AMD Carrizo / Toronto

Yes, all development effort on the Cat cores was probably axed in Zen's favor. That said, I don't think the market for these chips can be considered "very small". They offer acceptable performance for very low end machines at attractive prices. That's not to say that they're very profitable for AMD, but I think I've read somewhere that they make up 40% of AMD's unit sales.
 
AMD has been subject to attrition and/or poaching. I recall that one of the heads for the Jaguar team left, (Samsung maybe?). There were also other indications of people being hired out of AMD. The comments for the Techreport's video discussion with David Kanter included a note that much of that team is gone.
http://techreport.com/discussion/28...ideo-amd-zen-fiji-and-more?post=905853#905853

With Zen taking the forefront, we have piecemeal updates to existing cores or hardware revisions, and a K12 that has been delayed significantly.
The throughput of the core development pipeline seems very limited.
 
Big bunch of Carizzo slides here http://wccftech.com/amd-carrizo-apu-architecture-hot-chips/
On the face of it a reasonable IPC bump & a big power saving bump.

Apparently UVD & dual Encode blocks are a pretty big chunk of that Northbridge bit o_O

Moneyshot about same size as the earlier one but clearer.
AMD-Carrizo-Die.jpg
 
Last edited:
AMD has been subject to attrition and/or poaching. I recall that one of the heads for the Jaguar team left, (Samsung maybe?).

Wasn't just one person. Bobcat was developed by a very small team with little resources, and given the resources the team had the result was considered to be absolutely stellar. Soon after Bobcat was released to public, Samsung poached most of the team and used them to form their own new CPU design group in Austin. The rumor I heard was that Samsung poached one guy from the top, asked him who his best employees were, and extended offers to all of them. Most of them took it. Then, as AMD refilled the positions, Samsung kept poaching the best ones, practically using AMD Austin as a recruiting filter. This is what caused the horrible delays and product cancellations in the bobcat line.

The team that Samsung formed started designing their own ARM core, called Mongoose. Initially, the plan apparently was mostly to sell it to Apple, but Apple built their own CPU cores instead (incidentally, also by a team that's heavy staffed by AMD alumni, only in their case from the K8 team whose remnants they acquired with PA Semi), and so Samsung is now targeting them to their own SoCs. The first chips with them are expected to be released later this year.

AMD is so cash-poor and the prospects of the stock are so low that they can do little to prevent this kind of talent loss when their employees are targeted by a company who is willing to throw lots of money at good employees.
 
Wasn't just one person. Bobcat was developed by a very small team with little resources, and given the resources the team had the result was considered to be absolutely stellar. Soon after Bobcat was released to public, Samsung poached most of the team and used them to form their own new CPU design group in Austin. The rumor I heard was that Samsung poached one guy from the top, asked him who his best employees were, and extended offers to all of them. Most of them took it. Then, as AMD refilled the positions, Samsung kept poaching the best ones, practically using AMD Austin as a recruiting filter. This is what caused the horrible delays and product cancellations in the bobcat line.

The team that Samsung formed started designing their own ARM core, called Mongoose. Initially, the plan apparently was mostly to sell it to Apple, but Apple built their own CPU cores instead (incidentally, also by a team that's heavy staffed by AMD alumni, only in their case from the K8 team whose remnants they acquired with PA Semi), and so Samsung is now targeting them to their own SoCs. The first chips with them are expected to be released later this year.

AMD is so cash-poor and the prospects of the stock are so low that they can do little to prevent this kind of talent loss when their employees are targeted by a company who is willing to throw lots of money at good employees.

I had found indirect accounts of the overall trend, although only in broader strokes than that.
The way alumni from other high-end processor design teams show up, and how design houses can frequently trace a lineage back to prior design teams, shows how uncommon that kind of skill set is.


Some information on delta colour compression:
http://cdn.wccftech.com/wp-content/uploads/2015/08/AMD-Carrizo-APU_Graphic-Color-Compression.jpg
"5-7% improvement on games [...]"
Oh.

The power savings seem to be the biggest target.
Being bandwidth-limited would happen for one fraction of the workload, but the placement of the compression process and the unknown ratio means that the ROPs do not see the benefit directly.
The color cache is going to flush and fill just as much as before, and if anything there may be some kind of latency hit from it.

Even if the compression is a significant share of that 40% figure, it's now a stream of just as many flushes and loads, just with gaps in it that now need something else to fill them.
The performance uplift per area increase seems decent, given that it is software transparent.
 
Wasn't just one person. Bobcat was developed by a very small team with little resources, and given the resources the team had the result was considered to be absolutely stellar. Soon after Bobcat was released to public, Samsung poached most of the team and used them to form their own new CPU design group in Austin. The rumor I heard was that Samsung poached one guy from the top, asked him who his best employees were, and extended offers to all of them. Most of them took it. Then, as AMD refilled the positions, Samsung kept poaching the best ones, practically using AMD Austin as a recruiting filter. This is what caused the horrible delays and product cancellations in the bobcat line.

The team that Samsung formed started designing their own ARM core, called Mongoose. Initially, the plan apparently was mostly to sell it to Apple, but Apple built their own CPU cores instead (incidentally, also by a team that's heavy staffed by AMD alumni, only in their case from the K8 team whose remnants they acquired with PA Semi), and so Samsung is now targeting them to their own SoCs. The first chips with them are expected to be released later this year.

AMD is so cash-poor and the prospects of the stock are so low that they can do little to prevent this kind of talent loss when their employees are targeted by a company who is willing to throw lots of money at good employees.
Damn, great observation and annoyingly accurate. Thank you.
 
Some information on delta colour compression:

http://cdn.wccftech.com/wp-content/uploads/2015/08/AMD-Carrizo-APU_Graphic-Color-Compression.jpg

"5-7% improvement on games [...]"

Oh.
I would have expected bigger gains too. But 5-7% improvement at negative power cost is not that bad. It seems that the GPU is just not big enough. Would be interesting to know how much this saves power.

It will be interesting to see whether AMD has anything to pit against the Intel 72 EU Skylake. 48 EU Broadwell (with 128 Mb EDRAM) was already almost twice as fast as AMDs flagship APU.
 
GPU in Carrizo has only 8 ROPs (just like Kaveri), so it isn't as BW limited as the DDR3 models of Cape Verde (16 ROPs). I think with 16 ROPs these gains would be significantly higher.
 
GPU in Carrizo has only 8 ROPs (just like Kaveri), so it isn't as BW limited as the DDR3 models of Cape Verde (16 ROPs). I think with 16 ROPs these gains would be significantly higher.
I don't think you'd really need more ROPs to show more benefits. Don't forget the thing runs at like 400Mhz (if configured for 15W) which gives it about twice the bandwidth per ROP (or per ALU) compared to the 128bit Mars chips (which, unlike the 64bit brothers, do not seem to be that badly bandwidth limited themselves). Assuming dual channel configuration (which still aren't quite the norm...) of course. I want to see some single-channel and 45W numbers for comparison, could well make more difference there.
 
First official Carrizo laptop?

Lenovo Ideapad 500:
http://www.anandtech.com/show/9576/...ad-lineup-with-windows-10-skylake-and-carrizo

$399 for a 14" laptop with a SSD, FHD screen, Carrizo FX-8800P, R5 M330 discrete GPU (for Hybrid Crossfire?) and 802.11ac wireless by default seems very good.
It's still a drop in the ocean compared to all the Skylake design wins made at the same time from Lenovo, though.

I think this is the first attractive AMD laptop I've ever seen. I'd like it better without the discrete graphics, a bit smaller and cheaper*, but it's the first time that an OEM has made an AMD laptop that fits my needs and is priced attractively.

*$399 is the price with an HDD, I think. With the SSD option it's probably more than that, but shouldn't be too high either.
 
It says 128GB SSD or 1TB HDD. At least in retail, a 1TB 2.5 HDD is going for about $50, which is actually the same as a standard 128GB SSD.
 
So, where can I buy a Micro ITX board with a FX8800P on it?

A shame, it would be ideal also for a compact, general purpose HTPC without a dedicated GPU to keep space and heat down to minimal levels...
 
Socketed Carrizo called Bristol Ridge is coming sometime in 2016, together with DDR4. I'm pretty sure there'll be a couple of socket FM3 miniITX boards in time for launch.
Of course by then, Bristol Ridge will have to compete in price and performance with Intel's Kaby Lake and its 2-core/4-thread Core i3 models with their updated GT2 GPUs.
 
Back
Top