XBox 2 - 3.5GHz PowerPC?

[MacRumors]

TheRegister reports on a leaked report from a Chinese forum that claims to detail Xbox 2 specs.

The specs seem to include three-core 3.5GHz CPU. This is presumed to be a PowerPC/IBM design, though it is not specifically reported to be a PowerPC.

Several other sites are referencing the chip as a 976 PowerPC. As far as we can tell, all references of PowerPC 975/976 appear to be the product of unreliable/fabricated rumor reports. As a result, the 975/976 is not considered to be an actual product (...yet).

 

[Veldek]

When is the Xbox 2 said to come out (assuming that it’s not delayed as much as Longhorn)? At that time we should have 3.5GHz processors if this rumor was true.

 

[stoid]

Already a thread. But I guess this one's the 'fficial thread anyhow


I'm looking forward to seeing this thing in action!!

 

[guyute]

they must be planning on taking a large loss on everyone sold. That is a very nice chip to be able to play games on. I can't wait.

 

[dho]

If the get priority over apple for chips or fab time, im gunna be pissed

 

[Kid Red]

Damn, does a video console need that power? Will your power bill go up by playing Halo now? That thing will be huge or have a giant fan to keep it cool. I think 3.5ghz for a little gaming box is a tad extreme.

 

[BWhaler]

Well, I guess this officially kicks off the 18 months of "don't buy a PMac now. Wait until the 3.5 Ghz comes out" pieces of advice.

 

[the_mole1314]

Most systems are half of the top speed of the computers, not just for cost, but also for heat, etc. I SERIOUSLY doubt that 3.5ghz PPC are made. It sounds like a PC fan's rumor that they think is true. "HEY! If the Intel procs are nearly 4ghz, then the PPC MUST be 3.5 alteast! Those Apple people just want to screw over their customers...."

 

[dongmin]

Seems pretty far-fetched.

Three cores of 3.5 ghz CPUs???

June delivery date? Even if that's June 2005, it seems too soon for a brand spanking new processor. Or is this the fabled 970fx-successor?

 

[otter-boy]

Damn, does a video console need that power? Will your power bill go up by playing Halo now? That thing will be huge or have a giant fan to keep it cool. I think 3.5ghz for a little gaming box is a tad extreme.

If the XBox2 doesn't get released until mid to late 2005, there is no reason to believe that 3.5 GHz will be at the top of the speed curve. When the first XBox was released, its processor was somewhere in the middle to upper range of the speed spread for Intel processors. The costs will come down over time and the process will probably go through a shrink during its lifetime which will bring down costs and heat and power requirements, though rumors already state that the XBox2 chip will be produced on the 65-nanometer process, so it should be reasonable to expect that power and heat will be mangeable. Remember, we're looking for a 3 GHz chip here in just a couple months. Imagine what IBM can do with an extra year to improve chip design and production.

 

[MorganX]

they must be planning on taking a large loss on everyone sold. That is a very nice chip to be able to play games on. I can't wait.

Since it's a single CPU allegedly at 65nm customized for games, it may not cost as much as it you might think. I really don't think MS can sell the next box at anywhere near the loss as Xbox 1.

 

[stingerman]

Most systems are half of the top speed of the computers, not just for cost, but also for heat, etc. I SERIOUSLY doubt that 3.5ghz PPC are made. It sounds like a PC fan's rumor that they think is true. "HEY! If the Intel procs are nearly 4ghz, then the PPC MUST be 3.5 alteast! Those Apple people just want to screw over their customers...."

Well, interestingly a three core PPC with 1MB L2 (according to diagram) will have about the same transistors than a 1MB single core Prescot. Though it will be able to handle 6 threads (with SMT) simultaneously.

I doubt MSFT will be ready by 2005. They need to re-write the Win32 API and DirectX to PPC first, then have the games ported over to the new platform, etc. We'll probably see it in 2007 timeframe.

 

[macnews]

This would have a lot more credibility if IBM had made a 2.5Ghz chip. Considering we have yet to see anything other than a 2.0 Ghz version, I am more concerned with IBM being able to get to 3.0, let alone 3.5. Then you bring up the issue of meeting supply which they have already have had problems with in their Fishskill plant.

Of course, to keep positive, a new Xbox might arrive on the scene in October for Xmas sales. Perhaps why IBM hasn't delivered was due to problems with production but have been able to bump up speed beyond the 3.0 range. One can always hope..... (come on IBM, don't be Moto, don't be Moto...)

 

[crenz]

65nm? Everybody's having problems with 90nm already. I don't think 65nm processors will be here so soon.

 

[MorganX]

65nm? Everybody's having problems with 90nm already. I don't think 65nm processors will be here so soon.

I think IBM has produce memory at 65nm. The dedicated use of the Xbox CPU may enable them to reliably produce that chip faster than they could a desktop chip.

Additionally, Intel is moving quickly (2005/2006) to produce Tejas and other dual core CPUs at 65nm. It will be interesting to see how much volume Intel can produce for Dothan (90nm Pentium-M).

 

[tristan]

Yeah, right.

"Hello? Ibm? Yeah, I'd like to order several million 3.5 ghz CPUs, each with three cores in each chip, low power enough to fit in a console, and cheap enough so I can sell the whole unit for under $199. We'll need them in about six to twelve months."

"Sure, no problem. Good thing we have that advanced alien chip manufacturing technology from the Roswell crash site. It really comes in handy for physically impossible orders like this."

 

[eSnow]

With a little bit of tweaking, this monster should be able to emulate XBox1 games easily.

 

[MorganX]

With a little bit of tweaking, this monster should be able to emulate XBox1 games easily.

I don't know why everyone thinks this chip is "all that." A single CPU G5 can't even compete with today's Pentiums. And because the 3 cores are sharing resources, it probably won't be 3x the performance of a 1.6Ghz single G5. It will be high performance for a console, expecially running a lean NT/XP kernel, but it won't be "All that."

I'm more interested in the OS and the graphics subsystem.

 

[ffakr]

the chinese posting is crap

Here's my analytic take on the chinese posting that sparced all this...

triple core, 3.5GHz cpu, each core with vector processing.
That's a TON of horse power. If each core supported SMT as they are claiming.. we could be seeing 2-3x the performance of today's Dual G5.

What's the performance going to be used for? It's a console that's going to be hooked up to a TV. I could only imagine two uses.. playing high detail video games on HD TVs. Though, it's clear now that you can play all current games at 1600x1200 pretty well with the upcomming Nvidia video chip (with a 2GHz Athlon64). Other possibility? Real time mpeg encoding so you could record movies and TV, or video camera streams to DVD? I suppose that you'd want a ton of power to compress video, especially HD video, at anything faster than real time (for the multi-speed DVD burners) but.. if you want to do mpeg compression you'd be able to do it a lot cheaper with a DSP instead of building a triple core general purpose processor.

The diagram posted also notes a 500+MHz 16 pipe video core. That would be faster than the next gen video chipsets we're just now seeing.. like the NV40. Samples available now (clocked slower) play current games with FSAA 4x on 1600x1200 monitors with aplumb. In fact, the benchmarks I've seen with the NV40 are clearly held back by the performance of the 2+GHz Athlon64s and the 3.2+ GHz P4s that they are running in. I can see the video chipset being required for fast, steady, HD TV quality video gaming... but the cpu power proposed would be overkill since today's high end desktop chips (though slower than the video) already handle very high rez game play quite well.

Other odd point.. the diagram apparently shows 10 MB of video ram, possible attached directly to the video chip die. Though this would be fantastically fast.. it would not be nearly enough for running HD games. The vram is typically a frame buffer and a texture cache. I'd expect a game system like this, with this much power, presumably playing games on HD 1040i systems, would need a minimum of 128 MB of dedicated video memory.

I think it's a fake. but that's jmho
ffakr

 

[Kirkland]

I don't think this is so far fetched, the XBox2 isn't supposed to be released until late 2005. We're supposed to have 3Ghz chips in a few months (WWDC), whose to say next year Apple/IBM won't be at 4Ghz.

I don't think IBM just makes a chip and sits on their hands, they have 3.5Ghz, 4Ghz, 10Ghz etc. on their roadmap I'm sure.

 

[airmac]

Damn, does a video console need that power? Will your power bill go up by playing Halo now? That thing will be huge or have a giant fan to keep it cool. I think 3.5ghz for a little gaming box is a tad extreme.

Did you try FarCry?

...

 

[spud]

spintronics

maybe IBM's made more progress than they're saying on the electron-spin–based chip architecture. 6 years, 6 months, what's the difference?
http://www.nytimes.com/2004/04/26/technology/26ibm.html

 

[ffakr]

I don't know why everyone thinks this chip is "all that." A single CPU G5 can't even compete with today's Pentiums. And because the 3 cores are sharing resources, it probably won't be 3x the performance of a 1.6Ghz single G5. It will be high performance for a console, expecially running a lean NT/XP kernel, but it won't be "All that."

I'm more interested in the OS and the graphics subsystem.

That's a nice troll.
A G5 has a higher IPC (instruction per clock cycle) rate than an Athlon64. It has a much higher IPC than a Pentium 4.
OTOH, an Athlon64 has a much lower memory latency due to the on board memory controller, and it is a powerchip otherwise.
...

There are benchmarks where an Athlon will school other processors, as there are benchmarks with a P4 will school other processors.. and yes, there are benchmarks where a PPC 970 will school other processors (even if they clock at over 3GHz)
.. not that it's easy to benchmark cpus running different OSes and different instruction sets... Athlons used to be monster processors compared to PPCs and x86 cpus.. but they had a small instruction set. If you picked a benchmark with an operation that wasn't easily performed with the Alpha ISA, it benched very poorly because you needed to have an algorithm to work around that lack of functionality. Look at old RC5 benchmarks.

Only an idiot thinks that their pet processor is better than the competition in every respect. Are you an idiot?

The processor in question is presumed to be a triple core CPU based off the Power5 family. the PPC 970 is based off the Power4 family. the Power5 based processors will be more powerful per cycle, they will include IBMs implemention of SMT (simultaneous multi threading, aka. Hyperthreading)... IBM is saying that their SMT is much more efficient than Intel's. IBM has said that SMT on the Power5 is providing a 25-40% increase performance already (in design still) while the P4 HT generally provides maybe +15% to -10% performance boost.

The current PPC 970 core performs very favorably to current breed of CPUs (even with the crappy gcc code, we see 35% performance boosts when using IBM's xl compilers). The cpus that are 'claimed' to be in design will be more efficient, support SMT, and run 3 cores at nearly double the speed of today's PPC 970 chips.

That is why they are 'all that'.
None of the other desktop CPU vendors have anything on the roadmap for the next two years that would even marginally compete. Try to imagine a triple core P4 running each core at about 4.5GHz.. with a redesigned SMT.. that's what we are talking about. Unfortunately such a chip [that imaginary P4] would likely dissipate about 300 watts.

 

[ffakr]

Did you try FarCry?

...

yea, it runs nicely on my 2GHz AthlonXP with my Ti4200. With a Radeon 9800 and a similar CPU, my brother is able to max out every single setting and it still runs quite nicely.
The silly diagram that was posted in China has video subsystem that would be nearly 4x as powerful as a Radeon 9800 and a CPU that is probably half an order of magnitude more powerful.

 

[Frobozz]

65nm? Everybody's having problems with 90nm already. I don't think 65nm processors will be here so soon.

The problem is not with the 90nm process, it with the Strained Silicon on Insulator, which delaminated at high clock frequencies due to heat and production problems. At least, that's what the reports say. The transition to 90nm went without issue.