AMD - Piledriver chitchat

[scaryjim]

... Perhaps now consumer motherboards are out the APUs will leak into the retail markets. ...

Certainly my assumption is that mobo manufacturers refused to push AM4 boards into the channel at a time when they'd have been launching alongside construction-core CPUs and APUs, so AMD couldn't release the Bristol Ridge CPUs and APUs to the channel as there were no supporting motherboards. My anticipation is that we'll see BR CPUs and APUs slowly drift into channel over the next few months to fill out the £120 and lower market that Zen won't be reaching down to (yet)...

[watercooled]

TBH I'd kinda forgotten about BR until I noticed most of the AM4 motherboards have display outputs on them. TBH though, aside from the DDR4 memory controller is it all that different from Carrizo anyway? I know OEMs love new part numbers for their product refreshes, but I can see it having limited appeal in the retail market.

[CAT-THE-FIFTH]

The problem is BR is going to look very ordinary against a Pentium G4560 unless you need a decent IGP.

[DanceswithUnix]

The problem is BR is going to look very ordinary against a Pentium G4560 unless you need a decent IGP.

As long as it is vaguely competitive I think that would be fine.

Get a cheap BR now, upgrade to Ryzen later can generate a fair few sales.

[scaryjim]

The problem is BR is going to look very ordinary against a Pentium G4560 unless you need a decent IGP.

I'm not so convinced - the G4560 is only clocked at 3.5Ghz; the A12 9800 is 3.8GHz/4.2GHz. It's almost certainly still going to lose in single-threaded benchmarks, but I reckon it'll be closer than we've seen from any other construction core.

Have a look at the 3.5/3.8GHz Athlon X4 845 against a 3.7GHz Core i3 6100: http://www.anandtech.com/bench/product/1684?vs=1682 - while there are some big wins for the i3 there, there are also some very close tests - significantly the dGPU gaming - and one or two wins for AMD. Take 5% off the Intel scores and add 10% to the AMD scores and it's not necessarily a clear cut race any more...

[CAT-THE-FIFTH]

As long as it is vaguely competitive I think that would be fine.

Get a cheap BR now, upgrade to Ryzen later can generate a fair few sales.

I'm not so convinced - the G4560 is only clocked at 3.5Ghz; the A12 9800 is 3.8GHz/4.2GHz. It's almost certainly still going to lose in single-threaded benchmarks, but I reckon it'll be closer than we've seen from any other construction core.

Have a look at the 3.5/3.8GHz Athlon X4 845 against a 3.7GHz Core i3 6100: http://www.anandtech.com/bench/product/1684?vs=1682 - while there are some big wins for the i3 there, there are also some very close tests - significantly the dGPU gaming - and one or two wins for AMD. Take 5% off the Intel scores and add 10% to the AMD scores and it's not necessarily a clear cut race any more...

The problem is the A12 won't be £65 and the G4560 will be - this is the issue.

Its really weird if Intel ends up better value at the low end than AMD - they really need to have a £70 CPU to compete with the G4560.

[DanceswithUnix]

The problem is the A12 won't be £65 and the G4560 will be - this is the issue.

Its really weird if Intel ends up better value at the low end than AMD - they really need to have a £70 CPU to compete with the G4560.

I don't expect an all out price war, simply because AMD don't have a war chest of money to fight with, but £120 for an A10 is looking on the high side after Intel's latest price movements so I think it has to come down.

The current Carrizo Athlon X4 at £57 would seem about right, except it is on a dead motherboard platform. AMD can make those work on AM4, so they should.

[scaryjim]

The problem is the A12 won't be £65 and the G4560 will be - this is the issue. ...

No, the issue will be if the BR Athlon X4 is significantly more than G4560. You said yourself, the people who'll go for the APUs are the ones who want the better IGP than Intel are offering (not only are IGP clocks significantly higher on BR, they support faster RAM through DDR4). An X4 845 equivalent at ~ £60 and a k series for ~ £75 so there's an overclocker's option in the low end of the market should do it for AMD, IMNSHO.

Next quarter we'll be getting the R5 and R3 CPUs to fill out the mid-range, and in the second half we'll get Zen APUs (and I suspect we'll see Athlon equivalents from those), so any BR releases will only need to fill out the product line for 9 months at the most.

[watercooled]

I don't expect an all out price war,

Likewise - another reason being Intel can probably ride their branding and marketing to demand a higher margin than AMD. Also, AMD are ultimately limited by fab capacity, and there's no point dropping prices beyond the point where they can't match demand. But I'm no expert when it comes to market economics.

Next quarter we'll be getting the R5 and R3 CPUs to fill out the mid-range, and in the second half we'll get Zen APUs (and I suspect we'll see Athlon equivalents from those), so any BR releases will only need to fill out the product line for 9 months at the most.

Have we heard what the 4C ryzen chips will look like? I've read the 6C parts will have symmetrical core complexes i.e. one core disabled on each, presumably to prevent minor performance differences between the different die configurations. I can't see them doing a separate 4C Ryzen die (apart from the APUs of course), but it seems like quite a lot to fuse off (albeit not unheard of). In theory a load of configurations would be possible, but I'm guessing a single active core complex is probably what they'll do if they choose just one, with my second choice being two cores per complex.

...and just while I'm typing this I remember the 'leaked', but fairly accurate so far, model list. The 6 cores have the full 16MB L3 cache with the 4 cores having 8MB which would point to a single active core complex. Interesting, as it doesn't allow much room for cache binning, but it's not unlike AMD to leave all of the cache available even for the lower core count CPUs.

[Xlucine]

If the memory controller is on-chip, could ryzen 2 support faster ram?

[watercooled]

CPU memory controllers have mostly been on-chip since Phenom. What is it you mean by faster RAM exactly? It looks like Ryzen already supports DDR4 3600 out of the box, but TBH memory speeds make very little difference anyway.

We probably won't be seeing DDR5 for a number of years yet if that's what you're referring to, but again there's not really much demand for it, on the desktop at least.

[scaryjim]

CPU memory controllers have mostly been on-chip since Phenom. ...

AMD's has been on chip since Athlon 64, in 2003 - Phenom didn't turn up until 2007 (it was notable for being the first monolithic quad core - Intel's Core 2 Quads were MCMs of 2 Core 2 Duos linked together by FSB). Intel brought in IMCs with Nehalem in 2008, a full five years behind AMD.

I have a vague feeling that back in the Athlon 64 days trhere may have been platform limitations on memory speed - I can't remember if Phenom/Phenom II based processors were able to run 1066MHz DDR2 in original AM2 motherboards by default, or if something in the motherboard setup limited them to DDR2-800 (it may have been as simple of tweaking the settings in the BIOS, of course, if the board didn't support 1066MHz by default).

At the very least I'd expect it to be fairly straightforward to "overclock" the memory on a first gen AM4 board if future Ryzen chips gets higher supported memory clocks; but given the degree in integration Ryzen brings (full SoC, remember) I wouldn't be at all surprised if the AM4 boards read the supported memory speeds off the chip and adapt accordingly...

[shaithis]

IIRC, they were limited when using 4 sticks in some cases....

I am of the opinion though that for most the memory speed isn't going to make any real-world difference. Not sure what all the kerfuffle is about.

[scaryjim]

... Not sure what all the kerfuffle is about.

1) There are some workloads that benefit from faster memory. Not many, mind you, and the relevance of those tests also shrinks when you start getting big L3 caches involved, but nonetheless those tasks do exist. If thos etasks line up with the ones AMD reckon Ryzen is good at, then faster memory becomes a key consideration in speccing a system.

2) Ryzen-based APUs should only be half a year away. They'll love fast memory. We could see another step-change in IGP performance if AMD cram more shaders and 3600MHz+ memory support into a Ryzen APU...

[shaithis]

2) Ryzen-based APUs should only be half a year away. They'll love fast memory. We could see another step-change in IGP performance if AMD cram more shaders and 3600MHz+ memory support into a Ryzen APU...

Lets get them out before getting carried away buying 1000000MHz RAM

[CAT-THE-FIFTH]

Thats the thing - once go above 2400MHZ you are officially out of spec and overclocking the memory controller anyway. Plus even if RAM makes a difference,is it really worth spending so much money on those high end RAM kits?

The only reason that its being made a big deal,is since KL probably can have more of a memory controller overclock - but I hate the maximum overclocks argument since I actually want to keep my CPU for a few years,not have it end up degrading since people are playing a game of one-upmanship.

Intel might want to try and push KL but a sub 130MM2 chip selling for £350 is a joke.

They on purpose made sure the Xeon E3 route was locked out so instead of £175 to £200 for a 4C/8T Core i7 its now nearly £300ish. The same company could easily allow BCLK overclocking but made sure KL "fixed" the SKL "bug".

But I expect single core Super Pi will be the rage again,since the Core i7 7700K runs at XYZ percent higher clockspeed and has a XYZ IPC improvement.