AMD - Piledriver chitchat

[AETAaAS]

AMD dropped the price on 3 chips, the A6 6400k ($72->$62), the A8 6500 ($112->$97) and the A8 6600k ($112->$97). Representing a saving of around 13% on each.

Also 6790k has been priced in at $122, slotting in just under the 6800k which costs $142. Same graphics part, just a drop of 100mHz to clock (4.0) and turbo (4.3).

I like the price cuts at the lower end, where every dollar you can save can make a difference but the 6790k is confusing. It's so close to the 6800k and delivers most of its performance but costs 14% less. It's unlocked too so you could probably sneak past the 6800k with a few tweaks. the 6790k might be a price cut for the upper end, just not explicitly so.

Hopefully we see the changes reflected over here.

[CAT-THE-FIFTH]

http://semiaccurate.com/2013/11/09/g...-benchmarking/

I remember the week before the GTX680 launched,that I mentioned the GPU boosting mechanisms would be used to inflate benchmark scores,and I don't think many believed me. It looks like 18 months later it seems that was a correct observation!!

[CAT-THE-FIFTH]

I posted this a few days ago:

http://hardforum.com/showpost.php?p=...&postcount=212

This sounds exactly like an NDA meeting I was in last week. The fan was referred to multiple times as "jet engine" etc., but all performance comparisons shown were with the card in Quiet Mode. Funny stuff when you are reaching for straws.

According to Guru3D,Toms Hardware did not discover anything and got the information from Nvidia.



Tested in quiet mode of course!

BTW,performance we can trust!!

Yes,Nvidia that is why multiple review sites showed throttling with extended test runs on the GTX660TI,GTX760 and Geforce Titan in open air test rigs.

LOLCOPTER!

[abychristy]

This sure is a long thread !

[kalniel]

This sure is a long thread !

Thanks for your contribution! If you haven't read every single post then go back and start again from the beginning

[DanceswithUnix]

I like the price cuts at the lower end, where every dollar you can save can make a difference but the 6790k is confusing. It's so close to the 6800k and delivers most of its performance but costs 14% less. It's unlocked too so you could probably sneak past the 6800k with a few tweaks. the 6790k might be a price cut for the upper end, just not explicitly so.

The 6790K was briefly discussed a few pages back. It lacks the support for DDR3-2133 which AMD are supposed to test specially for on the high end chip. Cheaper RAM makes for a good chip for OEMs.

[AETAaAS]

The 6790K was briefly discussed a few pages back. It lacks the support for DDR3-2133 which AMD are supposed to test specially for on the high end chip. Cheaper RAM makes for a good chip for OEMs.

Ah missed that. Makes it less attractive for me now since faster RAM is a nice way to get more out of the APUs.

[scaryjim]

Don't forget that not officially supporting DDR3-2133 isn't the same as not being able to run DDR3-2133. If you're looking at it as a cheaper alternative to the 6800k to try to tweak the performance up to match the 6800k, it could still be a worthy investment - it might just need a little more coaxing to get the best out of it. But then, that's kind of the whole point of overclocking, isn't it?

[AETAaAS]

Don't forget that not officially supporting DDR3-2133 isn't the same as not being able to run DDR3-2133. If you're looking at it as a cheaper alternative to the 6800k to try to tweak the performance up to match the 6800k, it could still be a worthy investment - it might just need a little more coaxing to get the best out of it. But then, that's kind of the whole point of overclocking, isn't it?

Sir, I like your attitude. Just because something isn't towing the 'official line' does not mean we aren't going to do it.

Would be interested in a review though because I'm guessing there is some binning going on and the 6790k doesn't tick the boxes the way the 6800k does somehow?

[scaryjim]

Yeah, it'll just be second bin silicon: presumably it won't certify either at 2133MHz memory to ship as a 6800k, or at the lower voltage used by the 6700 to keep to the TDP down to 65W*. I guess at that point you're looking at either disabling parts of an otherwise good core to make a 6600k, or you launch an almost-top-tier APU with the memory and core speeds notched back one step to provide product differentiation.

As DwU said, most big OEMs and SIs won't bother with 2133MHz RAM anyway - they'll likely stick to 1600MHz (or maybe 1866) for cost purposes, so a processor that will certify at 4GHz/1866MHz at 100W TDP makes a certain amount of sense - a cheaper APU to go with the cheaper memory but still retaining much of the performance.

*EDIT: indeed, it could easily be an otherwise fully functioning chip with a slightly sub-par memory controller. Of course, the question then would be *how* sub-par: could it be coaxed to higher speeds with a small bump in voltage, or would you be flogging a dead horse. No way to tell without trying, sadly.

[kompukare]

There probably is some binning going on with the 6790K but that just means the 6800K's have been tested and passed. Doesn't guarantee that the 6790K's were tested and failed though. Sort of a false positive vs a false negative.

Testing really has to do with demand. Companies are unlikely to test more than is required. Pure speculation this, but I would imagine that the likes of AMD, Intel etc., do have ways of testing without someone sitting there and running MemTest on a bunch of CPUs with tons of RAM kits; or at least a way of finding dies which are likely to successfully pass. 900+ pins is a lot of pins and at least some of them might be used for diagnostics like this.

[watercooled]

Having a few people manually running memtest isn't a terribly efficient way of testing thousands/millions of dies.

They use custom testing jigs to perform diagnostics on dice before they're even cut from the wafer - ones which fail will be flagged, often by marking with some sort of dye, and they're later discarded. Binning may start at this stage, for example to sort mobile from desktop parts, from which they'll be sent on to packaging, soldering die on to substrate. From there, further testing and binning will likely be carried out, both to check the packaging stage went successfully, and to further sort because the substrate may have an impact on the electrical characteristics of the chip.

There's an awful lot more to it than that of course, but it's highly automated and thorough, it's highly unlikely guesswork or skipping involved considering the way the production line works, and the relatively insignificant costs of this stage of testing. Even if a company gets a load of highly-binned dice, they may still sell them as lower tier parts based on supply+demand. However even within a product tier there may be several ranked bins.

[DanceswithUnix]

Sir, I like your attitude. Just because something isn't towing the 'official line' does not mean we aren't going to do it.

Would be interested in a review though because I'm guessing there is some binning going on and the 6790k doesn't tick the boxes the way the 6800k does somehow?

It may be simpler than that.

There was a comment from AMD when the 6800K came out that they test the memory interface specially to bin the ones that can go that fast. That implies that part of the chip is a bit on the edge, but the important bit may simply be this. It takes time to run tests. Time on any production line costs money. Cut out the extended tests, just do the basic ones they all do that you expect all parts to pass with flying colours, you automatically get a cheaper part.

So I suspect this hasn't failed the test, as *only* 6800K parts run the test, so it hasn't even run the test.

Now here is what I think about the CPU frequency. How can you have a product that just hasn't had as much testing? Hard sell, hard to rate. No, they have to make it faster or slower even if just a bit. They can't make it faster, so knock the smallest about of frequency off it and you have a new part.

So, I think it is a down clocked 6800K with less testing. Feeling lucky?

[watercooled]

Of course, we can't know for sure, but as I said I doubt skipping the automated testing for certain chips would save much/anything in the way of time/money, and could cause more trouble than it's worth.

Depending on yield and sales volumes for the various parts, they risk scenarios like giving away a load of unknowingly very high binned dice at low prices, and simultaneously hit supply issues for the higher end products if they get a lower-than-expected yielding wafer or two back which do go through the testing, which would eat into profits as they'd need more wafers ordered to make up for it, and end up with an excess of lower-binned parts. IMO it makes more sense to correctly bin everything and, if the need arises, use some of the higher-binned dice to satisfy demand.

I can't imagine the difference between minimal and 'extensive' testing to be that different - they still need to run through the testing stations and sit there while it's fault tested at various settings - you can't just skip tests or you risk wasting money and reputation by releasing chips with fatal manufacturing flaws which could corrupt data, etc.

[DanceswithUnix]

I spent a few years working on test systems on production lines for electronic goods.

If it is an additional test for higher speed memory, then it can be skipped. Every fraction of a second matters on assembly lines, so any sane system is very finely grained and tunable. The implication from AMD was that only candidates for being a 6800K run that test, the rest already skip it.

As for stock of 6800K, I would imagine they are trying to get shot of them around now, with the replacement part on the horizon.

[watercooled]

OK I've just re-read your previous post and spotted the bit about *additional* testing - I must have misread it first time.

Yeah I can see that working, again depending on yield/demand; they could have a load of candidates that pass initial/base testing for binning, then they pull a certain number of those to further test the IMC for stability at higher clocks - as long as they have some sort of buffer, so to speak, they could just return non-passable dice to the previous bin and pull a few more to make up for them.

IMO this sort of testing makes more sense in a scenario where you expect most candidates to pass (high yielding) - if you end up struggling to meet demand because of low yield, it may make more sense to run everything through this test in order to capture as many high-performing chips as possible. Then again, I suppose there could be some level of initial candidate selection/filtering based on performance of previous tests i.e. predicting which are more likely to pass based on, for example, voltages needed to reach the 'standard' clocks.