop considering how fast 890M/140V are it basically destroys the battery life for not that improved performance.
I've been using them for a while. I got a cast off Raven Ridge laptop that I bios hacked to give higher power to the CPU and transplanted the fans from another laptop from the same line that had a dGPU option to improve the cooling. It's still working great driving a TV in my garage by my workbench. I later got, for a brief time, an ASUS ROG Flow X13 GV301 (2022 edition) that had no dGPU, just the 8 core HS Rembrandt chip. It worked like a champ. The motherboard died within a year though, but I wasn't too upset as I got it used for about a third of what it was sold new for. That device showed that it was possible to build a device that would work well for the concept, but, it was also financed by the far higher end products that also used that platform that had dGPUs in them.
Two of my kids have Lenovo laptops that have Hawk point/Phoenix point chips with 4060 mobile dGPUs. When they are mobile, they both use them with the dGPU disabled with no issues and get plenty of battery life. Such a device is perfectly possible, but no one wants to suffer the tighter margins to make them. Then AMD goes and makes a chip that would have been almost perfect, Strix Point. The ONLY thing it lacked was 16MB of infinity cache to bring it's performance way up, and they put in an NPU instead. After that, they decided that, what, since a broken product didn't sell well, they weren't going to try again?
Two of my kids have Lenovo laptops that have Hawk point/Phoenix point chips with 4060 mobile dGPUs. When they are mobile, they both use them with the dGPU disabled with no issues and get plenty of battery life. Such a device is perfectly possible, but no one wants to suffer the tighter margins to make them. Then AMD goes and makes a chip that would have been almost perfect, Strix Point. The ONLY thing it lacked was 16MB of infinity cache to bring it's performance way up, and they put in an NPU instead. After that, they decided that, what, since a broken product didn't sell well, they weren't going to try again?
Saylick
Diamond Member
- Sep 10, 2012
- 4,035
- 9,454
- 136
SMDH. All wtftech will ever be is a glorified news aggregator...lol look at this silliness. The 'kek already picked up on our discussion this morning and has an article about it!! They must have a bot scraping "Kepler_L2" instances in these forums....
View attachment 129555
AMD's Next-Gen Ryzen "Zen 6" CPUs To Feature TSMC 2nm "N2P" CCD & 3nm "N3P" IOD
AMD's next-gen Ryzen CPUs based on the Zen 6 core architecture are reportedly going to use TSMC's 2nm & 3nm process for CCD & IOD.
Thunder 57
Diamond Member
- Aug 19, 2007
- 4,025
- 6,740
- 136
adroc_thurston
Diamond Member
- Jul 2, 2023
- 7,070
- 9,815
- 106
moooooooooooooom moooom mom get the camera i'm in the TVlol look at this silliness. The 'kek already picked up on our discussion this morning and has an article about it!! They must have a bot scraping "Kepler_L2" instances in these forums....
View attachment 129555
AMD's Next-Gen Ryzen "Zen 6" CPUs To Feature TSMC 2nm "N2P" CCD & 3nm "N3P" IOD
AMD's next-gen Ryzen CPUs based on the Zen 6 core architecture are reportedly going to use TSMC's 2nm & 3nm process for CCD & IOD.
adroc_thurston
Diamond Member
- Jul 2, 2023
- 7,070
- 9,815
- 106
10-15% speed ain't gonna make or break them, and N3p slots will be a lot more free.
It's a normal tradeoff.
Open a Twitter account first to become a renowned leaker xDd
adroc_thurston
Diamond Member
- Jul 2, 2023
- 7,070
- 9,815
- 106
Kolifloro
Member
- Mar 15, 2023
- 31
- 27
- 61
iGPU on N2P would dissipate less heat ... which is pretty good for a laptop ...
... it would also occupy less space ...
Anyway ...
...
adroc_thurston
Diamond Member
- Jul 2, 2023
- 7,070
- 9,815
- 106
Again, 10% speed is not a dealbreaker for gfx.
At a tiny bit more money per wafer and generally more limited capacity.
MLID made a reference some time ago that Zen 6c Venice base clock is in 4 GHz range and he made it sound like that was the base clock. Which is similar or higher than boost clock high core count Zen 5.
Makes sense, and there is no problem if desktop CPU has some RDNA3.5 graphics.
What I am wondering is if there are any notebook Zen 6 parts that might launch in 2026, with older generation GPU. There doesn't seem to be any info about that.
It's part of the IOD / SoC in most instances, so it would be very wasteful to have anything better than N3P.
If you look at standalone dGPUs, they are not moving to N2 either (including NVidia). The only exception is Mi400 datacenter GPU, and that one is chiplet based, so only the compute portion will be N2.
Nvidia responded exactly the same way they did back with Iris Pro - cutting the price of the low end parts so there's no cost advantage.
You get iGPUs for three reasons:
-Cost
-Battery life
-Form factor
The regular iGPUs achieve all three marvelously. The Halo iGPUs increase cost, and the smaller form factor generally further increases cost. In addition, it's pretty much useless in desktops because the price factor gets further compounded by having it all in a single chip.
Part of the fault is in the vendor that makes them. If Intel had Iris Pro priced at i3 + Iris Pro, then it would have been more competitive. Same with AMD. Why not offer Ryzen R3 level with the Strix Halo? Both vendors make Halo iGPUs with a fundamentally different reason than regular iGPUs - make more money. They pair halo iGPUs with halo CPUs to force upsell. Also, modern iGPUs are in reality not really "free" either. Because back in the chipset graphics days, you could pair the lowest CPU with the best chipset graphics. Now you have to spend $100 or more for the highest. Are you *really* saving money? It's all a clever marketing trick. Heck, if you really cared about consumer value, you'd offer Strix Halo/Iris Pro graphics with a Celeron CPU and price it accordingly.
Which Nvidia can attack easily since they are an established vendor.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 22K
-
News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products- Started by poke01
- Replies: 384
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
Facebook
Twitter