Next month something Battlemage dGPU related might come.
Raytracing improvements didn't result in lower loss of performance with Raytracing on, because a faster hardware needs faster Raytracing just to keep the same losses.That's roughly a doubling over A770 with the same amount or units. That's huge if true. I would expect that the advantage in dx12 titles will be much bigger because Alchemist struggled due to several hardware limitations, also Raytracing has been upgraded on Battlemage.
There is no 64Xe core variant highest is 32Xe cores same as alchemist but the clocks and performance per clock are higherRaytracing improvements didn't result in lower loss of performance with Raytracing on, because a faster hardware needs faster Raytracing just to keep the same losses.
Since Xe2 is ~50% faster per clock, if they can manage 33% clock improvement they can do it. Probably needs 300W to do so. If the earlier expectations of 64 Xe cores and massive cache was true, they might have not needed it to be clocked that high.
Ray tracing hw has been 1.5X - 2XA 50% faster per clock in rasterization matches with 50% increase in raytracing units.
There were many early leaks about 64 Xe cores and a massive cache over 100MB. There were even earlier leaks about a version with 128 Xe cores but 4x32 Xe core tiles.There is no 64Xe core variant highest is 32Xe cores same as alchemist but the clocks and performance per clock are higher
Xe2 has exactly 50% more Ray Tracing units with 3 traversal pipelines versus 2 for Xe. If the rasterization hardware performs 50% better than it'll end up no worse or no better than the predecessor. Ray Tracing takes a fixed amount of time in the graphics pipeline.Ray tracing hw has been 1.5X - 2X
Isn't there twice the number of ALUs per Xe core in Battlemage?There is no 64Xe core variant highest is 32Xe cores same as alchemist but the clocks and performance per clock are higher
No it gets twice the amount of ALUs per EUs, hence why they abandoned that terminology. So a 32 Xe core Battlemage has half the amount of EUs, but same ALUs as 32 Xe core Alchemist.Isn't there twice the number of ALUs per Xe core in Battlemage?
Isn't there twice the number of ALUs per Xe core in Battlemage?
Hope it is true they need a good launch in GPU LNL was good showing for Xe2Intel is almost there with GPU's. They said ARC was a flawed/broken architecture structurally. They did their best to fix what they could with software drivers and patches. They said Battlemage will fix the hardware problems. Assuming they are on N4P, efficiency will see a huge jump over N6 which was what alchemist was built on.
The vast majority of gamers are in the xx60 and xx70 series Nvidia cards. AMD could have really cleaned up with a 12-16GB version of the 6660XT. I do not see why they do not update silicon mid cycle on their cards. At least Nvidia uses premium silicon throughout their product line all the way down to the 4060. The efficiency the 4060 has which is amazing will be almost identical silicon to N4P that Battlemage should be based on. The 50 series cards will see no efficiency gains because they are still on 5nm silicon process. Nvidia was already using the best silicon with the 40 series cards. 4nm is still made on the 5nm node, it just signifies very advanced silicon for performance and efficiency vs. the original standard N5 silicon.
Intel's competition should be the 5060 and 5060ti cards. Any efficiency gains on Blackwell would have to be in the GPU design. Obviously Blackwell should be very good but power efficiency will be flat because of their silicon. This upcoming GPU generation will be the first with silicon parity if both Intel and AMD use N4P.
They already have Celestial on their Intel 3 Process in works along with 18A and N3EBecause Intel makes their own silicon. When 18A is ready, they would have a cost advantage over both AMD and Nvidia using their own silicon. Intel's integrated GPU's have always sucked. Their discrete GPU development has greatly improved their onboard graphics.
Why would 50 series see no efficiency gains? Yes it might be using the same process but they can update the architecture.The 50 series cards will see no efficiency gains because they are still on 5nm silicon process. Nvidia was already using the best silicon with the 40 series cards. 4nm is still made on the 5nm node, it just signifies very advanced silicon for performance and efficiency vs. the original standard N5 silicon.
Again it's a fallacy. uarch gains are even more important now because process gains are starting to flatline. Nvidia has excellent engineering team. I bet they are going to keep the efficiency lead and 50 series will be a decent advancement over 40.Obviously Blackwell should be very good but power efficiency will be flat because of their silicon.
For Battlemage? Well they are starting from a lower point so have more room to improve too. We know going from N5 to N3B along with the uarch improvements results in over 1.5x efficiency gains, sometimes close to 2x.it won't be a massive jump like from samsung 8Nm -> N4 though
Nvidia jumped from Samsung silicon to TSMC because their silicon was superior to Samsung. AMD had the upper hand on Intel for years because of TSMC 7nm and 5nm silicon. Nvidia has superior processor designs, process did not matter as much. Nvidia does not need the performance uplift from silicon. They benefited from the efficiency process of 4N with the 40 series. The 50 series should have been on 3nm but Apple bought up all the silicon. Nvidia is using 4NX which is the most dense silicon on the 5nm node. The efficiency gains are 1-3% at best.Why would 50 series see no efficiency gains? Yes it might be using the same process but they can update the architecture.
If process was the only determining factor, then Arrowlake should have had better efficiency compared to Zen 5. This is not the case. Then the primary focus is getting the latest process, forget everything else.
Again it's a fallacy. uarch gains are even more important now because process gains are starting to flatline. Nvidia has excellent engineering team. I bet they are going to keep the efficiency lead and 50 series will be a decent advancement over 40.
I meant fof Blackwell it was 8nm samsung for 3000 series yes 2X gains are on the tableFor Battlemage? Well they are starting from a lower point so have more room to improve too. We know going from N5 to N3B along with the uarch improvements results in over 1.5x efficiency gains, sometimes close to 2x.
If what Golden Pig says translates into 4070S level of performance with all the expected improvements, they might do much better than Alchemist.
Alchemist's requirement of ReBar makes it a dichotomy of "cheap" card needing bleeding edge systems to work. Same with the high idle power.
Plans always change and if Battlemage surprises, they might decide to continue on Celestial. That's what the rumors might be really indicating - a wait and see approach to see how Battlemage fares.
OEM wanting to use cheap tricks to save cost color me suprisedIt isn’t Intel’s fault OEMs don’t support the products they release. MOST UEFI systems support rebar. See this if your doesn’t: https://github.com/xCuri0/ReBarUEFI
EDIT: Not PCIE version dependent either. Supported on Sandy Bridge, so your old 2500k-2600k can use it.
It's an inexcusable blunder.Lack of OEM support does not change the reality for Intel: rebar requirement handicapped their GPUs when it came to value/budget shoppers.
It can be a PITA to get working depending on the vendor and system in question. Oz did it with a lenovo and it got rather involved. Video is time stamped -It isn’t Intel’s fault OEMs don’t support the products they release. MOST UEFI systems support rebar. See this if your doesn’t: https://github.com/xCuri0/ReBarUEFI
EDIT: Not PCIE version dependent either. Supported on Sandy Bridge, so your old 2500k-2600k can use it.