AMD's 2012 lines: Komodo, Trinity, and, Krishna/Wichita

[stmok]

Originally Posted by mugginz

They could release Komodo as well as Vishera and keep both camps happy.

From what I can see...

For the 1st gen Bulldozer, they had the choice of releasing two major versions: One for AM3 and the top-end for AM3+. They built the solution for AM3+ first and looked into whether it was worth back-porting the design to AM3. It was not...Mainly because it would cost another 18 months in delay (collide with 2nd gen schedule?) and their market research indicated the majority of the consumer market doesn't upgrade their systems by opening their boxes and swapping out chips. It was a smaller group of Enthusiasts that did that.

One also needs to observe AMD's behaviour in relation to Intel...I've already mention that they've aligned themselves to Intel's Tick-Tock release model. Ivy Bridge is backward compatible with the current LGA1155 infrastructure. Would it safe to presume AMD is doing the same thing with AM3+ in order to match and surpass Intel at their own game? Would it not be wiser to dedicate one's limited resources to such a focus? It would certainly get Piledriver-based solutions out quicker to the market. (Which aligns with the new CEO's thinking. Seeing as he comes from Lenovo; an aggressive OEM provider.)


On the other hand, one possible approach could be AMD bringing Komodo (FM2) to counter Sandy Bridge-E (LGA2011).

Then you have...

FX-series (Komodo) vs SB-E => Top-End
FX-series (Vishera) vs Ivy Bridge (i5/i7) => Mainstream-Performance
A6-/A8-series (Trinity) vs Ivy Bridge (i3/i5) => Mainstream
E2-/A4-series (Trinity) vs Ivy Bridge (Celeron/Pentium) => Budget

...That's one possible avenue. (I'm guessing. I don't really know what the heck the new CEO is thinking!)


Personally, I don't think AMD sees the Top-End market as worth pursuing. It has a low Return-on-Investment. Its an e-penis at best. Great for marketing, but you won't make money from it due to the small market. It would be wiser to undermine the LGA2011 platform by offering features at a much lower price point.
=> Continue the "bang per buck" focus.

...Which is what they've done in the leaked press packet for the FX-series. Point out the costly LGA1366 platform as a comparison.

[mugginz]

Originally Posted by loophole

I think releasing two separate platforms for the high-end (or mainstream high-end if you like ) would be too much for AMD. They've already got two platforms with AM3+ and FM1, three if you include their BGA FT1 used for Brazos.

That's a lot more engineering and platform qualification effort that has to be passed on to OEMs and motherboard makers which they wouldn't like.

Just my 2c

I was thinking somewhat along the lines of the Pentium Overdrive processors for the 486 platform.

Vishera has Piledriver cores.
Komodo was to have Piledriver cores, and an additional two going up to 10.

If AMD could've released the new Komodo FM2 platform with its advantages over AM3+ and additionally provided the Piledriver cores to the AM3+ users in the form of AM3+ compatible Vishera you'd have happy AM3+ users and also new sales from those after the best of performance.

Unfortunately it would seem AMD are quite stretched from an engineering point of view and so I can see why AMD are unable to do this, but at the same time Intel are marching on so while Oct 12 may see parity (or there abouts) from AMD for a little while, if they're going to sit on there hands until Q3 2012 till Vishera is released Intel by then may very well be eating their lunch again.

[mugginz]

Originally Posted by stmok

Personally, I don't think AMD sees the Top-End market as worth pursuing. It has a low Return-on-Investment. Its an e-penis at best. Great for marketing, but you won't make money from it due to the small market. It would be wiser to undermine the LGA2011 platform by offering features at a much lower price point.
=> Continue the "bang per buck" focus.

The top end enthusiest may be a loss leader but you can't deny the flow on effects having a hero product has on the lower end products.

Ask your garden variety salesdude at Harvey Norman "Which brand CPU should I buy?" and if he's not selling based solely on kickback/special buy price, he'll likely tell you Intel. That position would reflect what's reported as the right thing to do from the tech press. If the tech press started saying "Must buy Bulldozer, Intel looses its crown" that would filter down eventually to the sales drones and others who drive purchase decisions.

Many car companies develop hero cars to put a shine on the brand overall, even if those hero cars don't themselves turn a profit. It's a well used tactic.

I think there's be benefits to AMD if they can maintain a halo of performance superiority, though obviously this isn't going to happen.

Originally Posted by stmok

...Which is what they've done in the leaked press packet for the FX-series. Point out the costly LGA1366 platform as a comparison.

That'll only work for the most naive out there. Anyone who knows what they're looking at will see that the more cost effective 1155 platform can outperform Bulldozer with 2600K in certain situations, and then there's 2700K coming and Ivybridge as well.

It's a bad look for a company when they use shallow marketing tactics to try and create a picture of a market that doesn't really exist. For the most part Bulldozers competition isn't 1366, it's 1155.

[mugginz]

Originally Posted by stmok

From what I can see...

For the 1st gen Bulldozer, they had the choice of releasing two major versions: One for AM3 and the top-end for AM3+. They built the solution for AM3+ first and looked into whether it was worth back-porting the design to AM3. It was not...Mainly because it would cost another 18 months in delay (collide with 2nd gen schedule?) and their market research indicated the majority of the consumer market doesn't upgrade their systems by opening their boxes and swapping out chips. It was a smaller group of Enthusiasts that did that.

One also needs to observe AMD's behaviour in relation to Intel...I've already mention that they've aligned themselves to Intel's Tick-Tock release model. Ivy Bridge is backward compatible with the current LGA1155 infrastructure. Would it safe to presume AMD is doing the same thing with AM3+ in order to match and surpass Intel at their own game? Would it not be wiser to dedicate one's limited resources to such a focus? It would certainly get Piledriver-based solutions out quicker to the market. (Which aligns with the new CEO's thinking. Seeing as he comes from Lenovo; an aggressive OEM provider.)

But if Q1/Q2 release of Komodo is off the table and we're to wait for Vishera in Q3 2012 then that doesn't align with Intel's product release schedule from a customer point of view. Ivybridge is "coming soon to a socket near you" but without Komodo, AMDers have to wait an awful long time for Vishera as they watch their Zambezi machines look slower and slower by comparison.

Originally Posted by stmok

On the other hand, one possible approach could be AMD bringing Komodo (FM2) to counter Sandy Bridge-E (LGA2011).

But with Komodo off the table there's a big hole there in their time line.

I'm happy for Komodo to turn up in Q1/Q2 with AM3+ guys having to wait for Piledriver until Q3 with Vishera, I'm less happy but agreeable with Komodo waiting for Q3 with Vishera but as you point out, that stretches resources at AMD, and if what's being reported is accurate, I'm not very happy at the prospect of there being no Komodo at all.

Originally Posted by stmok

Then you have...

FX-series (Komodo) vs SB-E => Top-End
FX-series (Vishera) vs Ivy Bridge (i5/i7) => Mainstream-Performance
A6-/A8-series (Trinity) vs Ivy Bridge (i3/i5) => Mainstream
E2-/A4-series (Trinity) vs Ivy Bridge (Celeron/Pentium) => Budget

...That's one possible avenue. (I'm guessing. I don't really know what the heck the new CEO is thinking!)

I'd personally be happy with that outcome you describe in your table if the AMD vs Intel release dates are close to each other. I just hope Komodo's not in the bin.

[stmok]

Squeeze this update in while I have a few minutes...

Fellow OCAU forum-ner .Radiant has pointed out a new leaked slide...

From http://www.donanimhaber.com/islemci/...k-detaylar.htm


This confirms last month's information...

AMD Cancels Next-Gen Komodo Processor, Corona Platform in Favour of New Chips.
=> http://www.xbitlabs.com/news/cpu/dis...New_Chips.html

So what was originally...

Performance/Enthusiast (Corona platform)
* Codename: Komodo
* Architecture: Piledriver
* Up to 10 cores
* Socket FM2
* PCI-Express 3.0
* Integrated Graphics: N/A (Discrete CPU)
* Triple-channel (???) DDR3 memory configuration.
* Replaces FX-81xx, FX-61xx, and FX-41xx series (Zambezi).


Is now...

Performance/Enthusiast (Volan platform or Scorpius Refresh)
* Codename: Vishera
* Architecture: Piledriver
* Up to 8 cores
* Socket AM3+
* PCI-Express 2.0
* Integrated Graphics: N/A (Discrete CPU)
* Dual-channel DDR3 memory configuration. (Re-use 9xx chipset)
* Replaces FX-81xx, FX-61xx, and FX-41xx series (Zambezi).


The good news is that those who have invested in a Socket AM3+ will see it carry on for another 18 months or so.

[stmok]

While I have a few minutes...

First Look: AMD Trinity APU, Linux Already Runs Well
=> http://www.phoronix.com/scan.php?pag..._trinity_early

For some reason, AMD engineers (?) or its partners have been posting results at OpenBenchmarking of the Trinity APU. (No one knows why, but its been done in a care-free manner.) ...They don't mean too much as they're early engineering samples on test platforms. The folks in charge of OpenBenchmarking have pulled the results. (In fear of hurting any relationship with AMD).

One of the AMD Linux engineering systems for Trinity is running nicely even on Ubuntu 11.04 with the Linux 2.6.38 kernel. The CPU string is AMD Eng Sample 2M252057C4450_32/25/16_9900_609 and its graphics are the Trinity Devastator Mobile with 512MB of video memory and an AMD Pumori motherboard. The PCI ID on the Trinity Devastator appears to be 0x9900. This Trinity APU is quad-core and running at 2.50GHz. The current quad-core Llano offerings are clocked at 2.6GHz (A6-3650) and 2.9GHz (A8-3850), while this Trinity part is clocked slower, it's numbers are nice compared to my A8-3850 Linux system.

Trinity APU (quad-core at 2.5Ghz) is at least on-par or better than A8-3850 (quad-core at 2.9Ghz). ...Well, under benchmarks in Linux.

Remember: alpha quality silicon!

Another early Trinity APU is the AMD Eng Sample 2D332158A2350_36/33/16_9991_711, which is a dual-core APU at 3.3GHz. This is fairly interesting with AMD's dual-core A4-3400 being clocked at just 2.7GHz. This APU is for a desktop system with Trinity Scrapper graphics. It's evidently working fairly well with Ubuntu 11.10 on the Linux 3.0 kernel. It appears that AMD hasn't done anything crazy with their engineering boards for Trinity as the Hudson Azalia driver and other components are being detected fine already on Ubuntu.

There's also a 3.3GHz dual-core Trinity APU on the AMD Annapurna. Ubuntu 11.10 on the Linux 3.0 kernel isn't able to detect the graphics and its PCI ID is 0x9991. The open-source Mesa / Gallium3D and Radeon DRM support hopefully won't be too far behind once the Trinity APUs launch in 2012. At least the early binary driver support appears in place. Depending upon what the retail price ends up being on this 3.3GHz dual-core Trinity APU, it should be quite an interesting product as the few numbers I've seen on it were quite good in relation to other hardware on OpenBenchmarking.org.

Yet another Trinity configuration is with the AMD Eng Sample 2M252157C4450_32/25/16_9903_609, but it's graphics are slightly different. It's another quad-core part, but it's graphics (PCI ID: 0x9903) don't appear to be working yet with the proprietary Catalyst driver -- or the AMD partner doesn't know how to install the binary blob.

[stmok]

New slide...

From http://www.donanimhaber.com/islemci/...-detaylari.htm

Trinity APU...
* Up to 4 cores (2 modules).
* Up to 20% improvement in digital media workloads over Llano.
* About 30% improvement in GPU performance over Llano's IGP.
* Official RAM speed support increased from DDR3-1866 to DDR3-2133.
* Added Secure Asset Management Unit (SAMU) ...Takes off DRM decryption workload from CPU to dedicated circuitry.
* Added Video Compression Engine (VCE) ...AMD's answer to Intel's Quick Sync.
* Re-use A75 chipset from Llano.
* Socket FM2...Socket FM1 is a dead-end. (Replicating Intel's LGA1156 period.)

I'm estimating that Trinity's IGP would put it in the same league as a discrete Radeon HD 5570 video card. While Intel's Ivy Bridge IGP will likely be in the same class as Llano's IGP.

I suspect nothing major really changes in terms of status quo for 2012. ie: Intel will have the CPU side overall, AMD will have the better IGP...And consumers will have more capable computing solutions for the same price points as today.

[AEKaBeer]

It really does seem like trinity could benefit from more memory bandwidth, I know it's a low cost solution but how much extra would it cost (total platform) for them to up it to tri or even quad channel memory?

[stmok]

Honestly, I have no idea.

They would have to change the memory controller (which is on the processor) and need to create a new reference design (for OEMs) that involves extra traces for the additional memory slots. Its something they won't do. AMD is re-using the chipsets on both FX and A-series platforms for 2012...It's a hint that they're trying to minimise development costs as much as possible.

I highly suspect the Trinity APU design is pretty much finalised in order to meet its planned release schedule next year.

Overall, don't expect the world from AMD any more. The glory days of K7/K8 architecture (and their competent engineering teams) are really over. The way they design processors today is more akin to System-on-Chip approach. (No longer focusing on pure performance)...That isn't going to change even with the recently announced CTO (Papermaster) at the helm of engineering. I guess my point is that one can expect more PR smoke blown in our faces for the next few years.

Intel will respond accordingly in terms of their tempo of price cuts. ie: They aren't going to drop prices on anything when there is no incentive to! (Consumers will see a stagnation.)

[itsmydamnation]

IB will not be anywhere near the high end llano class IGP on performance. Let alone features or IQ.

[dmandn]

Originally Posted by mugginz

The top end enthusiest may be a loss leader but you can't deny the flow on effects having a hero product has on the lower end products.

Ask your garden variety salesdude at Harvey Norman "Which brand CPU should I buy?" and if he's not selling based solely on kickback/special buy price, he'll likely tell you Intel. That position would reflect what's reported as the right thing to do from the tech press. If the tech press started saying "Must buy Bulldozer, Intel looses its crown" that would filter down eventually to the sales drones and others who drive purchase decisions.

Many car companies develop hero cars to put a shine on the brand overall, even if those hero cars don't themselves turn a profit. It's a well used tactic.

I think there's be benefits to AMD if they can maintain a halo of performance superiority, though obviously this isn't going to happen.





That'll only work for the most naive out there. Anyone who knows what they're looking at will see that the more cost effective 1155 platform can outperform Bulldozer with 2600K in certain situations, and then there's 2700K coming and Ivybridge as well.

It's a bad look for a company when they use shallow marketing tactics to try and create a picture of a market that doesn't really exist. For the most part Bulldozers competition isn't 1366, it's 1155.

....Though I can say from industry experience, the average buyer wants intel i7's, even those which do nothing more than check emails. Top end performance results and advertising do wonders for sales.

[stmok]

Some info from Chinese source...
=> http://www.chiphell.com/thread-307338-1-1.html









As you can see...
=> Trinity offers much higher clocks for both CPU and IGP compared to Llano.
=> Up to 30% better CPU performance compared to Llano under "digital media workloads".
=> About 30% better IGP performance compared to Llano's IGP.
=> Removed L3 cache from the 1st gen Bulldozer design.
=> Socket FM2 adds 125W TDP spec. Has one less pin compared to Socket FM1.
=> Dual channel memory (up to DDR3 2133Mhz).
=> 1 to 2 modules (2 and 4 core models).
=> Desktop (Socket FM2) and notebook (Socket FS1r2) models.
=> Hudson-D2 (A55 chipset); -D3 (A75 chipset); -D4 (A85X chipset).
=> The new A85X chipset appears to remove PCI support and adds RAID 5 capability.
=> Added Secure Asset Management Unit (SAMU) ...Takes off DRM decryption workload from CPU to dedicated circuitry.
=> Added Video Compression Engine (VCE) ...AMD's answer to Intel's Quick Sync.
=> As of October 2011, they are on A1 stepping.

...I guess they should be offering B0 or B1 stepping when the retail product comes along.


Questions I ponder on...
* Did they halve the L2 cache in this Piledriver-based APU compared to 1st gen Bulldozer? (From 2MB to 1MB per module).
* IGP: How many shaders at those speeds?
* Turbo mode algorithm: Llano had the issue of its turbo mode offering next to no effect as TDP budget was prioritised to the IGP side. Will Trinity be more balanced?

[bobbth]

0.7ghz seems like an aggressive turbo boost, does this suggest that we can expect cpu's with high heat output? ( and therefore the need for turbo boost, to keep temps down)

It'll be interesting to see how the bulldozer based cores fair against the stars cores, given bulldozers current performance. They wouldn't make the mistake of under-performing against their last gen products again, would they?

[stmok]

Originally Posted by bobbth

0.7ghz seems like an aggressive turbo boost, does this suggest that we can expect cpu's with high heat output? ( and therefore the need for turbo boost, to keep temps down)

Power consumption spec went up another 25W over Llano; to 125W. So, yes.

...So I'm guessing they're still held back by GlobalFoundries's production issues. Instead of waiting for GF to mature up its manufacturing, AMD decided to increase the TDP spec to meet their objectives. Well, at least for the desktop version.

Originally Posted by bobbth

It'll be interesting to see how the bulldozer based cores fair against the stars cores, given bulldozers current performance. They wouldn't make the mistake of under-performing against their last gen products again, would they?

Time will tell...But don't expect miracles.

[stmok]

According to this...
=> http://www.donanimhaber.com/anakart/...hazirliyor.htm

AMD is working on a chipset refresh for the 2nd generation FX-series (Vishera) to replace the current 9xx series in 2012. It seems they want to keep up with Intel's 7-series (for Ivy Bridge).

Top-end: 990FX => 1090FX
Middle: 990X => No refresh chipset in 2012!
Lower-end: 970 => 1070
South Bridge: SB950 => SB1060

They bring more 6Gbps SATA ports and native USB 3.0 support. But they don't offer native PCI-Express 3.0 support.

Remember, grain of salt for now...