• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

What about Skylake Iris Pro?

M

Mondozei

Golden Member
Piggybacking off the recent Haswell Iris Pro thread, I was wondering if there's any good guesstimate of where the Skylake top iGPU would land in terms of past GPUs. I think the basic specs have essentially leaked.
 
I remember the speculation that it had 20% more shaders(?) and 20% more clock speed... it probably won't be 40% faster... but likely 25%.

I was hoping for more... something to make the AMD APU completely redundant. 😉 AMD's still got a dog in the race there.
 
It will definitely be 40% faster than Broadwell Iris Pro, which we've only seen from the rMBP. Any numbers there? Notebookcheck says it matches an 820M.
 
There was no Iris Pro in Macbook Pro, because there is no such CPU from Intel with 45W of TDP, and Iris Pro from Broadwell line.

Iris Pro that is in MBPs is Haswell gen.

Just to clear things up, because you messed with information a bit.
 
Cloudfire777 said:
Iris Pro should reach GM107 ish, but will probably be a little slower.
Click to expand...

Given that iris was generally slower than GK107 in full form, and that GM107 is 75% faster than GK107, Iris Pro will not approach GM107 in performance at the estimated 25-40% improvement. Iris Pro will need a maxwell like jump to keep up.
 
It is coming on top of Broadwell of course. If they can't get even 75 per cent net over Haswell from a die shrink its slightly worrying in terms of keeping up with dGPU's.

One thought - isn't it also quite possibly going to get on a 95w model? That'd immediately give it a good chunk of extra power draw over anything else they've yet tried it in.
 
Depends on which Iris Pro. GT4e is supposed to have 72 EU, so if it's clocked at around the same speed as Haswell GT3e (40EU) it could be almost twice as fast. That might put it into 650Ti territory, which is starting to creep into the range of acceptable framerates at 1080p and medium settings.
 
broadwell does have a full iris pro with the new 65 watt desktop processors releasing june 2nd. Skylake is going to be 40% faster then the full broadwell iris pro releasing on the 65 watt desktop chips?
 
MrTeal said:
Depends on which Iris Pro. GT4e is supposed to have 72 EU, so if it's clocked at around the same speed as Haswell GT3e (40EU) it could be almost twice as fast. That might put it into 650Ti territory, which is starting to creep into the range of acceptable framerates at 1080p and medium settings.
Click to expand...

Well its supposed to have a faster clock so it may even be a bit higher than 50% increase compared to haswell but what about compared to the new broadwell desktop chips coming out the 65 watt unlocked ones?

Seems broadwell GT3e on desktop is 48 eu clocked at 1ghz with 64MB edram and 128bit memory interface. So yes there will be 50% more eu's and double the amount of edram. So even if it's clocked the same that 40% increase in performance seems very reasonable, maybe even 50% increase compared to broadwell iris pro.

I think it will have about a 60% increase over haswell and a 50% increase over broadwell iris pro. That is some serious gains which should allow full ultra 720p gaming and medium to medium high 1080p gaming at 40+ fps
 
Last edited:
The problem with Iris Pro will remain the same, edram cache tanks at 1080p. Also as soon as texture quality is turned up, we're back to being system ram bandwidth limited, so that hampers any gains in performance. You can cram 10x the EU and it won't make a difference.

Best case scenario is 720p gaming, where it should reach its max perf gains.

Basically suffers the same problem with AMD APUs, its just not viable for 1080p mainstream gaming due to bandwidth limitations. None of these issues will be solved until APUs use HBM for a high bandwidth vram for game assets.
 
Silverforce11 said:
The problem with Iris Pro will remain the same, edram cache tanks at 1080p. Also as soon as texture quality is turned up, we're back to being system ram bandwidth limited, so that hampers any gains in performance. You can cram 10x the EU and it won't make a difference.

Best case scenario is 720p gaming, where it should reach its max perf gains.

Basically suffers the same problem with AMD APUs, its just not viable for 1080p mainstream gaming due to bandwidth limitations. None of these issues will be solved until APUs use HBM for a high bandwidth vram for game assets.
Click to expand...

No I really think with the 72EU 9th gen cores 1ghz+ clock 128MB edram on skylake iris pro many games will have playable frame rates at medium settings. As long as the edram cache is smart enough to fill the 128MB with the most important most used data that should help relieve some of the data needed from system ram which should decongest the limited bandwidth by something at least. Medium 1080p for most games should be ok and some games even high may be reached. I'd say many games you'd be able to get medium 1080p settings and maintain a 30+ fps no problem 30 is playable but it is not perfectly smooth i try to shoot for 40+ fps (well of course 60+ on a dedicated card) but for the sake of arguement 30+ stable fps is playable. And this igpu should be having lotsa HTPC living room gamers having fun in 1080.

Not to mention 720p should be good to go for just about any game on the skylake igpu with high-ultra settings and even some aa to help smooth things out on the low resolution and be well in the target 40+ fps.
 
Last edited:
Well if you've noticed the recent rise in vram requirements for dGPU, I don't think a small edram increase is going to cut it in modern games on medium, forget about high.

1080p is still the mainstream gaming resolution and that's where next-gen APUs have to perform well at in order to displace the dGPU in a entry gamer's rig.

It also has to make sense when compared to the cost of a CPU + dGPU combo that gives similar or better performance. If Skylake Iris Pro ends up slower & more expensive, its a non-starter outside of special notebooks.

Intel needs to jump on the HBM tech asap.
 
Laststop311 said:
No I really think with the 72EU 9th gen cores 1ghz+ clock 128MB edram on skylake iris pro many games will have playable frame rates at medium settings. As long as the edram cache is smart enough to fill the 128MB with the most important most used data that should help relieve some of the data needed from system ram which should decongest the limited bandwidth by something at least. Medium 1080p for most games should be ok and some games even high may be reached. I'd say many games you'd be able to get medium 1080p settings and maintain a 30+ fps no problem 30 is playable but it is not perfectly smooth i try to shoot for 40+ fps (well of course 60+ on a dedicated card) but for the sake of arguement 30+ stable fps is playable. And this igpu should be having lotsa HTPC living room gamers having fun in 1080.

Not to mention 720p should be good to go for just about any game on the skylake igpu with high-ultra settings and even some aa to help smooth things out on the low resolution and be well in the target 40+ fps.
Click to expand...
if this can be done, it will be in my next laptop. I don't need high or ultra I just need medium settings for 1080p. a medium gaming latop that weights around 3 to 5 pounds would be heaven to me since I do move around alot.
 
boozzer said:
if this can be done, it will be in my next laptop. I don't need high or ultra I just need medium settings for 1080p. a medium gaming latop that weights around 3 to 5 pounds would be heaven to me since I do move around alot.
Click to expand...

The problem is the laptop ones dont have the thermal headroom and power to run like the desktop gt4e will run. Laptop one will have lower clocks for sure. I was talking about the desktop version of the gt4e. Laptops may top out at gt3e. Laptops should be able to run medium (sometimes high, rarely ultra) 720p + 4x aa at 30-40+ fps stably with the standard voltage 45w tdp skylake mainstream laptop cpus igpu.
 
That should be available right now with a dgpu like the GTX860M. Lower/mid gaming laptops are not really the huge monstracities they used to be. Only problem is the dgpu makes them expensive.
 
frozentundra123456 said:
That should be available right now with a dgpu like the GTX860M. Lower/mid gaming laptops are not really the huge monstracities they used to be. Only problem is the dgpu makes them expensive.
Click to expand...

The whole point was he wanted a lighter more portable gaming laptop. When you have strong igpu and no discrete gpu you gain a lot of flexibility as far as design goes. Not only is it more empty space it's also less watts of heat to disperse and less electricity drawn from the battery and lower overall temps and longer battery life.

Plenty of reasons for some 1 wanting an igpu that can game at medium 1080 settings no AA and a stable fps that does not drop below 30 fps at the very least. A 1080 igpu that can do that with the current latest games and is dx 12 ready really would be a game changer. It would make light portable longer lasting gaming laptops finally possible and actually good. And maybe a manufacturer will even do the unthinkable and leave it as thick as a dgpu system and just make the battery massive. Would be incredible to have a 6-7 hour gaming time on battery
 
Silverforce11 said:
Well if you've noticed the recent rise in vram requirements for dGPU, I don't think a small edram increase is going to cut it in modern games on medium, forget about high.

1080p is still the mainstream gaming resolution and that's where next-gen APUs have to perform well at in order to displace the dGPU in a entry gamer's rig.

It also has to make sense when compared to the cost of a CPU + dGPU combo that gives similar or better performance. If Skylake Iris Pro ends up slower & more expensive, its a non-starter outside of special notebooks.

Intel needs to jump on the HBM tech asap.
Click to expand...

Edram needs drivers to do well. Haswell Iris pro was pretty overspecced in terms of L4 cache according to intel so 128MB should be plenty for Skylake.

As far as Nvidia increasing cache sizes, GM204 is 2 MB and GM200 3 MB, both of which had significant impacts on bandwidth requirements and are significantly smaller than iris pro, especially when normalizing to performance.

Xbox one (32 MB ESRAM) performs well with edram significantly reducing bandwidth requirements.

I agree that HBM is pretty much a must for future products as long as costs are reasonable. I also hope intel reworks their cache structure.
 
Enigmoid said:
I agree that HBM is pretty much a must for future products as long as costs are reasonable. I also hope intel reworks their cache structure.
Click to expand...

What I expect to eventually see is HBM used as L4 cache for both CPU and GPU on the same die, just like the eDRAM in Iris Pro except faster and with larger capacity. You might see a standard i5 have 1GB-2GB of HBM on-die, with 8GB or so of traditional DDR4 DIMMs backing it up.
 
JDG1980 said:
What I expect to eventually see is HBM used as L4 cache for both CPU and GPU on the same die, just like the eDRAM in Iris Pro except faster and with larger capacity. You might see a standard i5 have 1GB-2GB of HBM on-die, with 8GB or so of traditional DDR4 DIMMs backing it up.
Click to expand...

It all depends on price and density. I dont think "combo" solutions will last long, if they even get to appear in the consumer space. With HMC/HBM2/HBM3 you can easily add 8-16GB. And since mobile is king, why even use DDR4 to begin with. We already see a lot of laptop manufactors soldering the memory while using more expensive LPDDR.

Fixed memory sizes for anything but server CPUs is coming fast.
 
Last edited:
ShintaiDK said:
It all depends on price and density. I dont think "combo" solutions will last long, if they even get to appear in the consumer space. With HMC/HBM2/HBM3 you can easily add 8-16GB. And since mobile is king, why even use DDR4 to begin with. We already see a lot of laptop manufactors soldering the memory while using more expensive LPDDR.

Fixed memory sizes for anything but server CPUs is coming fast.
Click to expand...

For once I agree wholeheartedly with you. I do not see the point in system RAM using DDR4 when a single HBM gen 2 stack with a 1024 bit wide bus can provide 8 GB with a 8Hi stack by H2 2016. This is how Pascal GP100 will get to 32 GB with a 4096 bit bus and 4 HBM gen 2 (8 Hi) stacks each of 8 GB capacity.

For atleast a mainstream gaming notebook using a Zen APU with HBM gen 2 I don't see the need for more than 8 GB unified memory in 2017. We will see higher density HBM stacks as time goes by. HBM is a JEDEC standard which is expected to have a long life and capacities of upto 32 GB per single HBM stack in future generations.

http://www.cs.utah.edu/thememoryforum/mike.pdf

The average PC user definitely including mainstream gamer is not going to need more than 8GB in the near future and definitely 32 GB through future versions of HBM is more than enough for a long long time. There is a host of advantages in eliminating DDR4 -
1.) Reduction in system power as all memory is on interposer so no need for driving I/O through motherboard saving power.
2.) Motherboard size and form factor reductions - Imagine a PC in a card which resembles something Jen Hsun Huang had in his hand.

http://wccftech.com/nvidia-generation-pascal-gpu-feature-3d-stacked-memory-nvlink-1-tbs-bandwith/

3.) Exciting new notebook and desktop form factors.

Other than the hardcore high end enthusiast gamer who wants a ton of PCI-E 16x slots for multi GPU gaming and a high end server CPU in a desktop version the vast majority will be perfectly satisifed with a integrated solution which works very well for their needs. In summary other than servers, high end workstations and high end enthusiast gaming PCs there is no need for system RAM other than APUs or SoCs with unified HBM memory.
 
Its 8 GB shared RAM so take that into account. If 3 GB of that is used as vram (big zen APU) that leaves 5 GB for the game + OS + background progams which may be cutting it in two years (add to that the fact that windows doesn't like going above 90-95% utilization).
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top