Should AMD include an unlocked multiplier on Carrizo BGA SKUs?

[cbn]

AMD should lock-down the multiplier and engineer different CPU options based on thermals and power efficiency and work with OEMs to ensure the entire package is put together in the 'right way' for different applications.

Ensuring the right CPU in the right product is important for both market perception as well as performance. Intel does a very good at this and AMD should make this a focus, especially as we move away from older, thicker 'standard laptops' into more 'creative' designs and increasingly-thinner and lighter designs.

Having one or more Carrizo SKUs converted from locked multiplier to unclocked multiplier does not change what your writing.

What it does is increase freedom and versatility in DIY without affecting OEM.

The question, in my mind, is what SKUs can be changed from locked to unlocked multiplier without affecting AMD's existing product differentiation/segmentation. I am thinking definitely the top 35 watt SKU would be a candidate as well as an Athlon x4 SKU (which I doubt exists, but could be made from dies with defect to display, media, etc)

I would also like to see a ULV SKU with unlocked multiplier as well (being able to run passive at stock would be awesome), but I am guessing for this to happen it would need to have the same price and core/iGPU config as the top 35 watt SKU.

 

[cbn]

If my eyes aren't completely off, Excavator sits at ~0.95 at 15W in the chart, and ~1.05 at 25W. The curve even seems to droop at the end. In other words, frequencies increase by about 10%, while power draw increases by 66%. How is that anything other than horrendous scaling?

Meanwhile, we know Kaveri can do 3.1GHz base/3.3 Turbo @ 45W (A8-7600), 3.3 base/3.8 Turbo @ 65W (A8-7600)

I think 10% higher clocks for 20 extra watts is worth it. This especially as the additional power required is still well within the realm of low end components.

Also your example of Kaveri at 45 watts vs. 65 watts doesn't even scale this well if going by base clocks (3.3 Ghz @ 65 watts is only 6.5% higher than 3.1 GHz @ 45 watts).*

We also have to remember having an unlocked multiplier allows easier overclocking of the iGPU and the scaling on that should be good to 1000+ MHz. (eg, R7 250 clocks @ 1000/1050 Mhz, R7 250X clocks @ 1000 Mhz, R7 260X clocks @ 1100 Mhz). If Carrizo iGPU clocks are similar to Kaveri Mobile iGPU clocks (which are up to 686 MHz for the 35 watt SKU and 553 MHz for the 19 watt SKU) that could amount to some good gains (particularly if DDR4 is part of the package).

*Although in all fairness it is further along on the scaling curve.

 

[CHADBOGA]

When I saw the thread title, I thought to myself, "Who the hell would vote NO", then I check the poll results and the NO Vote is winning.

 

[DrMrLordX]

Normally I'm a big fan of overclocking options and unlocked multipliers. But for reasons that have been stated, I think Carrizo is a poor candidate for this option. Also, we are in a special case in which AMD is struggling for survival. Missteps at this point could cause them a good bit of harm. For my own personal preferences, I'd much rather see everything unlocked, but putting myself in AMD's shoes, I can see why they would not want an unlocked Carrizo part at launch (and possibly not at all).

 

[ShintaiDK]

I cant see the purpose of unlocked parts in with those properties. Its just counterproductive.

 

[cbn]

Also, we are in a special case in which AMD is struggling for survival. Missteps at this point could cause them a good bit of harm.

I see things the reverse. Converting an locked multiplier SKU to unlocked multiplier SKU costs AMD nothing and increases the value of the Carrizo Inventory in the eyes of DIYers.

For my own personal preferences, I'd much rather see everything unlocked

I'm sure that is a common opinion that would be shared by almost any DIY shopper.

 

[Valantar]

I think 10% higher clocks for 20 extra watts is worth it. This especially as the additional power required is still well within the realm of low end components.

Also your example of Kaveri at 45 watts vs. 65 watts doesn't even scale this well if going by base clocks (3.3 Ghz @ 65 watts is only 6.5% higher than 3.1 GHz @ 45 watts).*.

Remember that those Kaveri clocks are probably at somewhere around 20W (for the 45W) and 30W (for the 65W) per core pair. And yes, that was my point exactly - even Kaveri doesn't really scale well above this power level. AMD is saying that Carrizo will be worse. If the previously posted graph is anything to go by, this architecture might see zero frequency gain past 25W per core pair, and the gains past 20W are negligible at best. Sure, they might make a 45W "refresh" SKU when the design has matured a bit. But for now, with AMDs survival as the focus, low power, low cost, mainstream chips need to be the only focus. They actually need to get some laptop design wins, unlike Kaveri, which is nowhere to be found. And releasing an unlocked chip which - at best - gives you a 10% overclock, that is a surefire plan to piss off as many DIYers as you can.

 

[Abwx]

Remember that those Kaveri clocks are probably at somewhere around 20W (for the 45W) and 30W (for the 65W) per core pair. And yes, that was my point exactly - even Kaveri doesn't really scale well above this power level. AMD is saying that Carrizo will be worse.

At full CPU load under prime 95 a 7850K comsume 67W with the cores using 50W, that s 25W per module, this is this latter case that is displayed on AMD s slide below, the vertical scale is normalized to 3.5GHz, at this frequency a Carrizo still has 10% better perf/watt than a Steamroller when including the IPC.

http://www.hardware.fr/news/14085/amd-leve-voile-carrizo-cote-technique.html

 

[cbn]

And releasing an unlocked chip which - at best - gives you a 10% overclock, that is a surefire plan to piss off as many DIYers as you can.

10% overclock @ 20 extra watts would be for the fastest SKU.....on the CPU side only*. But then we need to also factor in the effect of the unlocked multiplier on the iGPU as well.

If Carrizo clocks are anything like Kaveri mobile clocks, I would expect 50%+ gains to be possible for iGPU overclocking (even on the top 35 watt SKU)

Two things to consider:

1.) The availability of the DDR4 enabled Carrizo chips to better make use of the OC'd iGPU --> http://forums.anandtech.com/showpost.php?p=37224491&postcount=995

2. The power consumption of an APU with the iGPU overclocked to 800MHz to 1000 Mhz (or beyond) plus a pair of modules at 30 watts to 50 watts.

*and we still don't know if this represents the maximum CPU overclock possible (due to lack of data for a pair of modules beyond 50 watts).

 

[Valantar]

10% overclock @ 20 extra watts would be for the fastest SKU.....on the CPU side only*. But then we need to also factor in the effect of the unlocked multiplier on the iGPU as well.

If Carrizo clocks are anything like Kaveri mobile clocks, I would expect 50%+ gains to be possible for iGPU overclocking (even on the top 35 watt SKU)

Two things to consider:

1.) The availability of the DDR4 enabled Carrizo chips to better make use of the OC'd iGPU --> http://forums.anandtech.com/showpost.php?p=37224491&postcount=995

2. The power consumption of an APU with the iGPU overclocked to 800MHz to 1000 Mhz (or beyond) plus a pair of modules at 30 watts to 50 watts.

*and we still don't know if this represents the maximum CPU overclock possible (due to lack of data for a pair of modules beyond 50 watts).

Wow, you really are being as optimistic as possible about this, aren't you. So you believe that AMD is purposefully not launching a more powerful Carrizo SKU, even though it could do so at a reasonable power draw? That seems pretty crazy for a company barely managing not to lose money.

Also, "lack of data for a pair of modules beyond 50 watts" - so you're ignoring the fact that their scaling graph, on their own, curated, PR department-made slide, shows frequency gains basically leveling out at ~25W/module? Do you honestly believe that they would leave this detail in the slide if there was any chance of Carrizo scaling acceptably beyond 25W? Or do you believe that the graph will take a sudden upwards turn again at some point?

And yes, 10% at +20W is for the top SKU. Are you suggesting that they release an unlocked, lower clocked SKU? That would be the stupidest possible thing that AMD could do at this point. They need money, not budget DIY cred.

But sure, GPU OC, I'm with you on that. Thing is, you don't need an unlocked SKU for that. They could enable Overdrive on the chip in CCC, and do it through software. Easy. No hardware changes necessary, no new boards or hardware necessary, no extra expenses, just an extra prompt in software telling users that they're on their own if they brick the device. They could even make the software override the TDP constraints for DIY boards or laptops with decent cooling, making it so the chip only throttles with temperature, not power draw (as long as the device is connected to AC power). All this can be done in software, with no need for AMD to waste money on an unlocked SKU.

 

[cbn]

Are you suggesting that they release an unlocked, lower clocked SKU?

An unlocked multiplier on an ULV processor would be awesome.

http://forums.anandtech.com/showpost.php?p=37221372&postcount=14

Regarding small low profile heatsinks on Mini-ITX boards. A 15 watt unlocked Carrizo works with that. The heatsink can be the typically sized passive one at stock speeds (but with an added fan header on the board and provisions for mounting an optional case fan on top the heatsink for overclocking). Since overclocking a 15 watt processor to 65+ watts requires nothing beyond standard components price should not increase. One board design to handle both 15 watt and 35 watt unlocked SKUs, the only difference being the 15 watt board comes with a passive heatsink and the 35 watt one adds a fan.

http://forums.anandtech.com/showpost.php?p=37222600&postcount=26

I would also like to see a ULV SKU with unlocked multiplier as well (being able to run passive at stock would be awesome), but I am guessing for this to happen it would need to have the same price and core/iGPU config as the top 35 watt SKU.

 

[sm625]

Not everybody has seen a BGA board (usually there are in pre-built devices), but for anyone interested here are some DIY examples on Newegg:

http://www.newegg.com/Product/Produc...lt=True&page=1

The most common form factor for x86 is Mini-ITX with Micro-ATX coming in second place.

This is also a BGA motherboard:

They do not have to be small. It's too bad AMD doesnt just make boards like this one with fullblown APUs and DDR3 or even GDDR5. With the integrated FCH they can really cut down on the complexity and produce something fairly inexpensive.

 

[cbn]

But sure, GPU OC, I'm with you on that. Thing is, you don't need an unlocked SKU for that. They could enable Overdrive on the chip in CCC, and do it through software. Easy. No hardware changes necessary, no new boards or hardware necessary, no extra expenses, just an extra prompt in software telling users that they're on their own if they brick the device. They could even make the software override the TDP constraints for DIY boards or laptops with decent cooling, making it so the chip only throttles with temperature, not power draw (as long as the device is connected to AC power). All this can be done in software, with no need for AMD to waste money on an unlocked SKU.

Just being able to run an APU at CPU at base clocks with the iGPU @ stock speeds under load would greatly increase the utility of existing APUs. Of course, this increases power draw beyond TDP.

What you are mentioning further increases utility and power draw beyond the TDP spec. (Of course, I definitely like the idea).

So with these changes you have proposed, is having a unlocked multiplier in any way a cost adder?

The way I see things, with iGPU overclocking enabled and limits on power draw now removed the energy consumption potentially increases well beyond 35 watts. So now the board is going to need extra power circuitry/cooling just to handle the extra clocks from the iGPU. Therefore, at this point, I'm seeing the unlocked multiplier on the CPU as not only a zero cost adder from AMD's standpoint, but from the board manufacturer's standpoint as well. Maybe the only caveat is that max overclock on the CPU can't be used with the iGPU running overclocked at the same time (depending on how robust the board/cooling is). The user might have to restrict using the overclocked CPU with non-gaming loads only.

 

[cbn]

Regarding very small enclosures and overclocking, even this inexpensive 2.5L Mini-Box M350 is rated to 65 watts (for CPU coolers that have a fan). This without using any of its additional fan mounts for ventilation.

http://www.mini-box.com/M350-universal-mini-itx-enclosure

 

[cbn]

Regarding core configurations for the future DDR4 enabled Carrizo SKUs, here is what I am hoping to see:

1. quad core with 512sp
2. quad core with 384sp (for dies with defect in iGPU)
2. dual core with 384sp (for dies with defect in CPU module and iGPU)
4. quad core without iGPU (for dies with defect in display, media, etc.)

Any comments on the dual core config? (I am hoping AMD stops the practice of disabling so many iGPU units on the dual core SKUs. I think these weaker units need all the help they can get to make them worth more. 512sp should even be considered for dual cores without defect in iGPU.)

Any comments on the quad core without iGPU (ie, Athlon x4) config?

 

[Valantar]

JThe way I see things, with iGPU overclocking enabled and limits on power draw now removed the energy consumption potentially increases well beyond 35 watts. So now the board is going to need extra power circuitry/cooling just to handle the extra clocks from the iGPU. Therefore, at this point, I'm seeing the unlocked multiplier on the CPU as not only a zero cost adder from AMD's standpoint, but from the board manufacturer's standpoint as well. Maybe the only caveat is that max overclock on the CPU can't be used with the iGPU running overclocked at the same time (depending on how robust the board/cooling is). The user might have to restrict using the overclocked CPU with non-gaming loads only.

And here is exactly where the problem lies. As I hope has been made clear throughout this thread, OCing Carrizo's CPU cores would elicit HUGE amounts of increased power draw - close to doubling it for a puny 10% gain. In other words, yes, this would require strengthened power circuitry. OCing the iGPU on the other hand, would not come close to this level of power draw. Sure, it would increase, but bringing the whole package beyond 50W? I doubt it. And again, with a software implementation, this could easily be limited so that users don't fry their power delivery.

And yes, unlocking the top SKU would add costs, even if this was the only top-end SKU (as opposed to adding an unlocked one to the line-up). This due to OEMs having to factor in overclocking into their designs, making laptops and AIOs more expensive (due to needing bigger power supplies and other power delivery parts, cooling, and general added complexity in design). And in the end, the DIY ITX market is infinititesimally small. How many ITX boards would they sell, globally? A thousand? Two? No matter what, it would never, ever make them money, unless they priced them into oblivion. In which case, sales would be zero.



When it comes to your suggested core configurations, I'm completely on board. More large iGPUs, across the board please. As long as only the top configurations really work for gaming (which is the only real argument for these over Intel chips, as they have excellent video en-/decode and other functions), the lower specced ones are almost obsolete by design. I'd love it if the next generation AM1 chips had 256 or even 384sp, and of course lower clocked - cheaper - APUs with 512sp iGPUs would be a brilliant move.

 

[Shehriazad]

Why not both?

Carrizo BGA and
Carrizo K BGA.

Let their 2 fastest models be a K edition and the rest not....as simple as that.

Honestly I'm really hoping for the Carrizo BGAs to show even stronger iGPU power...because on that end no one can compete them in tthat price segment.

I mean you can match the price/performance of their APUs...but generally that means buying a dGPU and cheap CPU.

But if you want a tiny cube PC without additional GPU slot, then AMD seems to be the way to go unless you can afford going with Intels higher end chips that feature their acceptable iGPUs.




I don't see a real downside to making their strongest BGAs a "K" edition. leaves more room for people that do want to mess around with those chips...since at least I feel a lot more comfortable just messing around with unlocked multipliers instead of having to overclock the old fashioned way (which also affects the ram)

 

[cbn]

OCing Carrizo's CPU cores would elicit HUGE amounts of increased power draw - close to doubling it for a puny 10% gain. In other words, yes, this would require strengthened power circuitry. OCing the iGPU on the other hand, would not come close to this level of power draw. Sure, it would increase, but bringing the whole package beyond 50W? I doubt it.

Getting the iGPU to draw 50 watts all by itself would be easy.

Just look at the TDPs of these GCN Video cards:

HD7750: 512 GCN stream processors @ 800 MHz = 55 watts
R7 250: 384 GCN stream processors @ 1000/1050 Mhz = 65 watts

Now granted those TDPs include the VRAM and other components as well, but it isn't hard for me to imagine a 512sp Carrizo iGPU @ 1000+ Mhz drawing 65 watts. Even at a more reasonable 800 Mhz speed we are probably looking at ~45 watts for a 512sp iGPU. So to have this iGPU overclocking available (even at a modest level) means extra power and cooling is going to have to be there anyway. When the iGPU is not in use that extra power and cooling could be used for overclocking the CPU cores.

 

[DrMrLordX]

Why not both?

Carrizo BGA and
Carrizo K BGA.

Let their 2 fastest models be a K edition and the rest not....as simple as that.

Carrying an extra SKU introduces extra costs. You're talking about a separate validation process (unless they just go sloppy and toss in anything from the top non-K bin) costs, inventory costs, blah blah blah. It might only amount to a few mil, but really.

An unlocked 35W BGA part would probably not produce much interest. I personaly would love one, since I'm fairly certain the standard 35W Carrizo is going to be 1.7 ghz/2.1 ghz turbo (I could be wrong). If that guy could go all the way to 4 ghz then that would be beyond awesome, until I compared it to my FM2+ machine, but whatever. The iGPU would be better so it might be worth it.

In the end what you'd probably see is slow-to-non-existant adoption of the K part by laptop OEMs, so the K parts would sit around gathering dust, stuffing the channel up yet again. Once Carrizo went into AiOs and SFF/UCFF units, you'd see artificially discounted K parts coming from distributors sick of keeping them in inventory which would screw up AMD's ability to sell Carrizo parts at normal prices to the OEMs sticking them in cheap desktops-ish machines. That's a worst-case, mind you. AMD could sort of avoid that by holding back on the K parts for a little while.

It's possible that some oddball OEM would pick up an unlocked Carrizo for budget gaming laptops. It just don't see it as being a viable competitor for Intel + Nvidia dGPU laptops currently aimed at the gaming laptop market. Intel could probably use their clout to expand that segment if they really wanted to. AMD has to focus on making sure their products wind up in competently-designed, customer-friendly laptops. They don't have a lot of wiggle room for trying to expand niche markets and engage in other risky ventures.

Honestly I'm really hoping for the Carrizo BGAs to show even stronger iGPU power...because on that end no one can compete them in tthat price segment.

What you are going to see is less bottlenecking from limited memory bandwidth. If AMD pushes DDR3-2133 as a minimum in as many of the target OEM systems as possible, then that would be pretty great (though maybe not so much for battery life).

 

[cbn]

If AMD pushes DDR3-2133 as a minimum in as many of the target OEM systems as possible, then that would be pretty great (though maybe not so much for battery life).

On the desktop 2x 4GB DDR3 2133 usually carries ~$10 premium over 2 x 4 GB DDR3 1600 at Newegg, and on laptop it looks like the premium is $25.

I think that is going to be tough for AMD to ask for.

Besides, I have never seen an AMD desktop (from a major OEM) come with memory faster than DDR3 1600. Same with AMD laptops.

P.S. Looking for DDR4 rices to drop and for DDR4 3200 to become commodity. I would buy a DDR4 enabled Carrizo right now just because I know in the future the price will come down.

In fact, 4GB DDR 2400 is already $29.99 After rebate at Tiger Direct:

http://www.tigerdirect.com/applicati...48&CatId=11628

(Although with that mentioned, I am concerned.... after enough time passes by..... 2 x 4GB DDR4 vs. 1 x 8GB DDR4 is going to become 2 x 2 GB DDR3 vs. 1 x 4GB DDR3 all over again. That is the 1 x 8GB DDR4 configuration will eventually become cheaper than 2 x 4GB DDR4, and many folks and OEMs will be tempted to use DDR4 3200 single channel rather than DDR4 3200 dual channel.)

 

[cbn]

I'm fairly certain the standard 35W Carrizo is going to be 1.7 ghz/2.1 ghz turbo (I could be wrong).

Comparing the scaling we see on this slide,

to the 35 watt mobile Kaveri cpu clocks,

I'm thinking we should see slightly better than 2.7 Ghz base on the best 35 watt Carrizo SKU.

 

[DrMrLordX]

Don't count on it. Remember that Carrizo's TDP now carries the FCH along with it. The leaked 3DMark numbers were allegedly from a Carrizo laptop with speeds of 1.7 ghz/2.1 ghz turbo, though that's an ES leak so who knows what that really means for retail products.

If they can get 2.7 ghz (or higher?) on Carrizo then great, I'm just not gonna hold my breath waiting for it.

And neither you nor I know what notebook OEMs pay for SO-DIMMs, but on the DIY market DDR3-1600 carries practically no price advantage nowadays:

http://pcpartpicker.com/parts/memory/#s=301600,302133,302400&Z=8192002&sort=a10

$8 price delta between the cheapest DDR3-1600 and the cheapest DDR3-2400. Now, in SO-DIMM form, things are a little different:

http://pcpartpicker.com/parts/memory/#s=301600,302133,302400&Z=8192002&sort=a10&t=8,9

Much bigger delta, as you observed. Still, that $25 is retail. Take off the ~40% markup for retail and your price delta is now ~$18.

That's one of the areas where AMD has to put down their foot and say "use a non-bs BoM please". Even if the OEM doesn't give a darn how the thing really runs after its bought, AMD has to know that DDR3-1600 is a bottleneck for the iGPU while DDR3-2400 is less of one. Carrizo's nifty Tonga-like feature set should make DDR3-2400 even better. $18-$25 to transform what could be disappointment into delight could make all the difference.

 

[cbn]

Now, in SO-DIMM form, things are a little different:

http://pcpartpicker.com/parts/memory/#s=301600,302133,302400&Z=8192002&sort=a10&t=8,9

Much bigger delta, as you observed. Still, that $25 is retail. Take off the ~40% markup for retail and your price delta is now ~$18.

That's one of the areas where AMD has to put down their foot and say "use a non-bs BoM please". Even if the OEM doesn't give a darn how the thing really runs after its bought, AMD has to know that DDR3-1600 is a bottleneck for the iGPU while DDR3-2400 is less of one. Carrizo's nifty Tonga-like feature set should make DDR3-2400 even better. $18-$25 to transform what could be disappointment into delight could make all the difference.

Here is the current list of Kaveri notebooks on Newegg:

http://www.newegg.com/Product/Produc...0%282.20GHz%29

Not all the OEMs use dual channel, even at DDR3 1600 speeds

For example, only 1 out of 10 of the HP Kaveri notebooks had 8GB, the rest had 4GB. (Only two of these models were dual cores)

The sole Acer listing, a A10-7300 quad core, has 4GB.

Lenovo has its sole quad core model coming with 8GB, the two other dual cores listed have 4GB.

Dell has four Kaveri models listed, all quad core models, each having 8GB.

So in total eight quad core models are listed with 4GB (single channel) and only six have 8GB (dual channel).

 

[Abwx]

So in total eight quad core models are listed with 4GB (single channel) and only six have 8GB (dual channel).

Most if not all are dual channel, it s just that only one slot is populated with a 4GB stick.

 

[DrMrLordX]

That's what he meant. Even if the board supports dual-channel, if the OEM is too darn cheap/stupid to use a pair of DIMMs, it is running in single-channel mode. Only a select few users will notice the problem and pop in a second DIMM.

Again, it's up to AMD to lean on OEMs to cut the bs and start outfitting Carrizo laptops appropriately. Any OEM throwing in 1x4gb instead of 2x2gb is doing AMD a great deal of harm by showing off their products that way. If the laptops had dGPUs from AMD where AMD could control the on-device memory amount and speed, they wouldn't have to worry about the OEM's system RAM choice so much. But those iGPUs are a different story.