I think he's just guessing at the lowest power usage option, since consoles usually have tight power constraints.
Last edited:
What is the clockspeed delta between the highest clocked quad-core piledriver versus the highest clocked octo-core piledriver?
jaguar is not scaling beyond 2.0 ghz. If you look at the planned chips you will notice they stop at 2.0 ghz (quad core)
10% increase in clockspeed. 4 jaguar cores at 2.0 ghz is still gonna bottleneck pretty bad. (Please note that the 4 core version at 2.0 ghz is essentially a tablet chip).
jaguar is not scaling beyond 2.0 ghz. If you look at the planned chips you will notice they stop at 2.0 ghz (quad core)
10% increase in clockspeed. 4 jaguar cores at 2.0 ghz is still gonna bottleneck pretty bad. (Please note that the 4 core version at 2.0 ghz is essentially a tablet chip).
Note that it's 10% increase in clockspeed over the unreleased 28nm variants of the Bobcat.
Unless Sony were willing to bleed cash PS3-style again (all indications said they were not), an APU based on 4 core piledriver was probably not a viable option anyway.
Isn't it kind of confusing to make a presentation slide comparing against a product they never released and AFAIK never even announced SKUs for?
Consider the source (AMD). They live in a special powerpoint-engineering world like no other. Which is why it really is good advice to not believe anything AMD puts into a powerpoint until you actually see it on a 3rd party review site.
That will depend on whether its "native" code written for one game or if a cross-platform framework/engine is used. E.g. If you use a cross-platform physics engine it might compile to code optimized for the specific platform and/or use CPU dispatching at runtime.
But of course cross-platform titles will run into the lowest common denominator thing again, which is really, really sad. The FP throughput for PS4 will be less than PS3, unless the CPU is clocked at 3 GHz. In practice the non-FMA pipes will be more effective, but it should end up a wash for physics calculation and the like. Any Haswell i3 will be more powerful for FP code which can effectively use FMA. Bah. I can remember how impressed I was with CELL, because of the (then) insane FP throughput. This time, consoles are all-crap right from the start. :|
guskline
Diamond Member
- Apr 17, 2006
- 5,338
- 476
- 126
That's the unvarnished truth. All you have to do is go back prior to the Bulldozer release and review the slides AMD was touting about the Bulldozer.:$ I think they learned a little by "toning down" the PileDriver expectations.
CELL was very hard to program if you were used to normal x86 coding. Compared to GPGPU, it shouldn't be any harder; you still have to design your code so that you have batches of data which can get processed by vector units in one go instead of wildy interspersing integer & vector code. If people didn't manage to properly use CELL, I doubt their results with GPGPU will be much better; the code design problems are pretty much the same.
And yes, I suppose the GPU is supposed to take up the slack, but of course that will drop the graphics power. If you dedicate 500 GFLOPs of the PS4s 2 TFLOPs GPU to physics etc., you reduce your graphics performance by 25%. Just for comparison, the i5-4670 will have 460 TFLOPs in its CPU part alone; that is, it will reach 20+% of the total PS4 processing power of CPU+GPU combined. Intels i5-class iGCPUs will probably outperform the PS4 by the 10nm tock if programmers would only take a little care with their PC ports.
True, but instead they've now reached new heights of "deceptive footnoting". If you look at the SR slides, the big statements are "+20-30% ops/cycle", but the tiny footnote says "in SPECrate transaction processing". But of course all the fans don't ever read the footnotes, and so SR will have 20-30% higher IPC in single-threaded loads. I don't like that much better than the BD pre-launch hype, I have to say. But at least they're truthful, technically speaking. It's progress.
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 146
- 106
True, but instead they've now reached new heights of "deceptive footnoting". If you look at the SR slides, the big statements are "+20-30% ops/cycle", but the tiny footnote says "in SPECrate transaction processing". But of course all the fans don't ever read the footnotes, and so SR will have 20-30% higher IPC in single-threaded loads. I don't like that much better than the BD pre-launch hype, I have to say. But at least they're truthful, technically speaking. It's progress.
The 20-30% is also compared to BD and not PD to make it even better. But everyone assumes PD.
So thats where the 4-6 SR APUs will match/beat FX8350 in MT-taks hype train started? Wow, if you read some people's posts you might think its something 100% confirmed. Seems like they havent learned anything from Barcelona and BD.
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
Therefore the PS4 with a octo-core cannot exist according to you. In any case, this is not answering my question. I repeat: Who said you? Ref?
He's not saying that octocore can't exist. He's saying that the Jaguar architecture doesn't scale well beyond 2.0GHz (as can be inferred from AMD's own slide on the future of Jaguar: they stop the clockspeed at 2.0GHz for two generations); therefore it is a choice between 1.6GHz octocore and 2.0GHz quadcore. I'm also reading rumors that the final product will be octocore Jaguar at 2.0GHz.
A non-Jaguar chip, as we've speculated, would use up too much power for a console. There is no hard-and-fast source for this, so you are free to doubt him if you wish.
Last edited:
Dresdenboy
Golden Member
That's correct for nearly all cases (not talking about Atom here of course) as long as there is no second thread or process or OS task running on that Intel core.
well from what I have seen when it comes to i5 vs FX 8xxx at a similar clock/high OC, even on the best cases for the FX, the heavy MT stuff the i5 is not far slower, it's quite close actually
http://www.bit-tech.net/hardware/2012/11/06/amd-fx-8350-review/3
(while using 100w less or something)
now when we move to the inevitable less than ideal situations for the FX, the difference can be huge in favor of the i5,
now comparing a BD module (2c) to a SB core, it probably can vary a lot,
http://www.xbitlabs.com/images/cpu/amd-fx-8150/t4.png
(sure it's BD and Sandra and the Intel core with HT enabled)
coming back to the next gen games and i5s, I would agree with this from the DF article I guess
"CPU power is probably the least of the concerns the PC platform has, compared to the PlayStation 4 at least. After all, the AMD Jaguar cores in the next-gen consoles were designed to compete with Intel's low-power Atom architecture, created with tablets and low-power laptops in mind. Even with eight of them, today's quad-core and octo-core desktop processors outright own them in terms of processing power."
The same slide says that the maximum number of cores is four. Now you need to answer two questions instead of one:
Why the PS4 includes an eight-core Jaguar chip if the slide says that the maximum is four?
Who said you that is was 8 core jaguar at 1.6 ghz vs 4 core jaguar at 1.6 ghz? Reference please.
The TPD of the four-core Richlands APUs @ 3.5 is of 35 W whereas the the TPD of a four-core Jaguar @ 2.0 is of 25W. Eight core jaguar will have a higher TDP...
Olikan
Platinum Member
- Sep 23, 2011
- 2,023
- 275
- 126
well, piledriver is mere 7%* faster than bulldozer on average....20-30% is great
* for some odd reason it was 13% at games
Who said you everyone is assuming? There are claims of 45% over BD and 30% over PD.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter