AMD Carrizo APU Details Leaked

[el etro]

1) None of which have anything to do with ACTUAL REAL WORLD performance.

I don't care about the efficiency,die size, etc. My statement has everything to do with performance.



2) Kabini does well but its generally at half the clockspeed.
2.1) Having half the BW is generally not a problem.


3) DC won't help anything other than maybe igp.

1) Man, a Pentium 4 ipc processor, QC and clocked at 2Ghz can't handle more anything in modern days uses. Pentium DC can't even more handle WindowsXP with anti-viruses turned on....


2) Generally at half the clockspeed?
Look at same clocks: 1.91 for a8-5557M(2.1Ghz+Turbo and DC enabled) vs 1.9 for a6-5200.
2.1) "Generally": the range of Cinebench results shows that single-channel BW matters much on the score.
3) BW(And DC too) influence on the scores depends on the process type.

 

[Enigmoid]

1) Man, a Pentium 4 ipc processor, QC and clocked at 2Ghz can't handle more anything in modern days uses. Pentium DC can't even more handle WindowsXP with anti-viruses turned on....


2) Generally at half the clockspeed?
Look at same clocks: 1.91 for a8-5557M(2.1Ghz+Turbo and DC enabled) vs 1.9 for a6-5200.
2.1) "Generally": the range of Cinebench results shows that single-channel BW matters much on the score.
3) BW(And DC too) influence on the scores depends on the process type.

I don't care about clockspeed. A P4 @ 3-3.2 ghz will be better in a lot of ST tasks than a A6-5000. People keep bringing cores and clockspeed and IPC into this, THAT IS NOT RELEVANT, absolute performance is what I'm talking about on a single thread and that is not significantly greater.

I see minimal "range" on those numbers, and we are talking about mobile where you pretty much can expect to see +/- 10% due to heat, boost, mobo and statistical variations.

Put it this way, does scaling suddenly die on CB moving up the kabini SKU stack? No.

Show me where memory matters on CB. If its there is quite minimal (<5%).

 

[Arkadrel]

I don't care about clockspeed. A P4 @ 3-3.2 ghz will be better in a lot of ST tasks than a A6-5000. People keep bringing cores and clockspeed and IPC into this, THAT IS NOT RELEVANT, absolute performance is what I'm talking about on a single thread and that is not significantly greater.

I see minimal "range" on those numbers, and we are talking about mobile where you pretty much can expect to see +/- 10% due to heat, boost, mobo and statistical variations.

Put it this way, does scaling suddenly die on CB moving up the kabini SKU stack? No.

Show me where memory matters on CB. If its there is quite minimal (<5%).

In bold, I belive to be a lie.
Can you show me the benchmarks?

also why does single thread matter so much to you? when most programs now are multi threaded?

I rather have a 2.0 ghz A6-5200 than a Pentium 4 @3.2 ghz, because I believe the 2.0 ghz A6-5200 with 4 cores is MUUUUCH faster overall.

I think your forgetting how slow a pentium 4 really was.


look at

PassMark:

Intel Pentium 4 3.20GHz score = 384
AMD A6-5200 APU score = 2,435 (the A6-5200 is like x6 times as fast)


Cinebench R11.5:

Intel Pentium 4 3.20ghz (@3.6ghz overclocked) = 0,56 points
A6-5200 (@stock) = 0,50 (single thread)
A6-5200 (@stock) = 1,92 (multi thread)

(here its as fast (single thread), or x4 times as fast, when multi threaded).



Im sure if you bothered to look up other CPU benchmarks,
you would see the A6-5200 is faster than pentium 4 3.2ghz.

 

[monstercameron]

I don't care about clockspeed. A P4 @ 3-3.2 ghz will be better in a lot of ST tasks than a A6-5000. People keep bringing cores and clockspeed and IPC into this, THAT IS NOT RELEVANT, absolute performance is what I'm talking about on a single thread and that is not significantly greater.

I see minimal "range" on those numbers, and we are talking about mobile where you pretty much can expect to see +/- 10% due to heat, boost, mobo and statistical variations.

Put it this way, does scaling suddenly die on CB moving up the kabini SKU stack? No.

Show me where memory matters on CB. If its there is quite minimal (<5%).

why limit absolute performance to ST workloads only? seems a bit contrived.

also the pentium 4 at 3/3.4GHz barely beats kabini in some single threaded integer workloads

example #1
http://browser.primatelabs.com/geekbench2/compare/2437911/2423941

example #2
http://browser.primatelabs.com/geekbench3/compare/482270?baseline=478792

 

[Sweepr]

Geez, some people have really poor reading comprehension. He's not saying P4 is a better chip than Kabini in overall performance/power consumption/etcetera, just doing a simple ST performance comparison (considering IPC and clockspeed).

Intel Pentium 4 3.20ghz (@3.6ghz overclocked) = 0,56 points
A6-5200 (@stock) = 0,50 (single thread)

Aaaand you just proved the point. There's 6 of them for game devs, but that doesnt change the fact that XB1/PS4 CPU's per core performance is ridiculously low compared to any modern AMD and especially Intel desktop chip.

 

[AtenRa]

Why people always only take Cinebench to showcase a single thread performance of a CPU ???
ST performance is always different from program to program. Yes in Cinebench AMD CPUs in general and not only Kabini, have low single thread performance. That doesn't mean that Kabini at 2GHz is slower in ST than P4 in general. Run some other programs like PovRay, x264, 7zip etc and see what happens.

Also, why ST performance alone is so much important in 2014 ?? I would rather have a dual core P4 3.2GHz than any 1.5GHz Core 2 Duo single core CPU today.

ps: I still have a dual core 3.2GHz with 2GB of ram in Win 7 and it is still fine for web browsing and light office. But a Quad core 2GHz Kabini with a modern iGPU is 2-4x way better today.

 

[Enigmoid]

And my original point has already been lost.

Yes, Kabini when seen as an entire platform is pretty good.

But ST is pretty bloody low.

And that does matter for a lot of things.

ST is the weak line in the chain that is kabini. In terms of CPU progress it matches up well against a P4, showing how poorly it does in that area.

 

[Vesku]

If you could get a P4 3.4GHz quad core in a 25W TDP that is actually close to actual power draw, it wouldn't be a bad chip for small form factor. But that doesn't exist, instead we have Kabini and Baytrail.

If Carrizo can increase performance by 10+% while being more efficient on the same process as Steamroller that will be interesting. Our first sense of whether this is possible will be the cat core refresh which AMD is promising will do just that for the cat family.

 

[ElFenix]

running cinebench in ST mode is a non-real world benchmark anyway.

'ooo, i know, i'll do this client's commercial on just 1 of my many processor cores so that it takes 6x as long to finish the job' said no one ever.

 

[el etro]

Geez, some people have really poor reading comprehension. He's not saying P4 is a better chip than Kabini in overall performance/power consumption/etcetera, just doing a simple ST performance comparison (considering IPC and clockspeed).

Scaling/core(that not relies on ST only performance, but on ST and per-thread performance) counts too on Cinebench MT test.


Think the MT performance/TDP of the 8-core packed Kabini in a <50w TDP headroom. Is a thing that can't be achieved by the 84W P4(That is even OC'ed to 3.6Ghz on Cinebench test).

Aaaand you just proved the point. There's 6 of them for game devs, but that doesnt change the fact that XB1/PS4 CPU's per core performance is ridiculously low compared to any modern AMD and especially Intel desktop chip.

Steamroller ST performance is closer to jaguar ST performance. And all games can use well all of this threads. Nor the most basic OS(Explorer, Email, Forum navigation) nowdays uses so bad many threads. How the PS4/Xone games will use so bad this many threads??

 

[NTMBK]

And my original point has already been lost.

Yes, Kabini when seen as an entire platform is pretty good.

But ST is pretty bloody low.

And that does matter for a lot of things.

ST is the weak line in the chain that is kabini. In terms of CPU progress it matches up well against a P4, showing how poorly it does in that area.

And this is why all modern game engines and APIs are massively emphasising multithreaded scaling, because we need to get away from the paradigm of just hammering on one core and hoping for the best.

 

[Insomniator]

The fact that we are talking about P4's in any sense with regards to any benchmarks or limitations given, is pretty sad.

 

[Fjodor2001]

Regarding the assumed use of 28 nm instead on 20 nm for Carrizo, does it really matter that much on a desktop CPU?

Sure, the TDP will be a bit higher, but that is not so important on the desktop while were staying within reasonable limits. Also, it should be cheaper buying chips with X billion transistors on 28 nm compared to 20 nm, right? If so, it seems an obvious choice to make, until the price for 20 nm has come down (I know TSMC/GF 20 nm for big cores is not available yet, but likely it will be around the time Carrizo is released).

 

[Homeles]

Regarding the assumed use of 28 nm instead on 20 nm for Carrizo, does it really matter that much on a desktop CPU?

Sure, the TDP will be a bit higher, but that is not so important on the desktop while were staying within reasonable limits. Also, it should be cheaper buying chips with X billion transistors on 28 nm compared to 20 nm, right? If so, it seems an obvious choice to make, until the price for 20 nm has come down (I know TSMC/GF 20 nm for big cores is not available yet, but likely it will be around the time Carrizo is released).

AMD is severely lacking in performance. 20nm provides more performance. New nodes are just as big of a deal as architecture updates, if not more so. 20nm is a bit of an outlier in this regard because of the somewhat higher cost, but the performance benefits are still there.

However, 20nm at GloFo is not a high performance process. The only flavor of 20nm they offer is 20nm Low Power Mobility.

 

[Fjodor2001]

AMD is severely lacking in performance. 20nm provides more performance. New nodes are just as big of a deal as architecture updates, if not more so.

Is that really true nowadays? Do you mean it brings more performance due to node shrinks allowing for higher clock frequency? Because the latest node shrinks have not brought much of that.

 

[Vesku]

Is that really true nowadays? Do you mean it brings more performance due to node shrinks allowing for higher clock frequency? Because the latest node shrinks have not brought much of that.

Unless there are some major breakthroughs, it is not true past 32nm. Look at Sandybridge to Ivybridge, Carrizo staying on 28nm mainly means less opportunity to add more CPU cores and beef up the GPU. Given the slow roll out of DDR4, still not here for consumers, I don't think it's as bad of a blow as it would be if DDR4 was in full stride. That is if AMD's computer designed dense libraries work as they are claiming.

 

[inf64]

28nm or not, AMD WILL need 3 module parts next year, even if that means 95W SKUs and deviation from their current roadmap (that states 45/65W only).

If they could offer a part that has 3-4 modules and no GPU, on a new socket that supports DDR4 then that part could potentially be a worthy successor to FX83xx series. Hopefully Excavator core will have beefed up FP units that can execute 2x more SIMD instructions which would make it more competitive with Haswell and Broadwell. ST boost cannot hurt either, but we would have to see Excavator in action 1st before thinking it would bring anything worthy on the table (for a desktop user).

 

[mrmt]

28nm or not, AMD WILL need 3 module parts next year, even if that means 95W SKUs and deviation from their current roadmap (that states 45/65W only).

Will need =! Will be able to afford.

Plus if they were to field this SKU they would field it on the server market, not in desktops.

 

[OCGuy]

AMD's Roadmap shows FX dead for 2014. That is huge, and you have to wonder if they are giving up on the market segment all together.

 

[Enigmoid]

And this is why all modern game engines and APIs are massively emphasising multithreaded scaling, because we need to get away from the paradigm of just hammering on one core and hoping for the best.

Very true and I hope they do.

But some tasks are fundamentally serial.

It can even be difficult to break up very serial tasks in a way that scales well.

Conversion at 640 x 266

Minimal scaling over 8 threads.

3840 x 4320 conversion

Much better scaling.

 

[jpiniero]

28nm or not, AMD WILL need 3 module parts next year, even if that means 95W SKUs and deviation from their current roadmap (that states 45/65W only).

If they could offer a part that has 3-4 modules and no GPU, on a new socket that supports DDR4 then that part could potentially be a worthy successor to FX83xx series.

AMD doesn't have the money. What OEM would buy this?

 

[bullzz]

@inf64 - amd will not run the same problem it is currently with a8/a10. market segmentation
with a 3 module part, it may cost less than i7 but it might still lose in some MT benchmarks
with 4 module part, it may cost more than i7 where will compete with haswell e

 

[Homeles]

Is that really true nowadays? Do you mean it brings more performance due to node shrinks allowing for higher clock frequency? Because the latest node shrinks have not brought much of that.

I'm talking about power and price. That's all AMD's biggest customers care about.

 

[Fjodor2001]

I'm talking about power and price.

But previously you said node shrinks improved performance. So now you're shifting to it lowering power consumption and price instead, because you could not provide any evidence of what you claimed before?

Also, does it really lower price? That assumes it is cheaper for AMD to buy a chip with X billion transistors on 20 nm than on 28 nm from GF/TSMC. I don't think that is the case at the time Carrizo is expected to be released.

That's all AMD's biggest customers care about.

Power consumption is not that important on a desktop CPU/APU. On mobile it's another story though.

 

[Vesku]

But previously you said node shrinks improved performance. So now you're shifting to it lowering power consumption and price instead, because you could not provided any evidence of what you claimed before?

Also, does it really lower price? That assumes it is cheaper for AMD to buy a chip with X billion transistors on 20 nm than on 28 nm from GF/TSMC. I don't think that is the case at the time Carrizo is expected to be released.


Power consumption is not that important on a desktop CPU/APU. On mobile it's another story though.

Also the price aspect isn't true with the latest node shrinks, see 40nm->28nm GPUs and grumblings from Nvidia about wafer prices, rumblings about TSMC 20nm costs, and the sheer amount Intel has had to invest for 22nm and 14nm. Past 32nm it's been density and efficiency with the efficiency part costing a lot of time and money (Finfets).