AleksandarKIn regular SKUs, VCCGT is a single-phase power rail for the integrated graphics inside the processor. However, since Intel aims to enhance its iGPU performance with "Nova Lake," a single phase is insufficient.

Will we have motherboard makers trying to beat one another in VCCGT phases quantity?

wNotyarDWill we have motherboard makers trying to beat one another in VCCGT phases quantity?

I can imagine so :)

AleksandarKThis SKU includes four P-Cores, eight E-Cores, and four LPE-Cores, along with 12 Xe3P GPU cores. This 4+8+8+12 core

With words must be 4+8+4...not sure that Arc EUs must be counting together with CPU cores.

And there goes Intel’s efficient integrated Xe GPUs

Assuming that Intel doesn't charge too much for this SKU it's a relatively exciting prospect. Of course it would also require affordable motherboards with requisite VRM. It shouldn't have the PCIe/cache disadvantage that AMD's Zen APUs have had. Allowing up to 65W for graphics has the potential to put the performance somewhere in the desktop RTX 3050 range. Since entry level discrete graphics are dead this could make for a solid starting point or low power gaming system.

Let3rAnd there goes Intel’s efficient integrated Xe GPUs

They are paving the way for the Nvidia GeForce integration into Intel CPU; we all know they'll need all the power...

Pretty toasty power consumption! Still, 65W does not seem that bad considering it will likely deliver decent performance. I wonder if the iGPU is also built on 18A like the computer tiles.

Let3rAnd there goes Intel’s efficient integrated Xe GPUs

Yet another user who doesn't understand the definition of efficiency.

TheinsanegamerNYet another user who doesn't understand the definition of efficiency.

What do you mean by that!

thestryker6Assuming that Intel doesn't charge too much for this SKU it's a relatively exciting prospect. Of course it would also require affordable motherboards with requisite VRM. It shouldn't have the PCIe/cache disadvantage that AMD's Zen APUs have had. Allowing up to 65W for graphics has the potential to put the performance somewhere in the desktop RTX 3050 range. Since entry level discrete graphics are dead this could make for a solid starting point or low power gaming system.

I think you're right on there, in fact you can get RTX 3050 6gb as a low profile/half height SFF card. Either single slot or dual slot, and they are around that 60-65w limit since they are slot powered with no auxiliary power. Intel also has an A380 8gb in that same form factors and power limit. They are all going to do 1080p medium and hit that 60-90fps range in most modern games without having to enable frame gen to do it. So absolutely going to be a solid gaming machine if its any improvement on those. With some combination of xess and framegen it'll probably bounce off the refresh rate limit of a lot of budget gaming panels. Panther lake's B390 IGP is already impressive in handhelds, and that gpu gets about a third of that power.

Let3rWhat do you mean by that!

It can draw *up to* 65W, that doesn't mean it *will* draw 65W.

Efficiency is the performance to power ratio. It can be measured in frames per watt.

A more efficient GPU is capable of better performance at a given power draw.

Example, the RX 9070 XT defaults to 304W, with a maximum (on most cards) of *up to* 375W, and yet you can get about 90% of the performance at 250W (~80% power) with an undervolt.

Example 2: Despite being a much less powerful GPU, the B390 can actually outperform the 8060S at low power limits (15-25W) due to its better efficiency.

Let3rWhat do you mean by that!

Power use =! Efficiency.

Efficiency is a measure of work done compared to power drawn. You cannot judge this based on the power number alone. A 65 watt GPU that is twice as fast as a 40 watt GPU is more efficient, because it would take the 40 watt GPU 80 watts total to match the output of the 65 watt part.

Without performance metrics, a power rating is meaningless. You need perf/W, whether that be in frames per watt, or pixels rendered, or some kind of workload.

Let3rAnd there goes Intel’s efficient integrated Xe GPUs

functionally irellevant for these niche gamer desktop SKUs.

The fact that it doesn't "waste" with too high end of the CPU potentially makes this an actually great Desktop iGPU.

Previous big iGPU attempts were irrelevant as they always paired it with the i7's, the highest of the lineup. So if you wanted the iGPU, you were paying for the extra CPU, which is useless to you. I always wanted them to have a i3 lineup + decent GPU. So if the i3 itself costs $150, then they can charge $250, or extra $100 for the bigger iGPU. Which many people will see it as reasonable. I think even $150 extra is good. I can't see even at $100 how this is not attractive to Intel either. It's an i3, which they will sell anyway, but get to charge extra for graphics.

thestryker6Assuming that Intel doesn't charge too much for this SKU it's a relatively exciting prospect. Of course it would also require affordable motherboards with requisite VRM. It shouldn't have the PCIe/cache disadvantage that AMD's Zen APUs have had. Allowing up to 65W for graphics has the potential to put the performance somewhere in the desktop RTX 3050 range. Since entry level discrete graphics are dead this could make for a solid starting point or low power gaming system.

Xe3P should be at least 25% faster per Xe core according to rumors. I think it should be close to 50% faster as RDNA5 competition coming just few months later will be much faster than Pantherlake's GPU, and Xe4 is much further away.

Regardless, in a desktop format, with improved Xe3P, and uncorked in regards to power limit, this is potentially an attractive part for value gamers.

Let3rAnd there goes Intel’s efficient integrated Xe GPUs

It IS efficient. You are confusing low power use with efficiency. Efficiency is a relative metric, with performance being one part of the equation. A 486 CPU is very low power, but it isn't "efficient". A 250W 285K is more "efficient" than the 4W 486, because it's lot faster than the power difference.

Good chance of GPU being bottlenecked by DRAM more than actual capabilities of GPU itself. APU promised on this front but most failed due to DRAM bottleneck when compared to dedicated GPUs. Anyways good to see another iGPU(hopefully one that wont have asking price of Strix Halo) that actually can kill entry level dGPUs.

DavidC1I think it should be close to 50% faster as RDNA5 competition coming just few months later will be much faster than Pantherlake's GPU, and Xe4 is much further away.

I wouldn't make any bets on when RDNA 5 launches, its relative performance, when it will appear in APU form or what it will look like in said form. The Medusa Point APUs are still using a RDNA 3.5 based architecture and Medusa Halo is the only known RDNA 5 APU which absolutely will not appear in socketed form. Unless AMD flips the script on how they design APUs Intel will be playing by itself with regards to integrated graphics.

Rumors on the Intel side point towards a Halo competitor with RZL, so late 2027-mid 2028, which would be using Xe3p. This would indicate Intel believes it can compare favorably enough with RDNA 5 or else it would have been canceled.

ChaitanyaGood chance of GPU being bottlenecked by DRAM more than actual capabilities of GPU itself. APU promised on this front but most failed due to DRAM bottleneck when compared to dedicated GPUs.

That's why PTL's Xe3 implementation has 16MB cache. It's also why Strix/Gorgon Halo have Infinity Cache unlike their regular APU counterparts despite having twice the memory bandwidth.

ChaitanyaGood chance of GPU being bottlenecked by DRAM more than actual capabilities of GPU itself. APU promised on this front but most failed due to DRAM bottleneck when compared to dedicated GPUs. Anyways good to see another iGPU(hopefully one that wont have asking price of Strix Halo) that actually can kill entry level dGPUs.

Yes I easy imagine another $1000 for fast memory just so the "cheap" iGPU can show its potential.

XeonsNOpteronsI wonder if the iGPU is also built on 18A like the computer tiles.

Considering the compute is the 4+8 tile, seems like the compute tile will be on N2.
Honestly I expect the "large" variant of the iGPU tile to be external more than I do the 4+8 tile though. Even if Intel can gain PPA by moving the 4+8 compute tile to N2P, it's so much volume that IFS is going to lose by not making it on 18A. Unlike the large iGPU tile which will be comparatively more niche. Plus, I expect the iGPU to gain more from the move to N2P than the compute tile might.

Let3rWhat do you mean by that!

My 1080ti consumed 200-250w. My 4090 consumes 400w. The latter is tons more efficient than the former. Power draw isn't efficiency. That's what he means.

JustBenchingMy 1080ti consumed 200-250w. My 4090 consumes 400w. The latter is tons more efficient than the former. Power draw isn't efficiency. That's what he means.

You 1080 ti make more FPS on games in it launch years without any fake frames and scaling.

TumbleGeorgeYou 1080 ti make more FPS on games in it launch years without any fake frames and scaling.

Not true. 4k was completely out of the question in most games with the 1080ti. Even at 1080p it struggles in some heavy ones (wd2, ghost recon etc.).

But that's completely irrelevant to the efficiency discussion.

JustBenchingMy 1080ti consumed 200-250w. My 4090 consumes 400w. The latter is tons more efficient than the former. Power draw isn't efficiency. That's what he means.

Oh, ok.

thestryker6I wouldn't make any bets on when RDNA 5 launches, its relative performance, when it will appear in APU form or what it will look like in said form. The Medusa Point APUs are still using a RDNA 3.5 based architecture and Medusa Halo is the only known RDNA 5 APU which absolutely will not appear in socketed form. Unless AMD flips the script on how they design APUs Intel will be playing by itself with regards to integrated graphics.

Rumors on the Intel side point towards a Halo competitor with RZL, so late 2027-mid 2028, which would be using Xe3p. This would indicate Intel believes it can compare favorably enough with RDNA 5 or else it would have been canceled.

I don't really count halo. They always had bad value proposition. You lose upgradability part while paying for the high cost of a dGPU. I like this Xe3P config because it will be affordable. Moore's Law's primary thing is making compute accessible to everyone. I don't think it's a coincidence where in the past 5+ years computer prices have been increasing, because Moore's Law started to slow down. In the "14A" generation that will further accelerate.

Your Halo system will be a powerhog, expensive value part in 3-4 years, whereas a dGPU can be replaced and part of the cost recouped by selling. iGPUs have to be cheap. This 4+8+12 config is a right combo, as using a mass produced, standard platform means you can swap the CPU out for faster CPU, or if it breaks(although CPUs rarely do).

The point I wanted to make with RDNA5 is that it's going into PS6 handheld at end of 2027, and a 15W part is said to perform like 30W Pantherlake. Now, the scaling isn't linear, but I expect it to be at least 50% better than Pantherlake Xe3P at the same power. That means Xe3P being 25% isn't enough, especially because we don't see Xe4 until 2028-2029. This is why I hope Xe3P is another 50% gain over Xe3.

thestryker6That's why PTL's Xe3 implementation has 16MB cache. It's also why Strix/Gorgon Halo have Infinity Cache unlike their regular APU counterparts despite having twice the memory bandwidth.

It's not just the cache. Xe3 improves Hi-Z depth culling performance:
Culling refers to eliminating scenes that are rendered but not seen by the player. By eliminating that you reduce memory bandwidth use for all layers of the memory subsystem, from the diminuitive L1, to the large VRAM/shared memory. The specific microbenchmark for Depth Writes, related to occlusion culling has improved by a dramatic amount on Xe3. Intel's work paid off, as Pantherlake with pretty much same memory config outperforms predecessor Xe2 Lunarlake by 50%.

So whereas compute increased, they did work to offset the increase in bandwidth needed by the extra compute. Otherwise, they would have needed LPDDR6-1200 to beat Lunarlake by 50%, if it wasn't for combo of Depth Writes improvement + doubled GPU cache. For the 8-->16MB GPU cache, Intel says there's about 25% reduction in traffic, meaning it's equal to increasing memory bandwidth by 33%. 50% is also at same power level. Xe3 also scales with higher power, whereas Xe2 in Lunarlake kinda stops. Xe3 improves utilization of the shaders as well, which Xe1/Xe2 struggled as seen in Tom Peterson's overview. It's actually really nice, and Intel did a good job with Pantherlake.

Pantherlake SoC in general also brings memory compression, that's why it benches quite high in memory benchmarks.

www.notebookcheck.net/Framework-Laptop-13-Pro-review-The-king-of-repairability-is-back.1351246.0.html

You can see from the review above the Asus Expertbook with 8533 beats the XPS 14 with 9600 memory. The architecture is quite balanced, so using lower memory speeds isn't catastrophic for the iGPU. This is because Intel has an iGPU first approach(where it hurts them in dGPUs), whereas AMD has a dGPU first approach. It's easier and cheaper for AMD to repurpose the smallest dGPU die for the iGPU, which makes it not so optimal. Intel also gets iGPUs out first, whereas AMD gets dGPUs out first.

wNotyarDWill we have motherboard makers trying to beat one another in VCCGT phases quantity?

It's like looking for a board for an AMD APU all over again ☺️
But seriously, having one phase for this wouldn't work, so at least two, but tbh I'd like three or four.

DavidC1The point I wanted to make with RDNA5 is that it's going into PS6 handheld at end of 2027, and a 15W part is said to perform like 30W Pantherlake. Now, the scaling isn't linear, but I expect it to be at least 50% better than Pantherlake Xe3P at the same power.

There's a zero percent chance this is accurate in timeline or performance.

DavidC1I don't really count halo.

I only brought it up as a comparison point because there is no RDNA 5 APU coming any time soon except for Halo and Intel is rolling out a competitor for it using Xe3P. This indicates that Intel is not concerned about the level of performance RDNA 5 will bring compared to Xe3P.

DavidC1It's easier and cheaper for AMD to repurpose the smallest dGPU die for the iGPU

Except that they do not do this. There have been zero dGPUs without Infinity Cache since RDNA 2 and there have been zero iGPUs with Infinity Cache other than Halo. It's a glaring hole in AMD's iGPU design that I thought they would address with RDNA 3.5, but instead they opted for more CUs.