3DMark for Windows 8 to unify test suite between ARM and x86 solutions

[Exophase]

Intel has recently upgraded their integrated GPU performance pretty much by a factor of 2 every time they have introduced new products. I don't find it in any way hard to believe they achieved 60% boost by going to a much smaller process (and 3d transistors to boot). Anything less than that would be an utter failure.

It's not that I'm doubting that Intel will get a 60% performance boost with IB's IGP, it's that you were looking at the performance of a top bin HD 3000 with a much larger thermal budget than any ULV chip will have, meaning much higher clock speeds. A ULV IB at 17W won't have a GPU that's 60% faster than what's in a 95W Sandy Bridge.

That's a completely different product, based on older (and cheaper) technology.

I guess I missed that.. kind of easy to when after all this time there appears to be only one Sandy Bridge tablet, despite 17W CPUs being available for a good while. This one just came out and it's hard to find good reviews for it, much less anything gaming related. Got some benchmarks?

Kind of makes it hard to believe there will be many IB tablets at launch.

 

[rpg.314]

It's not that I'm doubting that Intel will get a 60% performance boost with IB's IGP, it's that you were looking at the performance of a top bin HD 3000 with a much larger thermal budget than any ULV chip will have, meaning much higher clock speeds. A ULV IB at 17W won't have a GPU that's 60% faster than what's in a 95W Sandy Bridge.

As long as the 17W IB is 60% faster than the 17W SB, intel is doing fine.

 

[Exophase]

As long as the 17W IB is 60% faster than the 17W SB, intel is doing fine.

... you are completely missing the point. Did you read the posts? >_>

 

[rpg.314]

... you are completely missing the point. Did you read the posts? >_>

I did read them. I am not arguing that a 17W IB will beat a 95W SB by 60%.

I am just saying that as long as that 60% improvement happens across the board, even a 17W IB should be fine.

 

[Exophase]

I did read them. I am not arguing that a 17W IB will beat a 95W SB by 60%.

I am just saying that as long as that 60% improvement happens across the board, even a 17W IB should be fine.

Fine for what? Do you have any benchmarks of a 17W SB that we can use for comparison?

 

[rpg.314]

Fine for what? Do you have any benchmarks of a 17W SB that we can use for comparison?

I am expecting the IB slates to be quite competitive with ARM slates in GPU performance, considering the lame memory bandwidth, process inferiority and smaller area budgets for ARM soc's.

 

[sebbbi]

And this is where we part. Apparently you and I have radically different notions of 'games running properly' on a tablet. What you consider as 'some developers might decide to go further' has been proven by Apple to be the very fundamental of successful tablet/touch-screen ecosystems. It's not optional in any way. You can technically neglect it, but that'd result in a badly designed software. So much for the AAA games.

You have to understand that Windows has already a huge user base, and all the new notebooks, netbooks and desktop computers will be selling with Win8 on it. Many old time users will also upgrade their existing computers to Win8 (this time it's also easier, since the system requirements haven't increased). Windows app store isn't just for tablets, it's for all Windows computers.

Now if you are designing a Win8 metro application to the Windows app store, you should expect that almost 90% of your user base will have just mouse + keyboard connected to their computer (at Win8 launch). If Win8 tablets become really popular, maybe in a few years we will have 50%/50% distibution, but this is just a guess game and not something you should rely on if you are spending a lot of money in developing a new game to the platform.

Here is a link to the Microsoft guidelines for Win8 user interaction:
http://msdn.microsoft.com/en-us/library/windows/apps/hh465370(v=VS.85).aspx

I picked one line from the documentation:
Touch-optimized applications must support efficient and intuitive mouse and keyboard interactions that expose equivalent functionality

Win8 is not equivalent to iOS and Android. You cannot design apps to it that only support touchscreen control. Those apps will not even pass the certification process. If you browse the Metro developer site a bit more, you will also notice that Metro apps have native support for Xinput API to control your games with a Xbox 360 controller. Metro also supports the Xbox 360 sound API and Microsoft also brings full Xbox Live to PC with Win8. I personally think all this was done to attract Xbox 360 developers to port their products to Win8. Microsoft has a huge developer base developing top quality games for their other products (Windows/PC & Xbox 360), why not use this resource to bring top quality entertainment to the tablets as well?

It will be interesting to see how things pan out. The platform will surely attract console developers (Xbox 360 APIs and Xbox Live), PC developers (full backwards compatibility of existing DirectX games) and handheld/tablet developers (touchscreen input). Lets see who will adapt to the platform best. I personally predict that games that are very natural to control by all the three possibilities (touch, kb+mouse, gamepad) will be the ones that will be the most successful.

Fine for what? Do you have any benchmarks of a 17W SB that we can use for comparison?

There's plenty of Macbook Air reviews around the net. It has the same 17W i5 2764M as the Samsung Slate. Some previews compare performance in Windows 7 as well (installed with Boot Camp). Gaming performance in general is better under Windows, since GPU manufacturers have spent much more time in optimizing their DirectX drivers.

Now this is just speculation (or extrapolation if you want to use that term): Top of the line (four core) Sandy Bridge parts have 95W power consumption. Top of the line Ivy Bridge will have 77W. Now is we assume that the 60% faster GPU and 20% faster CPU figures apply to the top of the line (like you previously estimated), we can calculate a rough estimate to the power consumption to the energy efficient models as well. 17W * 77/95 = 13.7W. Also in this estimate I assume that power usage grows linearly as you scale clocks. The growth is actually more than linear, so this linear estimate is likely too conservative (the energy efficient chips likely consume even less). Now if the manufacturerers are happy with the 2764M perfomance, they might instead opt for lower clocked Ivy Bridge products. Now assuming linear dependence of clocks to performance, a 20% down clocked Ivy Bridge should have equal CPU performance to a 2764M with similar specs and 28% higher GPU performance (1.6*0.8). This would bring (linear estimate again) power consumption down to 10.9W. But as all these linear estimates are conservative (clocking CPU up doesn't bring linear gains in performance, and brings more than linear increase in power consumption), it wouldn't be a bad estimate to hope for a 10W Ivy Bridge that has slightly more powerful CPU and around 30% higher performance GPU compared to the 2764M.

Look at the iPhone. Apple has increased it's GPU performance by 5x every year (two years in a row), and the battery life of the device is still the same. Intel has the most advanced manufacturing processes in the semiconductor industry. It's not hard to believe that they reached 60% increased performance + lower power consumption in their newest chip design. Intel is telling us that the new tri-gate 3d transistors alone would halve the power consumption and allow up to 37% higher clocks. Add in all the improvements they have implemented to their next generation GPU core, and the 60%+ performance estimate starts to looks a bit conservative, and clearly indicates that power efficiency was a bigger goal for Intel this time.

 

[Ailuros]

Look at the iPhone. Apple has increased it's GPU performance by 5x every year (two years in a row), and the battery life of the device is still the same.

I decided not to further participate in this conversation, but it's high time you get all of your facts straight. The above is merely one example out of many.

 

[sebbbi]

I decided not to further participate in this conversation, but it's high time you get all of your facts straight. The above is merely one example out of many.

Of course the iPhone is just an example. It's a very good example to show you that 60%+ GPU performance + reduced power consumption is not an impossible thing to accomplish in a year of SoC hardware development. As Apple (Samsung/IMG tech) can reach 5x GPU improvement in a year (twice in a row) I woudn't doubt for a second that Intel could not reach these (relatively smaller) improvements in the same time frame.

I am eagerly waiting for the new Win8 3DMark. For the first time we can start comparing the reduced PC hardware to the boosted up phone hardware. Only good things can happen when the competition in this hardware segment heats up.

Here's a youtube vid of someone playing Crysis 2 on his 2011 Macbook Air (it features the same i5 2764M and same 4GB of memory than the Samsung Slate):
http://www.youtube.com/watch?v=W80RqSMU9ME

 

[Ailuros]

Of course the iPhone is just an example. It's a very good example to show you that 60%+ GPU performance + reduced power consumption is not an impossible thing to accomplish in a year of SoC hardware development. As Apple (Samsung/IMG tech) can reach 5x GPU improvement in a year (twice in a row) I woudn't doubt for a second that Intel could not reach these (relatively smaller) improvements in the same time frame.

Do you actually understand what I'm writing? Get your facts straight. Apple DID NOT increase GPU performance by 5x times in a row, since they use some sort of tick tock strategy.

iPhone3GS = SGX535
iPhone4 = SGX535 (A4 SoC)
iPad = SGX535 (A4 SoC)

(yes with varying frequencies but it doesn't mean any Nx times performance between those)

iPad2 = SGX543MP2@250MHz (A5 SoC)
iPhone4S = SGX543MP2@200MHz (A5 SoC)

Even worse since A4 Samsung is barely manufacturing the SoCs afaik; design and development is a pure Apple affair ever since.

As for Intel it's not they can't, it's so far a bunch of idiotic design decisions that kept them from entering seriously the embedded market and not whether they can or not. GMA600 for smart-phones is barely a SGX535@200MHz (iPad1 GPU) and since I checked a couple of days ago it wasn't even shipping yet.

I am eagerly waiting for the new Win8 3DMark. For the first time we can start comparing the reduced PC hardware to the boosted up phone hardware. Only good things can happen when the competition in this hardware segment heats up.

You're still haven't grasped the concept for anything embedded yet and it's so bleedingly obvious that it hurts. There's no reasonable comparison between a device that consumes barely 1W, another one that consumes a handful and other ones that consumes a few dozens of Watts. If that still isn't understandable then there's no hope in repeating it over and over again.

Performance will continue to scale for all markets in the future, but power consumption NOT significantly as stated above. Therefore there are always going to be lightyears in differences between a netbook and a smart-phone or a tablet and a low end PC SoC as simple examples. Since win8 will scale from embedded to desktop in its own way, it makes sense to create a unified synthetic benchmark for all of the win8 platforms, but it's not going to mean that a smart-phone of the immediate future will suddenly burn a gazillion of Watts and you'll run around with a battery backpack to power it.

 

[sebbbi]

but it's not going to mean that a smart-phone of the immediate future will suddenly burn a gazillion of Watts and you'll run around with a battery backpack to power it.

I though we were talking about Win8 tablets and 3DMark here, not about mobile phones. Samsung has clearly shown that a Win7/8 tablet powered by Intel's Sandy Bridge can be produced. It has a rated battery life of 7 hours, and it has been getting good reception from the press. I have said nothing about putting a Sandy Bridge (or Ivy Bridge) inside a mobile phone.

Try the device before you claim something ridiculous like needing a battery backpack to run it. In addition to all the performance goodies (and a very nice display) it has a Wacom pressure sensor with 1024 levels of sensitivity (graphics artists love it). We have enough iPad clones in the market already. I hope that other Win8 tablet manufacturers would come up with devices as innovative as this.

But as you said earlier, this discussion leads nowhere. Lets wait for Win8 launch, and see how things proceed. Intel isn't the only one bringing their guns at the show. AMDs forthcoming 28nm Wichita chips (9W) sound also very promising (their previous Zacate APU has very good GPU performance for watt).

 

[Exophase]

I'm still looking for hard evidence that IB's CPU performance will actually increase by 20% at the top bin. I think this is an incorrect attribution of Intel's announcements about the 22nm node, that it can achieve 18% performance improvement at 1V levels. There's no way that IB can increase clocks by 20% AND decrease power consumption by nearly 19% all while increasing GPU performance by 60%. Some uarch shifts (register aliasing, more cache) boosts IPC a little but not nearly enough to cause any of this to hold.

But as you said earlier, this discussion leads nowhere. Lets wait for Win8 launch, and see how things proceed. Intel isn't the only one bringing their guns at the show. AMDs forthcoming 28nm Wichita chips (9W) sound also very promising (their previous Zacate APU has very good GPU performance for watt).

Wichita/Krishna are rumored to have been cancelled:

http://semiaccurate.com/2011/11/15/exclusive-amd-kills-wichita-and-krishna/

(I take S|A rumors with a ton of salt but this one seems to have a lot of weight behind it, and fits awfully well)

I'm all for the market receiving more than iPad clones, but when you shift up to $1500 devices which are much larger, weigh 2-3 times more, and need active cooling, you're talking about a very different kind of device that'll sell to very few people. And that 7 hour rating becomes pretty useless in the context of gaming, there is absolutely no way that device with a 40Wh battery can achieve more than 2 hours while playing a demanding game (which is practically anything on that platform). It'll probably be under 1.5.

 

[sebbbi]

I'm all for the market receiving more than iPad clones, but when you shift up to $1500 devices which are much larger, weigh 2-3 times more, and need active cooling, you're talking about a very different kind of device that'll sell to very few people.

Samsung Series 7 Slate is 934g, iPad 2 (3G) is 610g. And Samsung has a 11.6" display while the iPad 2 display is only 9.7". In comparison the 10.1" Motorola Xoom Android tablet is 720g (it's still not quite 11.6" but getting closer). Samsung Slate weights only 29% more than the Xoom. That's pretty far from "weights 2-3 times more".

The current price point is so high (1099$), because there's very limited market for Windows 7 tablets (Win7 isn't exactly a good OS for touchscreen use). With proper mass production (and more competition in the Windows tablet segment) the price would be slashed down. The Samsung Slate also includes nice set of accessories: the Wacom pen and digitizer + a dock + a bluetooth keyboard. Make these separate purchases and the price could be lowered even more.

Also the 11" Macbook Air with identical hardware (identical CPU, GPU, memory, screen, SSD) costs slightly more than the Samsung Slate. It's not overpriced for what you get.

And that 7 hour rating becomes pretty useless in the context of gaming, there is absolutely no way that device with a 40Wh battery can achieve more than 2 hours while playing a demanding game

That's true. But tablets based on mobile phone hardware have same issues when playing highly demanding games (such as Infinity Blade).

 

[mczak]

Wichita/Krishna are rumored to have been cancelled:

http://semiaccurate.com/2011/11/15/exclusive-amd-kills-wichita-and-krishna/

(I take S|A rumors with a ton of salt but this one seems to have a lot of weight behind it, and fits awfully well)

Not sure I buy that story. Last time I heard TSMC was going to fab them, not GloFo? That said, the latest leaked roadmaps indicated indeed a delay, so if AMD really has a successor to wichita/krishna in Q4/2012 killing it might make some sense.

 

[Laurent06]

About that Samsung Slate, The Verge mentions the noisy fan. I'd never buy a tablet with a fan and I guess I'm not alone

 

[Exophase]

Not sure I buy that story. Last time I heard TSMC was going to fab them, not GloFo? That said, the latest leaked roadmaps indicated indeed a delay, so if AMD really has a successor to wichita/krishna in Q4/2012 killing it might make some sense.

The cancellation part they claim is backed by several sources, while the attribute to GF is their own speculation. It could just as well be that TSMC's 28nm isn't great for it right now either.

On the other hand, the low shelf life argument doesn't seem that airtight considering that people are claiming Trinity is right around the corner, a scant 6 months after Llano was released.

 

[Ailuros]

I though we were talking about Win8 tablets and 3DMark here, not about mobile phones. Samsung has clearly shown that a Win7/8 tablet powered by Intel's Sandy Bridge can be produced. It has a rated battery life of 7 hours, and it has been getting good reception from the press. I have said nothing about putting a Sandy Bridge (or Ivy Bridge) inside a mobile phone.

Hello upcoming 3Dmark for win8 will also be able to test smart-phone SoCs if they run on win8. Besides it doesn't change a bit if you're talking about tablets; their typical power consumption is times lower than on a netbook, notebook or even worse desktop PC SoC in a similar analogy to performance levels between different platforms.

Try the device before you claim something ridiculous like needing a battery backpack to run it.

That's what a smart-phone would need if it would consume several dozens of Watts. Re-read my former post as often as necessary to understand what I meant. I've held several tablets in my hands so far; I'm probably to spoiled by notebook/PC performance but it isn't too hard to realize that they're times behind even compared to today's lower end notebooks.

In addition to all the performance goodies (and a very nice display) it has a Wacom pressure sensor with 1024 levels of sensitivity (graphics artists love it). We have enough iPad clones in the market already. I hope that other Win8 tablet manufacturers would come up with devices as innovative as this.

An iPad or any comparable tablet in that league is still bound to a specific upper level of power consumption which is analogue to a specific level of performance you can achieve within those boundries for tablets. Tablets range between smart-phones and netbooks.

But as you said earlier, this discussion leads nowhere. Lets wait for Win8 launch, and see how things proceed. Intel isn't the only one bringing their guns at the show. AMDs forthcoming 28nm Wichita chips (9W) sound also very promising (their previous Zacate APU has very good GPU performance for watt).

It's easy to Google what kind of platforms Intel has on offer right now from smart-phones over to tablets/netbooks and after quite a distance both in terms of power consumption and performance come notebook and PC offerings.

I'll have a long hard laugh if AMD doesn't in the end serve the embedded market with their own CPUs (due to the recent cancellations) but with ARM CPU IP instead.

About that Samsung Slate, The Verge mentions the noisy fan. I'd never buy a tablet with a fan and I guess I'm not alone

That aside those devices are typically anything but cheap; I keep asking myself as a consumer that by the time I'd have to shell out anything =/>500 bucks why I should even bother with a tablet/netbook whatever and not go directly for a notebook for the same amount and times higher capabilities and efficiency instead, even more so now that cheaper tablets appear which can be used for very specific needs like a fancy e-book reader or quick browsing device for folks that travel a lot.

 

[Exophase]

Here's an indication of how the 17W SB does in gaming:

http://techreport.com/articles.x/21551/7

It gets consistently and sometimes majorly beaten by AMD's C-50 Ontario, a 9W chip. So if you want an x86 gaming tablet (and can tolerate the controls barrier) Brazos is a much better choice, and will remain a better choice vs IB, even if Wichita really was cancelled.

I'd extend that a little further: if you're buying any mobile device with gaming as top priority nothing Intel based makes sense.

 

[sebbbi]

So if you want an x86 gaming tablet (and can tolerate the controls barrier) Brazos is a much better choice, and will remain a better choice vs IB, even if Wichita really was cancelled.

AMD has some really good graphics chips in their APUs. The CPU performance however isn't yet on par with Intel offerings. Some games are more CPU dependent and run slower on (low end) AMD APUs (Llano of course beats hands down all Intel chips). That review featured only 3 games, not a big enough sample to draw any definite conclusions who's going to be on top. These Windows 7 tablets can be pretty much considered prototypes, as the real battle is still over half year away (Microsoft still hasn't released Win8 launch date) and both Intel and AMD have new generation chips scheduled for early next year launch.

There are some (confirmed) AMD Desna Z-01 APU based Windows 7 tablets coming up before the Windows 8 launch. Z-01 has just 5.9W TDP, and has pretty much identical specs to the older C-50 APUs (dual Bobcat cores at 1 GHz and HD 6250 Radeon with 80 unified shaders). Z-01 is still fabbed on 40 nm process. According to AMD, Hondo APU should replace Desna Z-01 early next year and have a slightly lower 4.5W TDP. Enhanced Bobcat cores should give the next year APUs some kind of performance boost as well, but AMD hasn't yet spilled the beans. Bobcat in general is a very good processor architecture (this is a very good reading if you are interested: www.agner.org/optimize/microarchitecture.pdf).

Here's an indication of how the 17W SB does in gaming

Portal runs at 18-33 fps at 1366x768. At lower 720p consore resolution that would be pretty much par with console 30 fps. CoD4 runs at 35.7 fps at slightly lower than console resolution. According to some Macbook Air (2011) reviews and user posts in forums (same hardware) Crysis 2 runs at 25 fps (1028x600), Counter Strike Source runs at 40-45 fps, Starcraft 2 runs at 30 fps and Starcraft 1 runs at 60-70 fps. If Ivy Bridge adds 60% to those scores, things are looking really good.

 

[Dominik D]

Slightly OT:

Samsung Series 7 Slate is 934g, iPad 2 (3G) is 610g. And Samsung has a 11.6" display while the iPad 2 display is only 9.7". In comparison the 10.1" Motorola Xoom Android tablet is 720g (it's still not quite 11.6" but getting closer). Samsung Slate weights only 29% more than the Xoom. That's pretty far from "weights 2-3 times more".

Extra size + extra weight = less mobile, unfortunately. We've tested a bunch of tablets in my (still) recent job and IMO if this is the direction of Windows-based tablets, than it is bogus. :S