Rumor: Intel Skylake GPUs get 5xx naming
It looks like Intel will be revamping the naming schema for tyhe integrated procesorrs, at least that is what CPU-world is reporting today. "H" series CPUs will come with HD graphics 530 and Iris Pro 580. Xeon processors, based on Skylake-H design, will have a "professional" version of those GPUs, that will be branded as HD Graphics P530 and Iris Pro P580.
Here's the skinny from CPU-world:
Earlier reports indicated that the GPU will have up to 72 execution cores, or 50% more than Broadwell Iris Pro graphics. Having many more executions units, Skylake GPUs will be available in 4 tiers, from entry level GT1 to top-of-the-line GT4. Two intermediate tiers, GT2 and GT3, will have 24 and 48 execution units respectively.
To identify different GPU tiers and performance, consumer branding of Skylake GPUs will retain "HD graphics", "Iris" and "Iris Pro" names. Similar to previous generations, Skylake GPUs will also have model numbers, useful for rough (better or worse) comparison of GPU features and performance. However, there will be some differences.
The first HD model numbers, HD 2000 and HD 3000, were introduced in Sandy Bridge microprocessors. Ivy Bridge upped them to 2500 and 4000 to show their increased performance and improved feature-set. Haswell processors had GPU model numbers ranging from 4200 to 5200, and Broadwell had them in the 5300 - 6300 range. Skylake GPUs are not going to follow this trend, and they will employ three-digit numbers instead.
Celeron and Pentium CPUs will have HD graphics 510. This applies to both desktop and mobile budget processors. The "HD graphics" branding without model number will be reserved for Braswell ULV chips.
Core M systems on a chip will integrate HD graphics 515. "U" series SoCs will have higher power rating than Core M, consequentially their GPUs are going to perform better and will have higher model numbers. Overall, there will be three different GPU models in the "U" series, HD graphics 520, Iris 540 and Iris 550. The HD 520 will be available on 15 Watt "U" SoCs. We believe that Iris graphics will be available only on 28 Watt microprocessors.
Intel Core generation | GPU Model |
---|---|
Braswell |
HD Graphics |
Skylake Celeron, Pentium |
HD Graphics 510 (GT1, 12 eu's) |
Skylake Y-serie (Core M) | HD Graphics 515 (GT1.5, 18 eu's) |
Skylake U-serie |
HD Graphics 520 (GT2, 24 eu's), Iris 540 (GT3, 48 eu's), Iris 550 (GT3, 48 eu's) |
Skylake H-serie |
HD Graphics 530 (GT2, 24 eu's), Iris Pro 580 (GT4e, 72 eu's) |
Xeon Skylake H-serie |
HD Graphics P530 (GT2, 24 eu's), Iris Pro P580 (GT4e, 72 eu's) |
Skylake S-serie, desktop |
HD Graphics 530 (GT2, 24 eu's) |
First desktop Skylake Core i5 and Core i7 processors will have GT2 graphics, which corresponds to HD graphics 530. The same GPU type will be also used on Core i3 Skylake models, via CPU-world.
Senior Member
Posts: 528
Joined: 2005-02-23
Why can't they make discrete GPU with 4-6 times more execution units in it?
Senior Member
Posts: 13713
Joined: 2004-05-16
Mostly size constraint. Haswell dedicates 174mm^2 to the GPU. Simply multiplying this by four gives you a GPU chip 100mm^2 larger than the 980Ti. 600mm2 is the limit on reticle size.
Also while Intel's architecture is decent at shader performance, it's terrible at everything else. Pixel/Texel/Geometry. It's worse at all these things then Nvidia's offerings at the same level. So it wouldn't be as simple as just multiplying the slices, they'd need to redesign some stuff too in order to be competitive.
It may happen in the future, but as of now they aren't even close to competing with the other companies. Plus Nvidia owns a huge swath of patents on discreet graphics processing. It wouldn't surprise me if that also plays a role on Intel moving into that market.
Senior Member
Posts: 4186
Joined: 2004-09-28
What Intel needs to do as far as their IGPU goes is ditch the HD series and putting out their Iris Pro IGPUs.
Senior Member
Posts: 528
Joined: 2005-02-23
Mostly size constraint. Haswell dedicates 174mm^2 to the GPU. Simply multiplying this by four gives you a GPU chip 100mm^2 larger than the 980Ti. 600mm2 is the limit on reticle size.
Also while Intel's architecture is decent at shader performance, it's terrible at everything else. Pixel/Texel/Geometry. It's worse at all these things then Nvidia's offerings at the same level. So it wouldn't be as simple as just multiplying the slices, they'd need to redesign some stuff too in order to be competitive.
It may happen in the future, but as of now they aren't even close to competing with the other companies. Plus Nvidia owns a huge swath of patents on discreet graphics processing. It wouldn't surprise me if that also plays a role on Intel moving into that market.
Does die size really matter on discrete PCIE video card?
Haswell Iris Pro 5200 offers 830 GFLOPS having 28W TDP and 22nm tech process. It would be 3300 GFLOPS with TDP less than GTX 960. May be it won't beat GTX 970 in real games, but what about GTX 960? In real life Iris Pro 5200 provides higher fps than 930M.
And then there is 14nm and GT4e.
And isn't Intel has crosslicensing agreement with AMD? Or is it CPU-related only?
Senior Member
Posts: 1181
Joined: 2013-02-22
What Intel needs to do is use the full gpu core for the whole mobil cpu line since it would even bottle neck an i3 then we would have an interesting mobile market.