AMD - Piledriver chitchat

[watercooled]

With AMD confirming Richland is on 32nm, I wonder how it differs from Trinity? Surely they wouldn't bother with another platform name if it's just a clock speed bump?

Edit: It looks like they've improved the GPU? Is it a newer design or have they upped the shader count from Trinity? I've lost track amid all the rumours lately TBH...

[CAT-THE-FIFTH]

Maybe they changed over to GCN cores or have modified the CPU cores in some way?? They do say that Richland is meant to be 20% to 40% faster overall.

However,have you noticed that Kaveri is within the 2013 launch window for AMD though?? It would mean Trinity and Richmond are short lived products.

[CAT-THE-FIFTH]

The first comment on the TR coverage is informative:

http://techreport.com/news/24164/amd...esktop-roadmap

It seems DiRT:Showdown was running smoothly on a Temash based tablet!

10:44 PM
+-+14
#1
Since we're talking about CES 2013, I did a little digging on AMD's presentation...(I'm curious as to what they will bring in the APU lines).

CES: AMD Unveils Kabini, Temash Chip Families; Vizio Partnership On PCs, Tabs
=> http://www.forbes.com/sites/ericsavi...chip-families/

AMD Press Release: AMD Unveils Innovative New APUs and SoCs that Give Consumers a More Exciting and Immersive Experience
=> http://www.amd.com/us/press-releases..._new_apus.aspx


Basic gist...According to AMD Performance Labs:

* Temash (Jaguar-based APU for tablets)

Testbed:
* AMD Z-60 APU (Hondo) with Radeon HD 6250 IGP
=> 1.0Ghz, dual-core
=> 2x 2GB DDR3-1066
=> Windows 7
VS
* AMD A6-1450 (Temash) with Radeon HD 8280 IGP
=> 1.0Ghz, quad-core
=> 2GB DDR3-1066
=> Windows 7

FutureMark 3DMark Vantage P (Synthetic GPU test)
=> Z-60 (Hondo): 455
=> A6-1450 (Temash): 981

PCMark Vantage (Synthetic CPU test)
=> Z-60 (Hondo): 1914
=> A6-1450 (Temash): 3123



Kabini (Jaguar-based APU for ultrathin notebooks and SFF PCs)

Testbed:
* AMD E2-1800 APU (Brazos 2.0) with Radeon HD 7340 IGP
=> 1.7Ghz, dual-core
=> 4GB DDR3-1333
=> Windows 7 Ultimate.
VS
* AMD A6-5200 APU (Kabini) with Radeon HD 8400 IGP
=> ???Ghz, quad-core? (undisclosed!)
=> 4GB DDR3-1600
=> Windows 8 64bit.

PCMark Vantage (Synthetic CPU test)
=> E2-1800 APU (Brazos 2.0): 2807
=> A6-5200 APU (Kabini): 5271

AMD demonstrated working silicon of its first true system on chip (SoC) APUs, codenamed “Temash” and “Kabini,” which will be the industry’s first quad-core x86 SoCs. Both APUs are scheduled to ship in the first half of 2013.




Richland (Piledriver-based APU. Mobile versions compared.)

Testbed:
* AMD A10-4600M (Trinity) with Radeon HD 7660G IGP
VS
* AMD A10-5750M (Richland) with Radeon HD 8650G IGP

* AMD A8-4555M (Trinity) with Radeon HD 7600G IGP
VS
* AMD A8-5545M (Richland) with Radeon HD 8510G IGP

All configurations:
=> 4GB DDR3-1600 (Dual Channel)
=> Windows 7 Home Premium 64-bit

3DMark 11
=> A10-4600M (Trinity): 1150
=> A10-5750M (Richland): 1400
and
=> A8-4555M (Trinity): 780
=> A8-5545M (Richland): 1100

PCMark 7 Overall benchmark
=> A10-4600M (Trinity): 1965
=> A10-5750M (Richland): 2175
and
=> A8-4555M (Trinity): 1650
=> A8-5545M (Richland): 1850



Speaking of Kaveri APU...

The follow-on to "Richland" will be the 28nm APU codenamed "Kaveri" with revolutionary heterogeneous system architecture (HSA) features which is expected to begin shipping to customers in the second half of 2013.


We'll probably see something in either late-2013 or early-2014 ???



The AMD CES 2013 presentation slides can be found here...
=> http://www.slideshare.net/AMD/amd-ce...ess-conference

Most of it is fluff. Nothing much to talk about.

* Both Jaguar-based Kabini and Temash APUs will be officially marketed as A4 and A6 APUs. They are made on the 28nm process.

* The Piledriver-based Richland (aka: Trinity 2.0) APU will be officially marketed as A8 and A10 APUs. They are made on the existing 32nm process.



Overall?
* Richland is incrementally faster than Trinity in CPU performance. The biggest performance improvement is the GPU-based IGP.
* The Jaguar-based Temash and Kabini are a major improvement over existing Hondo and Brazos 2.0 respectively (in both CPU and IGP performance).
* AMD is getting rid of Z-series and E-series marketing designations. It'll all be A-series by the end of 2013.

[watercooled]

Yeah I spotted that earlier over at Anandtech, as I posted further up. Considering Kaveri was supposedly pushed back to 2014, it would be bad PR to bring it forward only to push it back again? Either way, it looks like Richland might only be on the market for six months or so! It seems like it was meant as a stop-gap while they ironed out issues with Kaveri, only to fix said problems soon after.

20-40% does seem a lot, even for marketing, if nothing much has changed?

I could speculate all day, but I guess we'll have to wait for Richland to see what the deal is.

[scaryjim]

Taking those Richland figures at face value, I'd guess that they've done a bit of a respin and managed to get less leakage. Most of the benefit is in the GPU, and a bigger gain is made by the low voltage parts than the high, so I'd put money on respin and higher clocks. I also strongly suspect that we might not see Richland desktop APUs at all.

Remember - Trinity mobile came out in May last year, so Richland looks like launching around a year on from that. That'd give a decent boost to OEM laptop designs coming out now, without having to rebrand (see today's news stories about the GPUs for how well that goes down ). Make it an incremental, mobile-only launch, with most of the performance gains coming from being able to clock higher at the same TDP - particularly on the GPU, and in ULV parts. Launch Kaveri for desktop in October, a year on from desktop Trinity launch, with a new core.Mobile Kaveri launch will then depend on how well received Richland is, and how Haswell turns out. Frankly, much higher performance in ULV parts is a big win for AMD, as mobile looks to be a bigger growth area than desktop.

[watercooled]

You could be right, but I keep seeing references to Richland on dekstop and FM2, along with desktop part numbers?

[CAT-THE-FIFTH]

Even a 20% improvement in the IGP for the desktop A10 would mean it is probably going to be faster than an HD6570 GDDR5 or HD6670 GDDR3. This would probably place it nearer to 9800GT level performance.

Also,I think some of you missed this post I made:

http://forums.hexus.net/cpus/241925-...ml#post2765825

It is an AMD document referring to the 6000 series APUs. Desktop versions now support 2133MHZ DDR3. Also the L1 cache is stated as being 3 cycle,although Trinity is supposed to use 4 cycle L1 cache,and the CPU now supports additional sleep states.

[scaryjim]

3 cycle latency isn't marked as a document change, so it looks like that's always been part of the Trinity specification (asuming that's what Family 15h is ). The only changes that lists are minor tweaks to the sleep state and memory capabilities (lower voltage DDR3, higher clock speeds). All of that would fit very nicely with my conjecture that they've reduced leakage so can run the silicon at higher clock rates or lower voltages.

The product data sheet certainly allows for Richland-based desktop APUs. However, that doesn't mean any will be released. If Kaveri is going to be ready before the end of 2013, it would make more sense to seek to move Trinity inventory rather than release another new APU (there's still a lot of Llano sitting in the channel - presumably unsellable since Trinity's release). The only way it makes sense to release Richland on desktop is if Trinity inventory is already low, and they release a limited amount of Richland so they don't end up with surlpus inventory when they do launch Kaveri.

[CAT-THE-FIFTH]

This post on SA sums up the information on the parts released so far:

http://www.semiaccurate.com/forums/s...&postcount=104

The Tamesh quad core has a 5.9W TDP for the entire SOC,so it will be interesting to see what the TDP of the dual core version is!

It looks like the Kabini die is 114MM2 for the entire SOC as opposed to the 74MM2 for Bobcat:

http://www.semiaccurate.com/forums/s...&postcount=121

The southbridge is 28MM2 for Brazos,so it appears the total die area used has not changed with Kabini.

DiRT Showdown was run on a Temash tablet at 1080P:

http://www.theverge.com/2013/1/7/384...a-1080p-tablet

Video:

http://www.youtube.com/watch?v=FruxOZ9Nfp0



The game could be smoother but still it is pretty impressive that it actually runs at that resolution on a tablet!!

[CAT-THE-FIFTH]

Taking those Richland figures at face value, I'd guess that they've done a bit of a respin and managed to get less leakage. Most of the benefit is in the GPU, and a bigger gain is made by the low voltage parts than the high, so I'd put money on respin and higher clocks. I also strongly suspect that we might not see Richland desktop APUs at all.

Remember - Trinity mobile came out in May last year, so Richland looks like launching around a year on from that. That'd give a decent boost to OEM laptop designs coming out now, without having to rebrand (see today's news stories about the GPUs for how well that goes down ). Make it an incremental, mobile-only launch, with most of the performance gains coming from being able to clock higher at the same TDP - particularly on the GPU, and in ULV parts. Launch Kaveri for desktop in October, a year on from desktop Trinity launch, with a new core.Mobile Kaveri launch will then depend on how well received Richland is, and how Haswell turns out. Frankly, much higher performance in ULV parts is a big win for AMD, as mobile looks to be a bigger growth area than desktop.

It looks like TR managed to get some more information about Richland from AMD:

http://techreport.com/news/24190/pro...d-richland-apu

Richland looks more like Trinity with process tweaks and more intelligent power management.

[CAT-THE-FIFTH]

It seems Temash has 128 GCN shaders:

http://semiaccurate.com/forums/showp...&postcount=153

Jim Keller is also working on the big cores it seems and leading development of them.

[scaryjim]

So pretty much what I said plus some firmware magic sauce - fair enough

If it is power management, it makes sense that the biggest improvements will be seen where power is at a premium - i.e. ULV mobile. I doubt this will improve on the 5800k that much, but imagine what the 5700 replacement will be like - the best desktop Trinity was undoubtedly the 5700, and that's the desktop product that will benefit most from the move to Richland? Can I get a Hell Yeah?!

This could make whatever replaces the 5500 pretty desirable too - and maybe we'll see 45W dual cores? Ooooh, so many possibilities...

[CAT-THE-FIFTH]

This video is funny:

http://www.youtube.com/watch?v=eu8Sekdb-IE

[watercooled]

Actually really interesting that, I was half expecting it to be another anti-AMD troll video!

He's running games at realistic, higher settings, rather than having everything including resolution at absurdly low/minimum settings to increase single-threaded stress on CPU, often meaning hundreds of FPS and showing Intel with an advantage. My theory anyway...

Unfortunately, hardware benchmarks are susceptible to reviewer bias, so this makes a refreshing change.

People often complain about non-real-world benchmarks in other areas, but then accept completely irrelevant CPU gaming benchmarks, or, only really relevant if you pair your £200 CPU with a £40 GPU and/or enjoy extreme screen tear!

[CAT-THE-FIFTH]

One of my mates has an FX6300 overclocked to 4.3GHZ and a HD5850 1GB. I have a Xeon E3 1220 quad core running at 3.1GHZ to 3.4GHZ with an HD5850 1GB too and in PS2,a game meant to run badly on AMD CPUs and is very CPU heavy, we were having a similar gaming experience. That was in online maps with hundreds of players,vehicles,explosions,etc and we running the game at 1920X1080 at high settings without motion blur and the complex shadows off,and the game seemed smooth enough. IIRC,the minimum framerates stayed consistently above 30FPS.

The funny thing is that after reading all the forums on how people needed overclocked Core i5 at 4GHZ and high end cards to run this game,even mates with lesser hardware were able to run it at lower settings fine.

Anyway,TR managed to run a benchmark on a Kabini SOC which also uses OpenCL:

http://techreport.com/news/24186/new...15w-kabini-apu

[CAT-THE-FIFTH]

A bit off topic but I saw this post about one of the Intel smartphone SOCs:

http://forums.anandtech.com/showpost...&postcount=114

That poster is actually has a decent technical background. It might explain why Intel is trying to integrated the VRMs into the CPU.

Edit!!

This core size comparison he did was interesting:

http://forums.anandtech.com/showpost...01&postcount=1