Richland ES now out in the wild.

Page 7 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Idontcare

Elite Member
Oct 10, 1999
21,126
0
0
Done

The Kabini ES looks pretty intriguing. It appears to be at 1.4 Ghz 1.7 Ghz turbo based on the OPN. So 4C 17W and it matches the low end 1.6Ghz 800mhz FSB Core 2 in single-thread in Cinebench R10 while matching a 2.6 Ghz Conroe Core 2 Duo, and a 2.8 Ghz Barcelona based Athlon II.
Thanks :thumbsup:

Assuming this is the real deal, and I do think it is reasonable to assume as much, do we know what stepping we are looking at here? Power numbers might improve if this is not a retail stepping.
 

Zstream

Diamond Member
Oct 24, 2005
3,176
7
116
So, whats the performance appear to be like?
 
Aug 10, 2012
68
0
71

Zstream

Diamond Member
Oct 24, 2005
3,176
7
116
The E-350 (1.6GHz) clocked in at 1171 with a single thread and 2250 with multiple threads.

http://www.anandtech.com/show/4023/the-brazos-performance-preview-amd-e350-benchmarked/3
An FX 8350 gets around 4300 in Single Threaded and 23500 in multi-threaded performance. Let's just say that IPC will increase at a 100% with MHZ increase.

Single Thread:
  • Richland chip running in turbo mode at 1700MHZ = 1575 (rounding) - .93 points per mhz
  • 8350 chip running in turbo mode at 4200MHZ = 4300 (rounding) - 1.03 points per mhz

Multi Thread:
  • 4 CORE
  • Richland chip running at 1400MHZ = 5655 (rounding) - 4.03 points per mhz
  • 8 CORE
  • 8350 chip running at 4000MHZ = 23500 (rounding) - 5.875 points per mhz

If AMD kept the definition of cores in place, we can maybe assume it's about equal to Piledriver in single thread performance while having a beefed up GPU core. If the same definition is in place for core (which I doubt) count, this thing will be a nice multi-threaded monster....
 
Last edited:

wlee15

Senior member
Jan 7, 2009
295
0
76
Thanks :thumbsup:

Assuming this is the real deal, and I do think it is reasonable to assume as much, do we know what stepping we are looking at here? Power numbers might improve if this is not a retail stepping.
Unfortunately no although if anyone could figure it out it probably Dresdenboy/Matthias Waldhauer.

An FX 8350 gets around 4300 in Single Threaded and 23500 in multi-threaded performance. Let's just say that IPC will increase at a 100% with MHZ increase.

Single Thread:
  • Richland chip running in turbo mode at 1700MHZ = 1575 (rounding) - .93 points per mhz
  • 8350 chip running in turbo mode at 4200MHZ = 4300 (rounding) - 1.03 points per mhz

Multi Thread:
  • 4 CORE
  • Richland chip running at 1400MHZ = 5655 (rounding) - 4.03 points per mhz
  • 8 CORE
  • 8350 chip running at 4000MHZ = 23500 (rounding) - 5.875 points per mhz

If AMD kept the definition of cores in place, we can maybe assume it's about equal to Piledriver in single thread performance while having a beefed up GPU core. If the same definition is in place for core (which I doubt) count, this thing will be a nice multi-threaded monster....
Bear in mind this is Kabini which uses the Jaguar core and is a replacement to the Bobcat based APUs.
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,367
156
126
I wouldn't see how Steamroller wouldn't perform similarly to Jaguar when one Bulldozer core performs the same as one Bobcat core.
 
Apr 21, 2012
151
0
81
BDv1 @ 1.7ghz - 4 cores

Single core : 1720

Multi-core : 5376
 

mrmt

Diamond Member
Aug 18, 2012
3,976
0
76
And Finally


If Richland really can stay at 3.2 Ghz at 35W TDP loaded then that is quite a bit of and improvement over Trinity.
It's just me or did anyone else find strange a 50% drop in TDP for the same frequency coming from a half-node SOI-less shrink?
 

inf64

Platinum Member
Mar 11, 2011
2,868
170
136
I wouldn't see how Steamroller wouldn't perform similarly to Jaguar when one Bulldozer core performs the same as one Bobcat core.
As seen above in one of the posts, in C10 benchmark, PD core is more efficient in ST task than Jaguar core (per core and per clock). If SR core improves as much as AMD claims then it might match MT performance of Kabini per core and per clock (PD is ~37% slower with MT "hit").

It's just me or did anyone else find strange a 50% drop in TDP for the same frequency coming from a half-node SOI-less shrink?
Do we know for sure Richland is 28nm bulk?

Also a good question would be: what is Vishera FX 2.0 if Richland is indeed done @ 28nm bulk node? Is it possible that AMD may introduce a "new PD" core for FX which is essentially PD done on 28nm bulk node? Refreshing the Opteron line in the process also? They sure need the power draw reduction and seeing how Richland achieves ~40% higher clock within same TDP as 32nm Piledriver for mobile , it almost looks like PD was meant to be produced on 28nm instead on leaky 32nm node.
Imagine 4.4Ghz/4.6GHz 95W 28nm Vishera FX chip. Who needs SR in 2013 :D.
 
Last edited:

mrmt

Diamond Member
Aug 18, 2012
3,976
0
76
As seen above in one of the posts, in C10 benchmark, PD core is more efficient in both ST and MT than Jaguar core (per core and per clock).
By more efficient do you mean better IPC or better performance per watt?
 

inf64

Platinum Member
Mar 11, 2011
2,868
170
136
By more efficient do you mean better IPC or better performance per watt?
I meant pure performance per MHz. But it does exhibit the MT performance scaling hit which negates that. Note that this perf. hit seen in C10 is rather high when compared to some other workloads.
Looking at perf./watt Kabini is probably much better than Vishera since it has ~10% lower performance(same clock) while consuming much less power. But since they are targeted at different segments they are not directly comparable in that metric. All in all Kabini has some impressive performance and perf./watt for a "small core" chip.

edit: Looking at 19W 4C Trinity model,it scores (roughly) 14% lower in C10 MT benchmark at the same 1.4Ghz clock(stock is 1.6Ghz). So 2W more and 15% lower MT performance than Kabini. All this is true of course if Kabini's numbers are from 64bit version of the benchmark. If they are from 32bit version than Kabini is even faster.
 
Last edited:

NTMBK

Diamond Member
Nov 14, 2011
8,300
280
126
Hmm, I just hope the chip isn't *too* crippled by memory bandwidth. Last I heard, Kabini only had a single memory channel, and if OEMs pair that with 1066 memory it's really going to suffer.
 

mrmt

Diamond Member
Aug 18, 2012
3,976
0
76
Looking at perf./watt Kabini is probably much better than Vishera since it has ~10% lower performance(same clock) while consuming much less power. But since they are targeted at different segments they are not directly comparable in that metric. All in all Kabini has some impressive performance and perf./watt for a "small core" chip
Thanks for the data. Well, I think that this explains why AMD is probably ditching big cores.

Not only Kabini was better performance per watt it also has better IPC, and all that in a far smaller package, and more important, manufactured out of GLF. That's much better economics than big core currently achieves. They will be better trying to scale Kabini up than trying scale Trinity down.

Brazos already sells more than Bulldozer in desktops anyway. A better performance chip build around the same premises should sell even more. Even if they are going to feel the heat from 10W Haswell Celeron and Airmont/Silvermont, this will happen only in 2014, giving them a nice respite.
 

inf64

Platinum Member
Mar 11, 2011
2,868
170
136
The problem with scaling Kabini up is that Jaguar is probably never going to run at 3.5+Ghz with 8 cores and stay in reasonable power envelope(say 65-95W). On the other hand 28nm Piledriver could do this without much problems. This is a pipeline design tradeoff. But since the IPC is always improving bit by bit with every generation of "big core" such as Bulldozer,we have come to a point when we will have SR core ,a high clocking machine, which also has (relatively) high performance.

If we were to compare Kaveri and Kabini at the similar TDP ranges ,I'm pretty sure Kaveri would come out on top in all metrics expect perf./mm^2. There it would be no comparison since Jaguar core is truly tiny. But I think that pure performance,performance/Mhz and maybe even perf./watt would be better on Kaveri( say 1.4Ghz 17W QC/2M Kaveri with very high Turbo core vs 1.4Ghz QC Kabini with 1.7Ghz Turbo).
 

mrmt

Diamond Member
Aug 18, 2012
3,976
0
76
The problem with scaling Kabini up is that Jaguar is probably never going to run at 3.5+Ghz with 8 cores and stay in reasonable power envelope(say 65-95W). On the other hand 28nm Piledriver could do this without much problems. This is a pipeline design tradeoff. But since the IPC is always improving bit by bit with every generation of "big core" such as Bulldozer,we have come to a point when we will have SR core ,a high clocking machine, which also has (relatively) high performance
I agree with that, but once you look at AMD product mix you'll see that 65W-95W are a shrinking part of the market. AMD biggest market is mobile and there Brazos is king. With 65W and Bulldozer performance levels you are pretty much limited to very big notebooks, which haven't big margins, and above that it's a no-go, you can only go with desktops there.

On desktops, situation is a bit better for power hogs, but not that much better. Over 60% of AMD sales are comprised of 65W and smaller TDP chips, and Brazos is on its way to becoming the most important, or the second most important desktop product out there. Brazos is on the verge of outselling Trinity.

In the end, regardless of my opinion about Brazos/Kabini performance levels, I see in them a much better marketing potential than the entire big core line up.

Brazos and Kabini are the perfect products for AMD. cheap, small, efficient, and capable to offer competition to Intel but on its own terms. It is where they can offer a significant trade off, less performance for a significant smaller price, forcing Intel to eat into margins to field Celeron IVB/Haswell there, at least until 2014 when the "monts" arrive.

OTOH big core will be troubled when Haswell arrives. If Haswell does not bridge the IGP graphics it will be close to, it also brings IPC and power consumption improvements that will put pressure on AMD prices. There isn't significant trade off there, with big core their offer competition to Intel on Intel terms, and they are too far behind the technology curve for that.

As for 28nm, I'm kinda skeptical about it. It's a half node shrink, which means nothing too significant to power consumption and die size. Also it is a SOI-less node, which means more leackage. I don't see huge clock increases here.
 

inf64

Platinum Member
Mar 11, 2011
2,868
170
136
I agree with your market analysis :). But AMD won't abandon the "big core" yet,maybe after 2015 they will reevaluate their startegy. I see intel is looking at the similar direction (data from the other thread, if true). The big cores are going away which is unfortunate for desktop users. The IPC benefits are diminishing since the x86 ISA is not very good in that regard. Also, the "more cores" approach is failing to yield more performance since workloads are not getting optimized (or cannot be optimized due to mostly serial nature of their code). There is a shift towards heterogeneous computing where workloads will be offloaded to special cores or dedicated hardware( look at intel's quicksync,a perfect example) and a few "fat cores" will do the serial jobs.

Sure,server segment can't get enough of "cores" but as we can see,intel is doing their xeon phi thing since it's just much better at specific workload scenarios and provides easy programming model. AMD is going for ARM+x86+GPU/HSA and we have to wait and see how that works out for them. In a few years the desktop segment will be mostly "APU" and a few server parts re-branded to fill in the "highest end" (like s2011 now).

As for 28nm,if the x264 dev really got a 28nm bulk 3.2Ghz 35W part then there is big increase in clocks versus Trinity. But that's a big IF. He probably either got 3.2Ghz part but it's not 35W rated or he got a 35W part with 3.2Ghz Turbo core clock.
 
Last edited:

mrmt

Diamond Member
Aug 18, 2012
3,976
0
76
I agree with your market analysis :). But AMD won't abandon the "big core" yet,maybe after 2015 they will reevaluate their startegy.
I think it's a fair question to ask how much cash does big core brings to AMD and how much cash will it bring after Haswell. If you look at their current numbers they are losing market exactly on big core, what do you think that will happen after Haswell if we still get a depressed PC market in 2013?

Right now AMD is on operating losses and it isn't the GPU unit making it, but the CPU. Given the shrinking big core numbers and the growing Brazos number, you can bet your money that losses are being caused by big core bad numbers. One reason for the cuts is to adapt big core cost structure to this tough reality. We also know that AMD isn't really happy with GLF, to the point it is paying to get out of the relationship.

So once you add Haswell, a depressed PC market and the significant amount of cash AMD will need to develop their ARM line up, where do you think managers will look for cuts? In Brazos/Kabini, a growing share product manufactured in a proven foundry, and a product they have a lot of price latitude, or in big core, where they have a rocky relationship with the foundry partner, shrinking numbers and bad margins?

I had exactly this idea of yours until reviewing AMD 2012 numbers. They need a very fast turnaround. This goes along the same lines of recent Rory statements, where he said he thought he had years to turn AMD around, he has to go much faster now. And once you are cash strapped, your choices are cut and/or sell.
 

NTMBK

Diamond Member
Nov 14, 2011
8,300
280
126
Also, don't forget that the current Brazos die is tiny. Even on the 45nm process node, it is only 75mm^2, whereas Trinity on 32nm is 246mm^2. Given that Jaguar will be moving to a significantly smaller node than its predecessor (28nm, iirc?), even with the extra two cores, better cores, and improved graphics, I can't see it getting anywhere near the size of Trinity. Smaller dies mean better yields, and hence lower production costs- meaning better profit margins.
 

Zstream

Diamond Member
Oct 24, 2005
3,176
7
116
I started to do some more calculations but I realized that without knowing the true size of the die and how much is allocated to the GPU, its almost impossible to come up with a proper conversion to another AMD CPU.

Can we confirm that this is a CPU/GPU or is it a CPU only? If it has the GPU mixed it, all I can say is that this would appear to be AMD's conroe.
 

inf64

Platinum Member
Mar 11, 2011
2,868
170
136
@ Zstream

If you were referring to Kabini ES and supposed perf. numbers from Cinebench10, then yes it's Jaguar based APU (meaning it has GCN GPU on-die too).

The topic is about Richland though. If it's true that the chip is QC APU rated at 35W and that it runs at stock 3.2Ghz then it's massive improvement over 32nm 35W 2.3Ghz QC we have with Trinity. 40% higher clock,same TDP rating and better GPU on-die. But too many IFs an little concrete facts.
 

Zstream

Diamond Member
Oct 24, 2005
3,176
7
116
@ Zstream

If you were referring to Kabini ES and supposed perf. numbers from Cinebench10, then yes it's Jaguar based APU (meaning it has GCN GPU on-die too).

The topic is about Richland though. If it's true that the chip is QC APU rated at 35W and that it runs at stock 3.2Ghz then it's massive improvement over 32nm 35W 2.3Ghz QC we have with Trinity. 40% higher clock,same TDP rating and better GPU on-die. But too many IFs an little concrete facts.
I'm referring to what IDC posted. Whatever CPU that is and based off the numbers posted, it should be a great chip.

Unfortunately, AMD couldn't market it's way out of a wet paper bag. They could be making and selling HTPC units like no other. Imagine if AMD sold one of these APU chips with a media center edition (XBMC) and worked with the vendors to improve upon this product or bla bla I'm rambling now...
 

Vesku

Diamond Member
Aug 25, 2005
3,745
0
86
@ Zstream

If you were referring to Kabini ES and supposed perf. numbers from Cinebench10, then yes it's Jaguar based APU (meaning it has GCN GPU on-die too).

The topic is about Richland though. If it's true that the chip is QC APU rated at 35W and that it runs at stock 3.2Ghz then it's massive improvement over 32nm 35W 2.3Ghz QC we have with Trinity. 40% higher clock,same TDP rating and better GPU on-die. But too many IFs an little concrete facts.
Yes, if the 35W AMD TDP rating and 3.2GHz non-turbo clock are accurate than that's quite a nice bump in efficiency. I very much doubt Kabini would be more efficient at almost double it's 17W clocks, assuming it could clock that high.
 

pablo87

Senior member
Nov 5, 2012
374
0
0
Yes, if the 35W AMD TDP rating and 3.2GHz non-turbo clock are accurate than that's quite a nice bump in efficiency. I very much doubt Kabini would be more efficient at almost double it's 17W clocks, assuming it could clock that high.
One should not doubt AMD's Honey Badger part. :p
 

ASK THE COMMUNITY

TRENDING THREADS