News Intel GPUs - we've given up on B770, where's Celestial already

Page 261 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
32,416
33,396
146
It was a bot post, most likely.

If UPS can overcome all the adversity delaying my package, and deliver it today. Instead of yet another reschedule, I'll get my B580 up and running with a 12600kf combo. I've been talking smack a long time, about an all Intel desktop gamer. Imma make good on it.
 

eek2121

Diamond Member
Aug 2, 2005
3,478
5,162
136
Intel is dropping their flagship pretty soon, and honestly, they could not have timed it better. Both AMD and NVIDIA are rumored to be cutting DYI production by significant amounts. All Intel literally has to do at this point is to sell these parts at a good target price (with decent margin, sorry budget gamers), make sure they perform, and make sure they are available.

They could leapfrog AMD within the next 2 years and possibly gain 5-10% additional marketshare depending on how they play their "cards".

If they pivot consumer volume to AI, they will wreck their GPU investments forever, so this one will be an interesting one to watch. If I were in charge, I would absolutely silently ignore AI (while pretending to be an AI company) and target the market that AMD/NVIDIA are seemingly handing them on a silver platter (provided that they can execute).
 
Last edited:

WelshBloke

Lifer
Jan 12, 2005
33,528
11,657
136
So, realistically, where is the new Intel GPU going to slot in performance wise with existing gpus?
 

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
32,416
33,396
146
sorry budget gamers
Right on. No big deal either. They already shut most of them out unintentionally, by giving them too many hoops to jump through.

First, some will have to learn to expose and enable rebar on their old setup. Then, they have to be willing to live with how much their old CPU is going to hold ARC back at times. Next, they are going to have to use DXVK, DXwrapper, dgVoodoo2, for some of the old games.

Finally, they are going to have to be cool with some old games not running. And subpar performance that red and green do not experience in more modern titles. UE5 seems to be a big sore spot.
 

dangerman1337

Senior member
Sep 16, 2010
439
77
91
Intel is dropping their flagship pretty soon, and honestly, they could not have timed it better. Both AMD and NVIDIA are rumored to be cutting DYI production by significant amounts. All Intel literally has to do at this point is to sell these parts at a good target price (with decent margin, sorry budget gamers), make sure they perform, and make sure they are available.

They could leapfrog AMD within the next 2 years and possibly gain 5-10% additional marketshare depending on how they play their "cards".

If they pivot consumer volume to AI, they will wreck their GPU investments forever, so this one will be an interesting one to watch. If I were in charge, I would absolutely silently ignore AI (while pretending to be an AI company) and target the market that AMD/NVIDIA are seemingly handing them on a silver platter (provided that they can execute).
Problem with B770 is that at this point ideally can it perform at 4070 Ti levels of performance, able to be sold at a relatively low cost (say $450) without a loss? Only way that can happen if Intel has somehow stockpiled plenty of GDDR6.

Though I think if Intel plays the mid to long game and is working on Xe3P they can take their time for a really good 2H of 2027 launch, likely a year-ish after the AI Bubble bursts (18 year economic cycle and the AI bubble together...) with cards with actually good PPA that compare to RTX 60/Rubin and RDNA 5 on 18A-P where Intel doesn't have to bother with competing against wafer space at TSMC and maybe pair it with 4GB GDDR7 modules. At least have a 128-bit and 256-bit die (hopefully a C770/780 competes against GB203's successor tech wise) meaning 16-GB (or 12GB cut down) or the entry-ish end and 32 or cut down 28GB SKUs for those performance range gamers where Nvidia has left out.

I mean if a C770/780 reach RTX 4090 performance in raster at least and higher in Ray & Path Tracing at 600 or even slightly less they'd have a killer card for the 1440P mainstream. Especially if XeSS evolves with Transformer models + Ray Reconstruction. I think that'd take the market by storm. Something noticeably faster than Xbox Magnus + AT2 (RTX 5080 in raster?) at a slighly lower price point would be really great and be the equivalent of the 3070 (except less memory gimped) was in 2020.
 
  • Like
Reactions: DAPUNISHER

DavidC1

Platinum Member
Dec 29, 2023
2,178
3,328
106
Problem with B770 is that at this point ideally can it perform at 4070 Ti levels of performance, able to be sold at a relatively low cost (say $450) without a loss? Only way that can happen if Intel has somehow stockpiled plenty of GDDR6.
I doubt they are going to lose money on a card basis, either with B770 @ $399 or B580 now. They are probably making a bit per card sold. But volume is what matters, and they have nowhere near the volume to make up for the fixed cost. Even if it's sold at $999 using B580 and doesn't lose volume share it'll lose money because it sells so few of them. It's barely at 1%.
Though I think if Intel plays the mid to long game and is working on Xe3P they can take their time for a really good 2H of 2027 launch, likely a year-ish after the AI Bubble bursts (18 year economic cycle and the AI bubble together...) with cards with actually good PPA that compare to RTX 60/Rubin and RDNA 5 on 18A-P where Intel doesn't have to bother with competing against wafer space at TSMC and maybe pair it with 4GB GDDR7 modules. At least have a 128-bit and 256-bit die (hopefully a C770/780 competes against GB203's successor tech wise) meaning 16-GB (or 12GB cut down) or the entry-ish end and 32 or cut down 28GB SKUs for those performance range gamers where Nvidia has left out.
The next bubble isn't going to be something that most of the population can ignore like the 2001 or 2008 bubble. It's being called Greater Depression for a reason. Most people are going to have the rug pulled completely under them. You won't have any market at that point to sell.
I mean if a C770/780 reach RTX 4090 performance in raster at least and higher in Ray & Path Tracing at 600 or even slightly less they'd have a killer card for the 1440P mainstream. Especially if XeSS evolves with Transformer models + Ray Reconstruction. I think that'd take the market by storm. Something noticeably faster than Xbox Magnus + AT2 (RTX 5080 in raster?) at a slighly lower price point would be really great and be the equivalent of the 3070 (except less memory gimped) was in 2020.
The volume market takes years and years of flawless execution and being on top to be swayed. Look how long AMD took with Ryzen. Look how they are fairing against Nvidia. Second rate performance no matter how good the perf/watt or perf/$ is, it won't happen. I wouldn't buy Intel either unless they show that they are above the short driver support or the disdain towards graphics in general attitude they had for 30 years. They just gave up support on 14th Gen iGPU, when Xe is largely similar to Xe2.
 

regen1

Senior member
Aug 28, 2025
362
455
96

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
32,416
33,396
146
If they really are going to release it to retail, they need to wait. Let the other vendors raise prices first. As I keep seeing comments speculating the profit margin would be razor-thin at the price it needs to sell for to be competitive. Rumor is Nvidia will up the price of the lowest tier 16GB cards first. Which is exactly what ARC needs to happen before launch reviews IMO.
 

dangerman1337

Senior member
Sep 16, 2010
439
77
91
B390 (Xe3) looks really, really good:

Hopefully we see handheld focused SKU that further performance per watt in gaming. B770 probably doesn't exist but if that's still on Xe2 I'd rather if that gets scrapped and Intel focuses on getting a Xe3P dGPU die(s) out.

EDIT: Raichu thinks Xe3P being 20-25% faster per Xe unit would be "suitable" and thinks it could be more 🤔:
For the IGPU part, I think 20-25%, even more perf uplift for the Xe3P-12C(NVL version) is suitable.Inclusive freq uplift and IPC.For the CPU part, I think the same uplift in CPU perf (MT part in peak perf).
 
Last edited:
  • Love
Reactions: DAPUNISHER

DavidC1

Platinum Member
Dec 29, 2023
2,178
3,328
106
Yea, they got a winner here. It's 82% in native rendering, and 50% per watt compared to Lunarlake.
EDIT: Raichu thinks Xe3P being 20-25% faster per Xe unit would be "suitable" and thinks it could be more 🤔:
The dynamic register allocation can improve further according to their overview.

The problem with Xe3P is that it's at least a year away, and that's for iGPUs. How long for their dGPU if ever?
 

vanplayer

Member
May 9, 2024
78
121
66
The problem with Xe3P is that it's at least a year away, and that's for iGPUs. How long for their dGPU if ever?

Would be lucky if you see Xe3P consumer dGPU or unless you are willing to buy Arc Pro. They have to fix so many things like PPA, drivers, etc. And current DRAM shortage is the last straw although it will also affect Nvidia/AMD.
 
Last edited:

mikk

Diamond Member
May 15, 2012
4,333
2,413
136
They specifically mentioned better PPA in their Xe3p docs Bionic said. Celestial Xe3p seems cancelled anyways.
 
  • Like
Reactions: dangerman1337

DavidC1

Platinum Member
Dec 29, 2023
2,178
3,328
106
I always enjoy TAP talks. ARC is improving fast.
I feel like with absence of B770 news if they give up on the dGPU market they will just going back to resting on their laurels. The dGPU market forces the vendors to try, because they are DIRECTLY responsible for their customers, while iGPUs are indirect. The higher performance level also matters in this regard, because with iGPU most people say "meh it's an iGPU" if it happens to have a mediocre experience. I called this years ago as well, saying they'll try way more on the dGPU. Look on the ARC reddit. Most people are in disdain about Pantherlake release, only caring about B770. I get it.

It's nonsense business to have Nvidia on one hand and expect ARC dGPU to continue. Mobile is also higher volume so unless Intel is going to stick Nvidia parts at a very niche, Strix Halo but no lower and no higher that makes no sense either. You either push ARC to be capable and take everything, or push Nvidia at the high end, if not for eventual replacement.

It's just like the fab utilization argument. Intel needs to move products back to their fab as fast as possible. As soon as they moved stuff to TSMC, they are taking a massive loss on the fab side.
 

poke01

Diamond Member
Mar 8, 2022
4,857
6,190
106
I feel like with absence of B770 news if they give up on the dGPU market they will just going back to resting on their laurels. The dGPU market forces the vendors to try, because they are DIRECTLY responsible for their customers, while iGPUs are indirect
I disagree. With handhelds now becoming mainstream, iGPUs are just as important as dGPUs now especially for low power gaming.