Celestial was leaked as being on N3 a while ago, plus Intel is on a two year cycle so it's a 2026-2027 product.
Nvidia will continue to have noticeable perf/watt lead in Battlemage generation. Process is only one factor.
Celestial was leaked as being on N3 a while ago, plus Intel is on a two year cycle so it's a 2026-2027 product.Intel would have a process advantage on cost alone with Celestial if they make their GPU's on 18A. Contrary to what people hype up in the GPU segment. The 4060/5060 and 4070/4070 is the real sweet spot of GPU's. Specially the Nvidia 60 series level of performance is where the volume sales are. Consumers are more likely to upgrade often or more often if they can get a GPU for $200-250.
true but intel want to be a foundry and if they want Nvidia, AMD or QQ as customers they better be able to deliver.That's a lot of expense to create something that won't be able to generate revenue in excess of Cost of Goods Sold. Maybe they could deliver a low level product with a tiny die, but pricing power there is VERY weak.
Yes the foundry needs to deliver and from the looks of it Intel Foundry is in better shape technologically and competitively vs their core designtrue but intel want to be a foundry and if they want Nvidia, AMD or QQ as customers they better be able to deliver.
I have heard that B580 Is 4060 level and with good optimization 4060Ti level but for me good improvement would be low idle power improved support vs ARC Alchemist 40-50% faster than A580 12 GB Vram and decent price of 240$B580 vs. A580 perf estimates anyone?
Mine:
2x or more in synthetics
1.65x or more in CP2077
B580 - 20 Xe2, 192-bit 18 or 20 GbpsB580 vs. A580 perf estimates anyone?
Mine:
2x or more in synthetics
1.65x or more in CP2077
That's a great figure for CPU but for GPU, it seems pretty meager. I hope you are dead wrongBased on regression in core count and memory bandwidth, compensated by improvements in Xe2 architecture, my guess is +30% perf.
B580 - 20 Xe2, 192-bit 18 or 20 Gbps
A580 - 24 Xe1, 256-bit 16 Gbps
Based on regression in core count and memory bandwidth, compensated by improvements in Xe2 architecture, my guess is +30% perf.
You have it backwards. A580 is the old one.bleh, 256 bit x 16 gbps? gross. That's barely an uplift over 192 bit 20 gbps. Unless there is other hardware tied to the memory controllers, why bother? That's 7% more or so?
Derp, my bad.You have it backwards. A580 is the old one.
And BM is more efficient in memory access so it'll be fine.
Not if it clocks higher like rumored.B580 - 20 Xe2, 192-bit 18 or 20 Gbps
A580 - 24 Xe1, 256-bit 16 Gbps
Based on regression in core count and memory bandwidth, compensated by improvements in Xe2 architecture, my guess is +30% perf.
Wait a sec isn't 8 Xe core are a part of single slice so it should be 24Xe2 physically they Just disabled 4 or is it really 20B580 - 20 Xe2, 192-bit 18 or 20 Gbps
A580 - 24 Xe1, 256-bit 16 Gbps
Based on regression in core count and memory bandwidth, compensated by improvements in Xe2 architecture, my guess is +30% perf.
Dunno but there's a rumor that the top part is called B980. Probably 32GB VRAM and may be as fast as a 4080.What are the specs for the top end BMG?
32 Xe2, 256-bit 20 GbpsWhat are the specs for the top end BMG?
Please tell me the source is not MLID cause it is not possible with 32Xe CoresDunno but there's a rumor that the top part is called B980. Probably 32GB VRAM and may be as fast as a 4080.
It was meant to be in between 4070-4070S anyway if they achieve anywhere near that it is good 4080 is ridiculous with their situationI didn't remember correctly.
View attachment 111891
So maybe 40% slower than 4080 (still a rumor though).
At 2740MHz it's clocked over 35% faster than peak A580 clocks. 20 Xe core version would then be like a 27-28 Xe core version.Hello Battlemage https://www.amazon.com/dp/B0DNV4NWF7?ref_=ast_sto_dp
That site is IMO even more of a joke than MLID. He basically rehashes information from MLID and RGT.It was meant to be in between 4070-4070S anyway if they achieve anywhere near that it is good 4080 is ridiculous with their situation
Yeah it should be between 4060-4060ti and if its 239$ it is a nice productAt 2740MHz it's clocked over 35% faster than peak A580 clocks. 20 Xe core version would then be like a 27-28 Xe core version.
In Lunarlake, it's clocked slightly lower and performs 30-50% faster than the predecessor. I would not be surprised at a solid 50% gain. That's 4060 Ti using TPU benchmark.
That site is IMO even more of a joke than MLID. He basically rehashes information from MLID and RGT.