ITT: We discuss processors for Steam & whether Westmere 2C/4T should be resurrected?

[cbn]

The reason I made this thread is because I am concerned about the future of Intel in the living room space. In this particular thread I will write in the context of SteamOS and Steam machines, but this could easily apply to other alternative Linux OSes as well (at some other point)

Why am I concerned?

I am concerned because while Steam is x86 gaming platform, the steam machine itself does not necessarily need to be x86 in order to stream x86 Windows games from a local network.

This, of course, opens to the door for alternate processor uarchs to make their way into the Steam Linux ecosystem slowly, but surely.

In fact, Valve has already ported some games to ARM already for Android---> http://shield.nvidia.com/android-game-portal-halflife2/

Therefore the potential exists for a Steam on alternative processor uarch (like ARM) provided the alternative hardware proves to be high enough value over what x86 has to offer.

So lets assume Intel wants to launch a very aggressive plan for cutting off any ARM processor competitors, what would the best way to do that be?

One idea that crossed my mind was to relaunch the Westmere 2C/4T with a new on package graphics chip/memory controller. This to compete in the ultra low cost desktop/console for living room category.

I'd imagine such a chip would be far superior to anything atom based or ARM based (in terms of processing power) for quite some time.

Nice thing about this is that I believe the 81mm2 32nm Westmere 2c/4T die could be used as is, with only a new graphics/memory chip needing to be created.

Maybe Intel could sell of the 32nm Westmere 2C/4T "as is" to interested parties (eg, Rockchip) and let these other parties design their on package graphics (using licensable IP from Nvidia, Imagination Tech, or even Intel) . This would help keep the Intel 32nm fabs filled beyond just making PCHs while solidifying Intel's position at the lowest end of x86 desktop.

Or it could be that Intel also launches their own version of this chip and uses the partnerships (Rockchip, etc) as way of lowering their own GPU/memory controller development costs?

Now some might be concerned this recycled x86 2C/4T would eat into or cannibalize existing x86 sales, but I think it would actually be the reverse. Instead of being subtractive, I believe the sales would be additive. Folks would still buy their office PCs and high powered Windows gaming rigs and add this smaller more affordable chip to their living room space in the same way they would add an ARM powered console chip.

So with all that mentioned and hopefully clear enough, lets move onto discussing the Westmere 2C/4T chip I mentioned in this thread---> http://forums.anandtech.com/showthread.php?t=2397599

The original 2C/4T Westmere was known as Clarkdale on the desktop. It was comprised of a 81mm2 32nm 2C/4T chip with a much larger 45nm Intel GMA graphics chip with memory controller on package:

One major flaw described by Anand was the moving of the memory controller off the processor die and onto the graphics chip:

http://www.anandtech.com/show/2901/2

Memory Performance - Not Very Nehalem

Let’s start at the obvious place, memory performance. Nehalem moved the memory controller on-die, but Clarkdale pushes it off again and over to an on-package 45nm graphics core.

To make matters worse, the on-package chipset is a derivative of the P45 lineage. It’s optimized for FSB architectures, not the QPI that connects the chipset to Clarkdale. Let’s look at the numbers first:

Processor L1 Latency L2 Latency L3 Latency
Intel Core i7-975 4 clocks 10 clocks 34 clocks
Intel Core i5-750 4 clocks 10 clocks 34 clocks
Intel Core i5-661 4 clocks 10 clocks 39 clocks
AMD Phenom II X4 965 3 clocks 15 clocks 57 clocks
Intel Core 2 Duo E8600 3 clocks 15 clocks



L1 and L2 cache latency is unchanged. Nehalem uses a 4-cycle L1 and a 10-cycle L2, and that’s exactly what we get with Clarkdale. L3 cache is a bit slower than the Core i7 975, which makes sense because the Core i5 661 has a lower un-core clock (2.40GHz vs. 2.66GHz for the high end Core i7s) Intel says that all Clarkdale Core i5s use the same 2.40GHz uncore clock, while the i3s run it at 2.13GHz and the Clarkdale Pentiums run it at 2.0GHz.

Processor Memory Latency Read Bandwidth Write Bandwidth Copy Bandwidth
Intel Core i7-975 45.5 ns 14379 MB/s 15424 MB/s 16291 MB/s
Intel Core i5-750 51.5 ns 15559 MB/s 12432 MB/s 15200 MB/s
Intel Core i5-661 76.4 ns 9796 MB/s 7599 MB/s 9354 MB/s
AMD Phenom II X4 965 52.3 ns 8425 MB/s 6811 MB/s 10145 MB/s
Intel Core 2 Duo E8600 68.6 ns 7975 MB/s 7062 MB/s 7291 MB/s


Here’s where things get disgusting. Memory latency is about 76% higher than on Lynnfield. That’s just abysmal. It’s also reflected in the memory bandwidth scores. While Lynnfield can manage over 15GB/s from its dual-channel memory controller, Clarkdale can’t break 10. Granted this is higher than the Core 2 platforms, but it’s not great.

What we’re looking at is a Nehalem-like CPU architecture coupled with a 45nm P45 chipset on-package. And it doesn’t look very good. If anything was going to hurt Clarkdale’s performance, it’d be memory latency

With that mentioned, maybe the fact the memory controller is on the graphics is not such a bad thing if the plan is to recycle 2C/4T Westmere for the future? Would it even be possible for purchasers of 2C/4T to design in a new DDR4 memory controller with whatever graphics IP they license for the on package chip?

Some other concerns I have about Westmere involve it being an old platform with features limited to usb 2.0, SATA 3 Gbps, PCIe 2.0, DDR3, etc. However, with that mentioned I am not sure how much of a detriment that is for a low cost platform? Or even how much of that could be updated to a new spec if necessary?

Please post your ideas, criticisms and thoughts.

If I am off my a country mile with my ideas (in your opinion) I want you to tell me. And if you have a better plan than me I want you to let us know about that as well.

 

[Ken g6]

Westmere? Westmere?! I see no reason for Intel to look backward for an ARM competitor.

This would help keep the Intel 32nm fabs filled beyond just making PCHs....

Intel 32nm fabs, once they're done making PCHs, are probably being retooled to do 10nm or 7nm. No, if Intel wants a new chip, it's going to be 16nm, or maybe 22.

If you're worried about die size, remember that Broadwell-Y is just 82mm^2. And that's with a nice, big IGP that probably takes 1/3 of the chip! Chop that off and it gets smaller still.

 

[Techhog]

I... um... What?

 

[VirtualLarry]

Westmere? Westmere?! I see no reason for Intel to look backward for an ARM competitor.

Intel 32nm fabs, once they're done making PCHs, are probably being retooled to do 10nm or 7nm. No, if Intel wants a new chip, it's going to be 16nm, or maybe 22.

This. Don't forget, one of the primary drivers behind shrinking chips is not just performance or power-consumption, but cost. Surely, Intel would not be willing to take a step backwards there, especially to fight low-cost ARM.

Edit: What about a 2C/4T Broadwell or Skylake, with GT2 or GT3? On 14nm, the cost probably wouldn't be too high. I could see such a chip being popular with Steamboxes.

 

[frozentundra123456]

Yea, dont get it either. Why go back to 32nm? Even if you port it out, I am not sure the volume would be enough to make a special chip for it. Doesn't atom already run Android? Maybe they could use atom in android steam boxes.

Personally, I would not be interested in an android steam box, and really am not interested in streaming either, but they are valid concerns for intel I guess. Although I am not sure if steam boxes per se will ever be a significant market compared to overall PC sales. Both steam OS and steam boxes themselves have gotten of to an incredibly slow start.

 

[cbn]

If you're worried about die size, remember that Broadwell-Y is just 82mm^2. And that's with a nice, big IGP that probably takes 1/3 of the chip! Chop that off and it gets smaller still.

But how much will Intel sell that Broadwell for?

I'll bet they will ask too much to be competitive at the level of desktop chips I am thinking about.

 

[cbn]

Doesn't atom already run Android? Maybe they could use atom in android steam boxes.

Personally, I would not be interested in an android steam box, and really am not interested in streaming either, but they are valid concerns for intel I guess.

No not Android.....SteamOS.

 

[Arachnotronic]

Braswell/Cherry Trail would be a better chip for this purpose than a resurrected Westmere

 

[cbn]

Although I am not sure if steam boxes per se will ever be a significant market compared to overall PC sales. Both steam OS and steam boxes themselves have gotten of to an incredibly slow start.

One problem I have with the idea of Steam Box as it currently stands is that the x86 hardware is just too expensive to justify limiteing the usage to just SteamOS.

Why does someone need to worry about the cost of a $100 Windows OS if they are spending $400, $500, $600 or more on hardware.

But lets say the hardware prices drop to a very low level, maybe even $100 or less. Now that $100 Windows is a rather large cost adder proportionally.

 

[cbn]

Braswell/Cherry Trail would be a better chip for this purpose than a resurrected Westmere

It just depends on how much the Westmere 2C/4T goes for.

P.S. One nice thing about Westmere is that Intel pretty much doesn't have to worry about a third party using it to make a laptop. (A laptop is definitely something that could eat into Intel's current chip sales, whereas I don't think this would be the case with living room x86 console/desktop).

 

[Ken g6]

But how much will Intel sell that Broadwell for?

Well, they're probably binning the Broadwells like crazy to get down to 5W. Also, some chips will have working CPUs, but not GPUs. Both kinds of cast-offs would be great for Steam consoles with discrete graphics.

And this would be going up against, what, 2GHz quad-core ARMs? Then the other thing they could do is make a new Broadwell die where they chop off the iGPU, cut it down to 1C2T, and clock it to 4GHz.

 

[frozentundra123456]

No not Android.....SteamOS.

Same goes for Steam OS. If I am going to spend the money for the hardware, I want Windows. Perhaps if you could dual boot Steam OS for games and linux for general use, that would be OK. Still though, can we be assured that all games will support Steam OS? I would bet that games sold only through Origin and UPlay will not support Steam OS. And what about older games? Are all of them going to be ported back to Steam OS?

In any case, in relation to the OP, as I said, I think atom or low end haswell (or something like Kabini) would make a lot more sense for a low end streaming type steambox, rather than some resurrected 32nm cpu.

 

[cbn]

Same goes for Steam OS. If I am going to spend the money for the hardware, I want Windows.

Yes, that is a problem folks are going to face if they spend too much money on x86 hardware.

Why invest so much in expensive hardware and then massively limit capability by restricting oneself to SteamOS?

And why dual boot SteamOS and Windows? If a person has already bought windows, why not just use Windows and operate through the steam client?

I think for SteamOS to make sense the x86 hardware needs to be cheap. This way a person doesn't tie up a large investment for such a limited usage.

 

[cbn]

Frozentundra123456,

Regarding dual booting SteamOS and Linux or perhaps installing the steam client on Linux, I think the advantage of using SteamOS by itself would probably factor in strongest when extremely limited storage is being used. Think something like an ARM or Intel HDMI stick or set-top box with 8 or 16 GB of eMMC.

 

[NTMBK]

To be honest, if you want an affordable x86 console chip, you're looking at the wrong company... AMD is the one with a successful track record.

I'm still hoping for them to release a console-style big APU with soldered down, high speed memory which runs Windows. Maybe with Puma+ cores on 20nm- hopefully they will be able to increase the clock speeds on the new process node.

 

[cbn]

Well, they're probably binning the Broadwells like crazy to get down to 5W. Also, some chips will have working CPUs, but not GPUs. Both kinds of cast-offs would be great for Steam consoles with discrete graphics.

Then the other thing they could do is make a new Broadwell die where they chop off the iGPU, cut it down to 1C2T, and clock it to 4GHz.

I'd hope the bargain Intel Broadwell chips would not need a discrete GPU. A video card is going to be a large cost adder.

In fact, with this plan I am mentioning, I am hoping just the opposite happens for Westmere 2C/4T. Instead of buying discrete graphics there would be many companies integrating their own graphics and memory contoller for the on-package chip.

 

[witeken]

I don't think Intel has anything to fear from ARM in anything above tablets, certainly not from the Steam Machine niche. Intel's current roadmap should be good enough.

 

[dave_the_nerd]

So... an i3 with Iris Pro graphics?

I'm sure Broadwell will have something in that bracket.

 

[cbn]

Another advantage to selling the Westemere 2C/4T and letting others design their own on package chip is that it could potentially bring in new graphics IP (Nvidia, Imagination Tech, AMD?) into Intel's fabs.

Though with that mentioned, I am not sure the on package chip would necessarily need to be fabbed at Intel, but maybe Intel would want (or insist) it to be if they had a large hand in designing the new memory controller?

 

[cbn]

And this would be going up against, what, 2GHz quad-core ARMs?

I think this would depend on at least two things:

1.) What is SteamOS able to offer over Android?

What perks does a SteamOS user have that the Android user doesn't have? This, will in part, determine the type of ARM hardware we could see on SteamOS.

2.) The timeframe for launching Westmere + third party on package graphics. Are we talking 2016 or later or could this be done sooner?

 

[MiddleOfTheRoad]

One problem I have with the idea of Steam Box as it currently stands is that the x86 hardware is just too expensive to justify limiteing the usage to just SteamOS.

Why does someone need to worry about the cost of a $100 Windows OS if they are spending $400, $500, $600 or more on hardware.

But lets say the hardware prices drop to a very low level, maybe even $100 or less. Now that $100 Windows is a rather large cost adder proportionally.

You can bolt together an entry level setup for around $200-250 if you use an APU to get started (and skipping the dedicated video card initially). That's the beauty of a PC -- you can start with a very basic system and gradually add parts to get it where you want it. An A4 can do some light gaming with the built-in Radeon, although I'd recommend the A8 if the budget permits.

SteamOS or Ubuntu (can also run Steam) are both free and are good ways to keep the build cheap.

https://www.youtube.com/watch?v=sSYTMhDoMS8

 

[cbn]

You can bolt together an entry level setup for around $200-250 if you use an APU to get started (and skipping the dedicated video card initially). That's the beauty of a PC -- you can start with a very basic system and gradually add parts to get it where you want it. An A4 can do some light gaming with the built-in Radeon, although I'd recommend the A8 if the budget permits.

SteamOS or Ubuntu (can also run Steam) are both free and are good ways to keep the build cheap.

https://www.youtube.com/watch?v=sSYTMhDoMS8

I priced that build out at $226.

Yes, I have used the steam client before with Ubuntu. It works great and the AAA titles have been increasing over time to include games like Civilization V, Metro Last Light, Witcher 2, etc (This in addition to the Valve and indie games, of course). Borderlands 2 is coming soon as well--> http://www.phoronix.com/scan.php?page=news_item&px=MTc3OTA

With that mentioned, one thing though we need to watch for with AMD is their Proprietary Linux OPEN GL driver performance.
Example: http://www.phoronix.com/scan.php?page=article&item=nvamd_binary_comparison&num=8

The NVIDIA proprietary Linux graphics driver was consistently delivering better performance. In some tests the Radeon R9 290 Hawaii graphics card on Catalyst was running around the performance level of the GeForce GTX 770 (a ~$70+ cheaper graphics card) and in the worse case around the speed of a GTX 760

Outside of these OpenGL test results, the NVIDIA proprietary drivers tend to have better maintained legacy drivers, the NVIDIA drivers more quickly support new X.Org Server and Linux kernel releases, and generally the consensus among both Linux gamers and developers is that there's far less bugs with the NVIDIA driver than AMD Catalyst.

The AMD Catalyst driver though has been improving quite a bit in recent months given all the commercial game studios now pushing Linux games, but it's still not to match NVIDIA's Linux dominance when it comes to proprietary drivers.

With that mentioned, AMD does have the cross platform Mantle API and OPEN GL-NG might be borrowing a good deal from Mantle as well: http://techreport.com/news/26922/amd-hopes-to-put-a-little-mantle-in-opengl-next

Huddy told us AMD has done a "great deal of work" with the Khronos Group, the stewards of the OpenGL spec, on OpenGL Next. AMD has given the organization unfettered access to Mantle and told them, in so many words, "This is how we do it. If you want to take the same approach, go ahead." Khronos is free to take as many pages as it wants out of the Mantle playbook, and AMD will impose no restrictions, nor will it charge any licensing fees.

So maybe we finally see this driver discrepancy on Linux disappear eventually? (I would just love it if one day Linux games played on AMD hardware had the same FPS as the same titles played on Windows. Otherwise some of the savings from Linux could be canceled out by the higher hardware spec need to play the game acceptably)

 

[cbn]

Intel 32nm fabs, once they're done making PCHs, are probably being retooled to do 10nm or 7nm. No, if Intel wants a new chip, it's going to be 16nm, or maybe 22.

Assuming the project was successful, Intel could always shrink down Westmere to a smaller node (at some later point in time).

The idea would be to allow some technological progress while keeping R & D expenses to a minimum so the platform could have maximum bang for buck.

The amazing thing about Westmere IMO is that while it is an older Intel big core I still think it would still be a good deal ahead of what ARM could field on the desktop in the near future. This while still being modern enough for its intended ultra budget price category.

(Though with this mentioned we do have to consider what Apple could do with Cyclone or a successor to Cyclone if the company were to juggle form factors and increase clocks. Maybe a new form of Apple TV that is more like a Mac Mini would warrant a higher clocked Cyclone cored processor than what we see in phones and tablets?)

 

[Vesku]

If Intel dedicated as much die space as they've hinted for 14nm Broadwell iGPU then a Broadwell i3 will be a decent Steambox APU. Seems that the main suggestion of using a 32nm Westmere 2C/4T paired with a dedicated GPU or large iGPU is price, but that's making a big assumption about Intel's Fab plans and costs.

 

[TuxDave]

Assuming the project was successful, Intel could always shrink down Westmere to a smaller node (at some later point in time).

I think once you open that door, you're better off using fuses on a current gen processor. The amount of design resources to build another core derivative (even if it's logically the same as a previous gen) is enormous.