[AT] Iris Pro in a socket

[CHADBOGA]

It's far from clear if the EU's in this product will be Gen 7 or Gen 8.

I suspect they may be Gen 7, so it isn't as exciting as it could be.

 

[AtenRa]

If this unlocked SKU has the full Broadwell IrisPro graphics with eDRAM and cost less than $300 then it would be worth considering. Otherwise it will completely be irrelevant for 99% of the users and only extreme Overclockers will find it a nice toy to play around.

 

[Sweepr]

Broadwell, 5th Gen Intel Core, Iris Pro... Pretty sure were're talking about a GT3e ''Gen 8'' iGPU part here, not Haswell.

 

[ShintaiDK]

Haswell GT3e cant fit in LGA1150 for that matter. So yes, pretty obvious ist Broadwell.

 

[AtenRa]

Broadwell GT3e is the one with 48 EUs and eDRAM correct ?? If that is Iris Pro then what about GT4 ?? Is GT4 BGA only (Apple and OEMs ??)

If the unlock SKU has the 48 EU iGPU part + eDRAM and the retail price is maximum of $300, then it should be worthy of recommendation for High-End SFF builds.

 

[NTMBK]

Are you sure ???
Lets eliminate the CPU and see what happens with graphics only. From AT Kaveri review.

Core i7 4770R is 65W TDP.

Even a low 45W Kaveri iGPU is faster than Iris Pro in Bioshock. 65W Kaveri is 17% faster.

In Tomb Raider Iris Pro is ~10% Faster than 65W Kaveri.

Well, even vanilla Intel HD graphics are faster than Iris Pro. It seams that eDRAM is not always the best solution or the Intel driver is even worst than we believe.

Equal to 65W Kaveri.

So, where did you see that Iris Pro creams Kaveri at the same power consumption ?? And this is 22nm FinFet vs 28nm. Intel is way behind in graphics performance.
Also to note here that Kaveri 45W is equal or faster than the 100W Richland parts. AMD should of bring a GDDR-5 SKU even if the price was at $300+, it would completely eliminate the competition.

That Kaveri review used the Gigabyte Brix Pro for its Iris Pro benchmarks. It's well known that the Brix Pro has MASSIVE problems with heat and throttling. Details here: http://techreport.com/review/26166/gigabyte-brix-pro-reviewed/4 Meanwhile every other processor was on an open motherboard testbed. When you see results like the 4770K beating the 4770R, you should start checking the test methodology. That Iris Pro was being throttled like crazy, because Gigabyte shoehorned a 65W processor into a tiny, poorly ventilated chassis. It will be very interesting to see how Iris Pro in a socket does, when mated to a decent CPU cooler.

 

[beginner99]

Sounds cool but then as long as the iGPU is only used as a GPU and basically dead silicon with a dGPU, it's not that interesting. I would rather prefer hexa- or octo cores in mainstream with weaker or no iGPU at all.

 

[NTMBK]

It's far from clear if the EU's in this product will be Gen 7 or Gen 8.

I suspect they may be Gen 7, so it isn't as exciting as it could be.

No chance that Intel would mate Gen 7 to Broadwell just for this low-volume SKU. Swapping out the graphics would mean a costly redesign of the chip.

 

[NTMBK]

Sounds cool but then as long as the iGPU is only used as a GPU and basically dead silicon with a dGPU, it's not that interesting. I would rather prefer hexa- or octo cores in mainstream with weaker or no iGPU at all.

I do wish that game engines would start using the iGPU for something when a dGPU is present- calculating physics perhaps, or something like that.

 

[ShintaiDK]

Broadwell will like CherryTrail have Gen8 GPU.

 

[mikk]

It's far from clear if the EU's in this product will be Gen 7 or Gen 8.

I suspect they may be Gen 7, so it isn't as exciting as it could be.

It is clear. Broadwell comes with a Gen8 GPU.

 

[CHADBOGA]

It is clear. Broadwell comes with a Gen8 GPU.

Francois Piednoel refused to confirm this on Twitter when asked if Gen 7 or Gen 8.

 

[IntelUser2000]

Never heard that one before - have a source for such? After all, there's no question that texture sampling is a weak point of Intel's graphics thus far.

Actually I can provide evidence:

http://www.anandtech.com/show/5871/intel-core-i5-3470-review-hd-2500-graphics-tested/4

3DMark Vantage Texture Fill:

2 samplers HD 4000: 6.1 @ 1.15GHz - 5.3 points/GHz
1 sampler HD 3000: 4.56 @ 1.35GHz - 3.38 points/GHz
1 sampler HD 2500: 2.93 @ 1.1GHz - 2.66 points/GHz

https://www.google.ca/search?q=Intel+Bay+Trail+mpixels+filetype%3Apdf&ie=utf-8&oe=utf-8&aq=t&rls=org.mozilla:en-GBfficial&client=firefox-a&channel=np&source=hp&gfe_rd=cr&ei=EiQsU8_oA-_L8ge024DIAg

Page 12: Bay Trail presentation.

@ 700MHz it claims 1050 MPixel/sec. Now that doesn't make sense, unless it runs at asynchronous frequency. Assuming 2 TMUs, it would run at 75% frequency, and above with Anand's bench we get 78%, which is pretty close.

Same trick Intel used back with GMA X3000: http://techreport.com/review/12195/intel-g965-express-chipset

"What we may be seeing here is the result of different clock domains for the shader processors and the IGP's back end;"

They "fixed" it later, I believe with GMA 4000 series.

That means real texture throughput of Iris Pro running at 1.3GHz is 15.6GTexels/s, not 20.8GTexels/s. The Geforce 650M which was a popular comparison had nearly twice the fillrate in some variants, like the Geforce 650M used in Anand's Iris Pro comparison.

 

[IntelUser2000]

Are you sure ???
Lets eliminate the CPU and see what happens with graphics only. From AT Kaveri review.

Also to note here that Kaveri 45W is equal or faster than the 100W Richland parts. AMD should of bring a GDDR-5 SKU even if the price was at $300+, it would completely eliminate the competition.

"Let's eliminate the CPU".

Why? At those settings its not playable at all. One of the bigger changes in Broadwell is supposed to be related to texturing performance, which is likely responsible for lack of performance in more extreme settings/resolutions.

Francois Piednoel refused to confirm this on Twitter when asked if Gen 7 or Gen 8.

They can do that, but it would be unnecessary work, since Broadwell uses Gen 8 arch.

 

[NTMBK]

Francois Piednoel refused to confirm this on Twitter when asked if Gen 7 or Gen 8.

It's Gen 8. Broadwell is Gen 8, as demonstrated by the commits to Linux graphics drivers detailing changes for Broadwell. I'm not surprised Francois stayed quiet, because if he starts blatantly leaking stuff before Intel are ready to talk about it then the PR department will have his head on a spike.

 

[frozentundra123456]

@ Aten Ra

At more playable settings, they trade blows. The graphs you show, Seriously???.

Lets look at the absolute differences: Bioshock, less than 3fps, Tomb Raider 0.5, etc, and all are under 15 FPS and also beaten by an ancient HD6750. Kind of like arguing about who is bringing the sharpest knife to a gunfight.

 

[crashtech]

The graphs you show, Seriously???.

My CPU can play a slideshow faster than your CPU!

 

[Shivansps]

Broadwell GT3e is the one with 48 EUs and eDRAM correct ?? If that is Iris Pro then what about GT4 ?? Is GT4 BGA only (Apple and OEMs ??)

If the unlock SKU has the 48 EU iGPU part + eDRAM and the retail price is maximum of $300, then it should be worthy of recommendation for High-End SFF builds.

Shoot in the dark here, but GT4 might has been a failed attempt by Intel to win console APU.

 

[mikk]

Francois Piednoel refused to confirm this on Twitter when asked if Gen 7 or Gen 8.

This guy is meaningless whether or not he confirmed it.

 

[Sweepr]

Shoot in the dark here, but GT4 might has been a failed attempt by Intel to win console APU.

They'd rather sell it as an expensive chip to Apple and others than low-margin <$500 consoles.

 

[mrmt]

Shoot in the dark here, but GT4 might has been a failed attempt by Intel to win console APU.

I doubt that Intel would have bothered designing and building a chip in a bleeding edge node for the kind of margins AMD gets on their console chips.

 

[AtenRa]

Why? At those settings its not playable at all.

@ Aten Ra

At more playable settings, they trade blows. The graphs you show, Seriously???.

We are not evaluating APUs here as a product but only the iGPU performance. Playable frame rates is not what we are seeking for.
Eliminating the CPU influence (lower resolutions and image quality settings) really show the iGPU performance of each architecture. And those AT graphs clearly demonstrate that GPU wise, Haswell IrisPro is not that much more power efficient than Kaveri.

 

[crashtech]

Intel fanboys cry about the weakness of their iGPUs being revealed, while AMD fanboys cry about the weakness of their CPUs being revealed.

At least Intel seems to be improving their iGPU (LOTS of room there); I wish the same could be said about AMD and their CPUs.

 

[Khato]

Actually I can provide evidence:

http://www.anandtech.com/show/5871/intel-core-i5-3470-review-hd-2500-graphics-tested/4

3DMark Vantage Texture Fill:

2 samplers HD 4000: 6.1 @ 1.15GHz - 5.3 points/GHz
1 sampler HD 3000: 4.56 @ 1.35GHz - 3.38 points/GHz
1 sampler HD 2500: 2.93 @ 1.1GHz - 2.66 points/GHz

And assuming that the HD4600 and Iris Pro numbers from here - http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/15 - are simply 30x those in the review you linked then we can add...
4 samplers HD 5200: 15.56 @ 1.3GHz - 11.97 ponits/GHz
2 samplers HD 4600: 7.64 @ 1.35GHz - 5.66 points/GHz

Though what precisely is the point of the above? Unless we know the exact breakdown of texture sampling calls in the test and how many textures per clock the design samples for each of those instructions it doesn't tell us anything really.

Page 12: Bay Trail presentation.

@ 700MHz it claims 1050 MPixel/sec. Now that doesn't make sense, unless it runs at asynchronous frequency. Assuming 2 TMUs, it would run at 75% frequency, and above with Anand's bench we get 78%, which is pretty close.

Once I found the presentation... It quite clearly states Fill rate in MPixel/sec there as well - what does that have to do with texture sampling? It's purely a ROP measurement. Regardless, the 3DMark vantage numbers for IVB show pretty much the exact same 1.5 pixel/clock as reported there. But that's not indicative of a separate clock frequency - if it was then why do the AMD and NVIDIA parts also show fractional pixel/clock?

 

[Khato]

We are not evaluating APUs here as a product but only the iGPU performance. Playable frame rates is not what we are seeking for.
Eliminating the CPU influence (lower resolutions and image quality settings) really show the iGPU performance of each architecture. And those AT graphs clearly demonstrate that GPU wise, Haswell IrisPro is not that much more power efficient than Kaveri.

Actually, all cranking up to maximum settings does is negate the majority of the advantage derived from the eDRAM due to its limited size. Which is why there's a pretty obvious trend of the AMD parts tested with 1866/2133 MHz DDR3 pull ahead of Iris Pro with 1600 MHz DDR3 as scene complexity/graphics memory footprint increases.

Of course, feel free to blame Intel for targeting their solution at playable frame rates. I tend to call that proper engineering.