The real reasons Microsoft and Sony chose AMD for consoles [F]

[Exophase]

Speaking of beyond3d posters, 3dillitante found that the profiling stats from the Killzone post-mortem pretty much confirm a 1.6GHz CPU clock:

http://forum.beyond3d.com/showthread.php?p=1734547#post1734547

I concur with this 100%, you can see other things like "time 305us, time 487,710 cycles" which equates to 487710/305 = 1599 cycles/us. Or 1.6b cycles/S, or 1.6GHz. It is however possible that this doesn't reflect release hardware.

 

[Cerb]

In a word, critical paths.

http://www.anandtech.com/show/1083/7

Bobcat never clocked much higher than its initial specs for this very same reason.

Not so. Bobcat never clocked much higher, because the clock used for CPU speed is also used for other parts of the chip, so we have never been allowed to know what the CPU could or could not do, speed wise. There was never a BE Zacate. It would be like blaming those critical paths for a non-K Core not overclocking more than 5-10%, when the actual reason is having to overclock other parts to also overclock the CPU cores.

It probably would be silly to overclock them much, since the performance would at best go up 1:1 with the clock, but power would reach the level of Richland or IB CPUs, but we have no way to know if they, in general, could do it, or not.

 

[krumme]

People bring news forward that they like no matter if it seems like nonsense. Its not the story its about. Its the persons perspective and the story ability to support that perspective.
If the news had been fx. It was clocked 1.2 GHz it would probably be ported to us by another person not so enthusiastic about ps4.

We filter our observations by our prior understanding. What doesnt fit the prior understanding we disregard as falsy. This is how we build meaning in life. More backwards so to speak. We tend to think we build meaning from observation but that is a slow process.

 

[VulgarDisplay]

People bring news forward that they like no matter if it seems like nonsense. Its not the story its about. Its the persons perspective and the story ability to support that perspective.
If the news had been fx. It was clocked 1.2 GHz it would probably be ported to us by another person not so enthusiastic about ps4.

We filter our observations by our prior understanding. What doesnt fit the prior understanding we disregard as falsy. This is how we build meaning in life. More backwards so to speak. We tend to think we build meaning from observation but that is a slow process.

Wrong forum?

 

[Arachnotronic]

People bring news forward that they like no matter if it seems like nonsense. Its not the story its about. Its the persons perspective and the story ability to support that perspective.
If the news had been fx. It was clocked 1.2 GHz it would probably be ported to us by another person not so enthusiastic about ps4.

We filter our observations by our prior understanding. What doesnt fit the prior understanding we disregard as falsy. This is how we build meaning in life. More backwards so to speak. We tend to think we build meaning from observation but that is a slow process.

Go home, krumme, you're drunk!

 

[Exophase]

Not so. Bobcat never clocked much higher, because the clock used for CPU speed is also used for other parts of the chip, so we have never been allowed to know what the CPU could or could not do, speed wise. There was never a BE Zacate. It would be like blaming those critical paths for a non-K Core not overclocking more than 5-10%, when the actual reason is having to overclock other parts to also overclock the CPU cores.

It probably would be silly to overclock them much, since the performance would at best go up 1:1 with the clock, but power would reach the level of Richland or IB CPUs, but we have no way to know if they, in general, could do it, or not.

You're right that we don't know for sure how high the CPU can clock. But what we do know is that AMD's engineers have discussed tradeoffs they've made to allow Jaguar to clock higher, tradeoffs that cost a small amount of IPC and area but were worth it in in impact on attainable clock speed. It'd be kind of dumb if AMD made these optimizations while leaving a ton of clock speed on the table.

Considering the highly synthesized design, layout decisions favoring efficiency over performance, and the low FMUL latency it's not that hard to see how the design could be limited clock to not much higher than 2GHz.

 

[Idontcare]

People bring news forward that they like no matter if it seems like nonsense. Its not the story its about. Its the persons perspective and the story ability to support that perspective.
If the news had been fx. It was clocked 1.2 GHz it would probably be ported to us by another person not so enthusiastic about ps4.

We filter our observations by our prior understanding. What doesnt fit the prior understanding we disregard as falsy. This is how we build meaning in life. More backwards so to speak. We tend to think we build meaning from observation but that is a slow process.

Wrong forum?

Go home, krumme, you're drunk!

And now we know how to take the long way to describe "confirmation bias" when we are out drinking late with our friends

 

[itsmydamnation]

You're right that we don't know for sure how high the CPU can clock. But what we do know is that AMD's engineers have discussed tradeoffs they've made to allow Jaguar to clock higher, tradeoffs that cost a small amount of IPC and area but were worth it in in impact on attainable clock speed. It'd be kind of dumb if AMD made these optimizations while leaving a ton of clock speed on the table.

performance/clock per watt doesn't equal absolute maximum clock. I think its safe to say , above 1.6-2ghz for jaguars power consumption is heading north at a very fast rate, but i wouldn't assume what the clock ceiling is.

 

[sm625]

Good god why does a game need to use 553MB of RAM for audio?

 

[insertcarehere]

Good god why does a game need to use 553MB of RAM for audio?

It's not the final game, and I would guess that a lot of this audio data is uncompressed, which happens a lot with consoles.

 

[Agent707]

I don't care. I'm not buying one.

Love my PC.

 

[krumme]

And now we know how to take the long way to describe "confirmation bias" when we are out drinking late with our friends

Some suppose i go out and drink right up until 8 in the morning ! Man i am to old for that s..t

This explanation was not a joke. But like explanations for jaguars inability to hit 2.7 can be explained or handled shortly like eg. Highly synthesized, short pipeline or fanboy, sometimes insight into the finer details can enlighten us. Some might learn a little Confirmation bias have a reason.
I explained a little of the reason from a personal psychology perspective. It applies to organazations too and explains why they often are slow to react to changes.
For more info try Karl E Weick sensemaking in organisations

Take care

 

[Exophase]

The people who want the 2.75GHz CPU claims to be true strike me as showing much more confirmation bias than those who are going with the much more likely explanation...

Here's another reason why I think 2GHz is near the limits for Jaguar: Correct me if I'm wrong here, but doesn't A6-5200 have no turbo and instead has 2GHz as its base speed? If it had more clock headroom why isn't AMD allowing even the slightest boost for when less than four, or particularly only one core is active? Should we believe that going to something like 2.1 or 2.2GHz is reliable but more than quadruples the power consumption?

It was conspicuous enough when the best Bobcats only got a very modest turbo boost, but with 4 cores instead of 2 it's even more so.

 

[pelov]

You're dismissing process tech as a possible example here. A Jaguar SoC on a low power process isn't going to clock as high as that same chip tuned for a high power process.

I'm not saying that's the reason, I'm just pointing out that it's a possibility

 

[Exophase]

Do we even know what exact process Temash/Kabini are made on? Given it goes all the way up to 25W parts, the less than amazing idle performance, and the GPU made for high performance parts I wouldn't be so sure it's made on 28LP. Maybe 28HPM, which would raise the question of how they're using their transistors.

Kind of funny that someone is telling me this now though, that's exactly what I was saying to the people who said that Cortex-A15 (which we know for sure is tuned for low leakage in Samsung's chips) wouldn't be a contender for performance. Speaking of which, now that we have much more certainly that PS4 is going to be running the Jaguars at 1.6GHz instead of 2GHz that lowers the bar for the hypothetical Cortex-A15 part from 2.5GHz to 2GHz...

 

[Sweepr]

2.75GHz refers to GDDR5 (WCK) frequency -> 176GB/s @ 256-bit which is exactly PS4's bandwidth.

 

[pelov]

Do we even know what exact process Temash/Kabini are made on? Given it goes all the way up to 25W parts, the less than amazing idle performance, and the GPU made for high performance parts I wouldn't be so sure it's made on 28LP. Maybe 28HPM, which would raise the question of how they're using their transistors.

I'm not certain, but I think it is LP? HPM would make sense as well but I'm not sure how well that would fly at the lower end. Given the big bump in TDP for the relatively meager bump in clock speeds between the lowest Temash and the highest Kabini, I'd wager LP.

Kind of funny that someone is telling me this now though, that's exactly what I was saying to the people who said that Cortex-A15 (which we know for sure is tuned for low leakage in Samsung's chips) wouldn't be a contender for performance. Speaking of which, now that we have much more certainly that PS4 is going to be running the Jaguars at 1.6GHz instead of 2GHz that lowers the bar for the hypothetical Cortex-A15 part from 2.5GHz to 2GHz...

Yea, a lot of people forget and assume it's directly attributed to the pipeline, but that's only part of the equation here.

Qualcomm has stated their Snapdragon 800 can clock up to ~2.5ghz on TSMC's HPm. That same chip on an LP process would probably be sitting around 1.5ghz? The A15's can definitely do more than the 1.5-1.7ghz we've seen given that these are most certainly on LP.

For a console maker the needs would differ drastically than that required of smartphone or tablet with respect to TDP and therefore clock speeds. If the performance is there and it's needed, meaning the chip can theoretically clock higher on HPm or HPP, then it could potentially be worth opting to go with the higher clock speeds. If it's a single die, it's going to built on HPP or HPm anyway... unless you're planning on getting a 7870 to sub-10W :biggrin:

 

[USER8000]

Why would the GDDR5 clockspeed be rated at 2.75GHZ??

It would be either too low or too high a speed?? GDDR5 clockspeed is either quoted as the actual module clockspeed or the effective QDR clockspeed.

If its the effective clockspeed it means the PS4 is using 687.5MHZ modules which seem quite slow and if it is the actual clockspeed,the PS4 GDDR5 would have effective 11GHZ(2.75GHZ X 4) clockspeed.

The highest clocked GDDR5 RAM used by an AMD GPU are 1.75GHZ modules(7GHZ effective) in the HD7970 Toxic,and I don't think commercially available 2.75GHZ modules exist as of yet.

Looking at the bandwidth figures quoted for the PS4 it is unlikely too,even if they exist.

If you look at the HD7870 it has 153.6 GB/S bandwidth:

http://hothardware.com/Reviews/AMD-Radeon-HD-7870-and-7850-GPU-Previews/?page=3

It uses 1.2GHZ modules(4.8GHZ effective QDR)

The PS4 is confirmed as having 176 GB/S bandwidth which equates to 1.375GHZ modules,or 5.5GHZ effective QDR clockspeed.

 

[Exophase]

Please, read the posts in the thread. This has been explained several times now. GDDR5 rated at 5.5GHz needs both 1.375GHz and 2.75GHz clock inputs. The highest physical external clock present in the design is all that matters here, this isn't a description of performance, it's something they need to know because they're doing tests for the device's electromagnetic interference characteristics. Which the high frequency clocks contribute to.

 

[Sweepr]

Ok, lets take a look at Sony's released (official) specifications for PlayStation 4:

Main Processor: Single chip custom processor. CPU is an x86-64 AMD "Jaguar", 8 cores. GPU is an AMD next-generation Radeon graphics engine rated at 1.84 teraflops with 18 unified Compute Units.
Memory: 8GB GDDR5 with 176GB/s bandwidth.

1.375GHz command clock, 2.75GHz write clock, 5.5Gbps -> 16 memory chips (very high density - 4 Gbit) giving 176GB/s.

 

[Exophase]

I'm not certain, but I think it is LP? HPM would make sense as well but I'm not sure how well that would fly at the lower end. Given the big bump in TDP for the relatively meager bump in clock speeds between the lowest Temash and the highest Kabini, I'd wager LP.

Check out these AMD slides, page 7 reveals a bit about the process:

http://www.hardware.fr/marc/ISSCC2013-Final-v5.pdf

TSMC's 28LP process doesn't use HKMG so it can't be that. They also say that only HVT, RVT, and LVT transistors were used. And on page 28 they say this:

Primary design optimization uses all Low Vt for speed and area.

So that says pretty much outright that their design isn't leakage optimized. And this is consistent with the idle power consumption being not very good at all - AMD themselves say > 1W for the 4 core parts. That would be mostly leakage.

Also, to those saying that it may still be able to clock much higher than 2GHz, they say outright that they achieved "> 1.85GHz", meaning 2GHz in well binned parts is probably not leaving an awful lot on the table (hoping no one chimes in to tell me that they could have meant much much greater than 1.85GHz by that statement ).

They do say there's an emphasis on density, maybe if they optimized performance at the expense of density they could have gotten more clock speed. Although from what I've heard a lot of area and timing optimizations go hand in hand, as you'd expect from decreasing average wire length between sections, although you can pretty easily think of cases where that wouldn't be true.

Qualcomm has stated their Snapdragon 800 can clock up to ~2.5ghz on TSMC's HPm.

Not to nitpick, but those cores (or Krait 300, almost the same thing anyway) are clocking fine at 1.9GHz in the 28LP Snapdragon 600 (as seen in Galaxy S4s). But yes, Qualcomm has made it very clear that changes in process allowed for higher clocks.

 

[SiliconWars]

As mentioned before, AMD already has codenames for higher performance >2GHz Jaguars - http://www.cpu-world.com/news_2013/2013062701_AMD_leaks_model_numbers_of_future_Kabini_APUs.html

A6-5250 and A6-5350, which might go up to 2.15GHz based on the numbers. That's still a long way from 2.75GHz obviously.

 

[Exophase]

I did see you mentioned that before, yet nowhere in that did I see anything about CPU clock speed. For all we know it could be nothing more than higher GPU clocks.

 

[strata8]

Most likely it's just the addition of turbo and possibly a slight clock bump.

 

[SiliconWars]

Well Bobcat eventually hit 1.75GHz with the E2-2000 (with a modest graphics increase included, but much higher graphics turbo) that's up almost 10% on release clocks of 1.6GHz. I don't think it's a huge stretch to get Jaguar to 2.15-2.2GHz, granted it would be in a much shorter time-frame assuming these higher parts release soonish.