News Intel GPUs - Falcon Shores cancelled

Page 235 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

ajsdkflsdjfio

Member
Nov 20, 2024
171
117
76
Dunno why you care so much. He could be right. You could be right. Only way to find out is to wait and see. I don't understand your bitterness over his refusal to agree with you.
He could absolutely be right, the issue is he makes these assumptions through reasoning so poor that they could be considered lies. Look at the original post he made which started this entire thing:
It's even worse than it looks based on size - N4 wafers that Intel uses are priced at $20k whereas N7/6 are $9500 (Tom's Hardware estimates), plus yields will be different.
Intel doesn't use N4 they use N5. N4 wafers aren't 20k based on toms hardware they are 18k$ according to his own article he later linked. N7 prices aren't 9,500 according to Tom's hardware they are 10k, and you cannot use N7 prices for N6. Through multiple days of arguing he still holds that N5 wafers are undoubtedly at least 18K and will be closer to 20k in 2025 and refuses to acknowledge even the faintest possibility that BMG isn't actively losing Intel money for every card they produce. He also continuously refuses to acknowledge that N5 and N4 are separate nodes that are priced separately by TSMC. He uses yields here as if they are likely to be a significant cost increase, but based on what? Aren't yields for these more mature nodes relatively good, and even the larger BMG die isn't at such a large size that its yields will be significantly worse than the rx 7600 especially when you add in the fact that the B580 is less dense of a design than typical N5 devices.

He later tried to claim that Intel's b580 die is up to 3.5x more expensive to produce than an rx 7600 die like what????

It's not about him agreeing/not agree with me, I'd just rather not see more overtly BS posts like these in these threads. Maybe it's an issue of me caring too much, but personally I joined this forum in hopes of actual good-faith discussions and learning more about semiconductors and future product designs, not people parroting the latest anti-XYZ company opinions.
 

Win2012R2

Senior member
Dec 5, 2024
647
609
96
so let's make another bet on top of our celestial bet
No thank you, one silly wager is enough, plus if Intel cancels BMG next year that would almost certainly mean no Celestial either, in which case you'll have to self delete from here, so no more any talk anyway.
 

Attachments

  • img_6690[1].png
    377.9 KB · Views: 13
Reactions: ajsdkflsdjfio

reaperrr3

Member
May 31, 2024
55
188
66
If Intel effectively or outright cancels BMG in the first 9 months of 2025 (no more sales)
MLID's (who might deserve some of the flak he's getting, but still has a better track record than many other rumor mills, including some exclusives that turned out to be correct, like the Zen5 IPC) alleged sources at Intel are saying they'll only produce the absolute minimum they have to to keep up appearances (because they're really losing money per card sold), but will trickle that low volume out over time instead of selling it all in one go.
So assuming that info might have some substance to it, nobody should risk taking that bet, even if they thought you're otherwise wrong...
 

ajsdkflsdjfio

Member
Nov 20, 2024
171
117
76
MLID's (who might deserve some of the flak he's getting, but still has a better track record than many other rumor mills, including some exclusives that turned out to be correct, like the Zen5 IPC) alleged sources at Intel are saying they'll only produce the absolute minimum they have to to keep up appearances (because they're really losing money per card sold), but will trickle that low volume out over time instead of selling it all in one go.
So assuming that info might have some substance to it, nobody should risk taking that bet, even if they thought you're otherwise wrong...
I really wouldn't put much stock into MLID info, I mean he's the same guy that claimed arrow lake would bring +40% higher single core performance and 25% ipc for lion cove, not even joking this is word for word. Also all this flak he's getting has been 10x worse for battlemage to the point where he's had to go to the comments and defend himself in a lot of his videos covering it. I wouldn't really trust him to be unbiased when it comes to battlemage even if he actually had connections with reputable sources. Also he's literally said this same no volume given to retailers thing for multiple cards from Nvidia and AMD in the past and even if he was right about those cards at launch day, they ended up being shelves and have continued to be produced for the majority of their product cycle. Watching his video, I think I'm starting to understand where so many people here get the idea that Intel is selling its cards at a loss. Seems like we got some MLID patreon subs here.

Even if he's right about battlemage, trickling out volume a little at a time would essentially mean having arc b580 out of stock for 80-90% of the time even well after the usual initial launch hype. It will be pretty obvious if Intel actually goes through with this game plan. I'd consider that to mean battlemage is essentially canceled anyways and would consider that to fall under the category of "effectively" canceled which I put in that statement specifically for situations like this.

This is really a side point, but if you watch Tom Peterson's interview, specifically from 34:20 onwards, you really get the idea that Intel is trying to gain a foothold, even if its small, in the market with battlemage and its pricing. To me that means two things:
1. Battlemage has a relatively decent amount of volume, which I define as significantly more than alchemist and enough so that If you want battlemage you will be able to buy it at decent prices (a little while after initial launch of course)
2. the B580 at 250$ is likely not actively losing intel for each card it sells. If intel is trying to gain a foothold in the market meaning significant supply, it is unlikely their card is losing them money for each one they produce as that would be unsustainable.

 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
22,369
12,175
136
And what's your Googling coming up with, have you got some alternative sources contradicting what I have offered?

So far it appears you've got squat whereas I've provided the best currently freely available.

You didn't actually present any evidence from your "Googling" that I'm aware of. But hey, whatever.

So back to the Aussie UK Steve review: why is he still using a 14900k @ 125W? Is this the CPU he thinks is appropriate the a B580? Let's be honest, we should expect to see this card paired with cheaper CPUs quite often, so maybe it's a fair pairing (or a 5800X3D or 7800X3D or whatever). But for a raw review of the card, pair it with a 9800X3D and let er rip!

Also if B580 is hard on the CPU then maybe it's got some budget caveats people need to think about before buying it
 
Last edited:
Reactions: DavidC1

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
30,379
26,927
146
It's UK Steve aka RandomGaminginHD not HUB Steve. He just did the 1440 tests and had less issues in town in RDR2. He is adamant that it's not a CPU issue because he tested other CPUs with it. 🤣 I don't know which ones he tested, but everything points to the CPU being the problem. I suppose it could be a driver bug, but playing in higher resolution making it run better is a big checkmark.

And if it is anything like Alchemist, CPU overhead is higher than even Nvidia.
 

DavidC1

Golden Member
Dec 29, 2023
1,362
2,222
96
And if it is anything like Alchemist, CPU overhead is higher than even Nvidia.
It definitely has high overhead. People with Zen 1/2 systems say it performs quite a bit lower in the Intel ARC reddit. One guy upgraded it to Zen 3 X3D and it improved a lot.

It's clear from 1080p tests and relative standing against A770 if anything the overhead is worse.

This is why I keep wondering if they can pull out a driver that puts out 10% general performance improvement in 1080p. The card definitely improved a lot on both perf/watt and perf/mm2 ISO-node but some expectations didn't pan out like UE5 performance and 1080p.

If the driver overhead is high with B580, then B770 is going to become critical.
 
Reactions: ajsdkflsdjfio
Mar 11, 2004
23,410
5,823
146
It definitely has high overhead. People with Zen 1/2 systems say it performs quite a bit lower in the Intel ARC reddit. One guy upgraded it to Zen 3 X3D and it improved a lot.

It's clear from 1080p tests and relative standing against A770 if anything the overhead is worse.

This is why I keep wondering if they can pull out a driver that puts out 10% general performance improvement in 1080p. The card definitely improved a lot on both perf/watt and perf/mm2 ISO-node but some expectations didn't pan out like UE5 performance and 1080p.

If the driver overhead is high with B580, then B770 is going to become critical.

Wondering if this might be something related to resizable BAR, I think Intel's review guide said to use it (think I saw somewhere that Intel said it was required even?).
 

mikk

Diamond Member
May 15, 2012
4,286
2,366
136
In this test there are 3 resolutions up to 4k: https://www.profesionalreview.com/2024/12/12/intel-arc-b580-review/

Big difference from 1080p over 1440p over 4k. So yes the utilization/overhead issue remains the same if not worse on Battlemage. I wonder if all overhead optimizations for Alchemist are enabled on Battlemage.

However the performance and load efficiency is a surprise to me, I did not expect 20 Xe2 cores that fast and that efficient at 2850 Mhz. And this is on N5 instead of the rumored N4P, a pity they didn't use N4P.

On my B580 and I have to run Furmark 2 to reach 190W board power and in this case the chip power goes to 140W. There is hardly a game which goes to 140W.

In this test of 16 games there is only 1 game over 130W chip power.

 

BurnItDwn

Lifer
Oct 10, 1999
26,233
1,717
126
Yay! Level1Linux post on B580!

Looks like it handles most things decently well on the open souce drivers. Some games like Cyuberpunk have problems, but, overall appears to be a strong option for Linux.
 
Last edited:
Reactions: igor_kavinski

gaav87

Senior member
Apr 27, 2024
452
794
96
Who at Intel was the mastermind behind this genious move to cancel the 24- core 12GB version ?
It would look so much better even at 329$ maybe streach it to 349$. They would deff profit on this vs b580 net-zero

 

DavidC1

Golden Member
Dec 29, 2023
1,362
2,222
96
Who at Intel was the mastermind behind this genious move to cancel the 24- core 12GB version ?
It would look so much better even at 329$ maybe streach it to 349$. They would deff profit on this vs b580 net-zero
You aren't going to get 20%, except at unplayable 4K settings. 10-12% is more realistic and you'd be right at 4060 Ti 8GB. A770 is not 33% ahead of A580 for the same reason.

And we don't know how much money they are losing/making per unit. Only that guy that seems to want to derail the thread conclusively believes this.
 
Last edited:
Jul 27, 2020
22,304
15,557
146
It's kinda weird the architecture design choice they have gone with. More pixels should strain their GPUs, not make them run faster. Seems the occupancy and scheduling stuff in their architecture is fundamentally broken or just more geared towards compute, which makes sense since Raja seems to be a compute nut.

I wonder if they can undo Raja's damage in Celestial or if it will take one more gen to de-Raja their GPU.
 
Reactions: Win2012R2

Hans Gruber

Platinum Member
Dec 23, 2006
2,448
1,316
136
They said the B580 was supposed to be the replacement for the A750. Will there be a B750 and a B770 coming soon? I was thinking the B580 is the Battlemage version of the A580.
 

GTracing

Senior member
Aug 6, 2021
275
645
106
It's kinda weird the architecture design choice they have gone with. More pixels should strain their GPUs, not make them run faster. Seems the occupancy and scheduling stuff in their architecture is fundamentally broken or just more geared towards compute, which makes sense since Raja seems to be a compute nut.

I wonder if they can undo Raja's damage in Celestial or if it will take one more gen to de-Raja their GPU.
It's not that higher resolutions make their GPUs faster. They seem to have more driver overhead. Which means at they run into CPU bottlenecks easier than Nvidia or Radeon GPUs. Lower resolutions are more CPU bottlenecked than higher resolutions are.

HWU has covered how Nvidia driver overhead is worse than AMD. It affects low end CPUs more than high end CPUs. It isn't really an issue because most people don't pair an i3 with a 3080 and a 1080 by 1920 monitor.

If the issue with Battlemage is indeed driver overhead, and if it's worse on weaker CPUs, that could put a damper on it's budget-king status.
 
Reactions: Tlh97 and Elfear

DavidC1

Golden Member
Dec 29, 2023
1,362
2,222
96
It doesn't make it run faster... It is only relative. The competition has less memory bandwidth with fewer TMU and ROPs so they fall off more rapidly.
No, it's a driver-related bottleneck, which shows as being "CPU limited". In higher resolutions it doesn't matter as much.
It's not that higher resolutions make their GPUs faster. They seem to have more driver overhead. Which means at they run into CPU bottlenecks easier than Nvidia or Radeon GPUs. Lower resolutions are more CPU bottlenecked than higher resolutions are.
Nvidia moved to driver based scheduling few generations ago and Intel did the same with the Iris Xe generation. Plus, Nvidia is much better at drivers.

Intel has a problem of needing to sell it super cheap with a more expensive BoM, while it's not a nice fit for budget systems because it needs ReBar and extra performance a higher end CPU needs.

Nvidia at least has a good reputation so they can sell it for the premium systems.
 

gaav87

Senior member
Apr 27, 2024
452
794
96
You aren't going to get 20%, except at unplayable 4K settings. 10-12% is more realistic and you'd be right at 4060 Ti 8GB. A770 is not 33% ahead of A580 for the same reason.

And we don't know how much money they are losing/making per unit. Only that guy that seems to want to derail the thread conclusively believes this.
My estimate was 12%-20% 65-69 fps over b580. With only 4 more cores same phys/bus/ram/cooler/board i doubt it would be much more then 20-30$ over b580.
If they priced it at 349$. It would be 100-150$ cheaper than 4060ti 16GB with same performance. And 50$ cheaper than 7700xt with slightly worse raster and way way better RT performance.
Also they would even make small amount of profit on it at 349$.

After brain-storming for, a while. G31 (32cores) has to have, a cut-down version for deffective dies anyway so maybe, a 28core (g31) ? So then G10 (28-core and 24-core) cancel was correct.
 

DigDog

Lifer
Jun 3, 2011
14,066
2,554
126
update: there is no update.

new mobo, new TIM, PC still dead. Lights up, ethernet flashes, no output video. Not from GPU, not from CPU. Did the whole "one ram stick at the time , in one different ram slot at the time" as well.
Obviously no CMOS to reset as no BIOS settings yet.
Monitor works but not getting any video.
New GPU has the "Arc" logo that lights up, fans spin, so PSU is fine.

Not really possible that i damaged the CPU, as i didn't even unmount it from mobo previously, when it had already died. Have used various anti-static tricks to be extra extra careful.

f* it, tomorrow i will be going, for the first time in 30 years, to the computer repair shop.
 
Reactions: igor_kavinski
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |