I know this. I thought DDR4 drops (as GDDR5 does) the burst chop mode of DDR3 which halves the access granularity (at some expense of bandwidth efficiency as the second part of the full length burst is simply masked). But it obviously still exists also in DDR4. But halving the granularity means that while it may be often tolerable with a 128bit controller (64byte granularity equalling the cache line size, but most situations probably prefer the individual controllers which is also the setting AMD recommends), but with 128Byte granularity for a 256Bit controller performance would take a dive in basically all use cases.
Haswell vs Kaveri
- Thread starter AnarchX
- Start date
Blazkowicz
Legend
This mobo is dealt with here, with info (in French) about fun USB and audio stuff
http://www.hardware.fr/news/13297/cartes-ma-res-fm2-gigabyte.html
It has an isolated USB port to supposedly not let go interferences go into the USB power supply for an external sound card or FAC, and isolated on-board sound card with op-amp.
Else, it's just an overpriced mobo.
FM2 CPUs go in it and the news let know it's apparently launched before Kaveri is available
http://www.hardware.fr/news/13297/cartes-ma-res-fm2-gigabyte.html
It has an isolated USB port to supposedly not let go interferences go into the USB power supply for an external sound card or FAC, and isolated on-board sound card with op-amp.
Else, it's just an overpriced mobo.
FM2 CPUs go in it and the news let know it's apparently launched before Kaveri is available
Charlie has an article on the Gigabyte mobo's, he reckons that their decision to go with high end mobo's is something that "things should be read into". Obviously he thinks that Kaveri is going to be a cut above the recent AMD CPU's but I'll reserve judgement on that until release I think.
I do believe Kaveris will end up above Richland, perhaps by a slightly more extensive slice than Trinity ended up above Llano (no, the Richland / Trinity thing is not accidental). However that is not saying much because that bar is still quite low, all things considered. I would be somewhat leery of reading much into motherboard makers releasing overpriced...err...I mean super-enthusiast boards, as those are pretty bad predictors of where the hardware ends up. Let us not forget that a Crosshair V (and a follow-up to that) exists, and none of the FXs were truly high end stuff.
Blazkowicz
Legend
What I wonder is.. can Kaveri beat a Phenom II X6 1100T? 
Dunno. I bet it beats Phenom II X4 and FX 4300, hopefully. Hoping it fares well against i3 3220, i3 Haswell and maybe i5 2300.
Badly needed is a triple module/six core Kaveri variant if they want to claim semi-high end status and consumer multimedia content creation.
Dunno. I bet it beats Phenom II X4 and FX 4300, hopefully. Hoping it fares well against i3 3220, i3 Haswell and maybe i5 2300.
Badly needed is a triple module/six core Kaveri variant if they want to claim semi-high end status and consumer multimedia content creation.
What I wonder is.. can Kaveri beat a Phenom II X6 1100T?
Dunno. I bet it beats Phenom II X4 and FX 4300, hopefully. Hoping it fares well against i3 3220, i3 Haswell and maybe i5 2300.
Badly needed is a triple module/six core Kaveri variant if they want to claim semi-high end status and consumer multimedia content creation.
Richland is already there for games, but it's almost 30% behind in productivity applications. I think 30% faster is what Kaveri ought to be; not necessarily what it will be, but what it ought to.
Architecturally it surely is much better than BD/PD in terms in single threaded performance per clock, but the worry is the clock part with the shift to 28nm bulk, and relying on globalfoundries to get their end of the bargain right.
As I said before there are an awful lot of things that can go wrong here, and some already have and some might be going wrong right now. I've learned my lesson regarding AMD cpu's though, so anything better than current APU performance is just a bonus and I'm neither hoping for or expecting much more.
As I said before there are an awful lot of things that can go wrong here, and some already have and some might be going wrong right now. I've learned my lesson regarding AMD cpu's though, so anything better than current APU performance is just a bonus and I'm neither hoping for or expecting much more.
Because of the new process, I too worry that it might have trouble reaching the kind of clock speeds that would be necessary for a major improvement over Richland, at least as far as the 100W desktop variants are concerned. Reaching 4.4GHz is no trivial matter.
25W notebook chips, however, are clocked no higher than 3GHz, and that's where I would expect (or hope for) Kaveri to really shine.
Blazkowicz
Legend
Will AMD go on making a single?, i.e. consider the 45nm gen, they had the Athlon II X2 die, the Athlon II X4 die and the Phenom II X4 die (then the X6 one).
Now a single Kaveri die will have to serve all desktop users.. Though I expect them to just continue selling Vishera into 2014 (till when?) and Kabini is here. But for low end, a single module Kaveri would run the web browser faster than a quad core slow Kabini..
The current low end Trinity/Richland sucks, and it's a huge die with most of it disabled, dunno the economics of that.
Now a single Kaveri die will have to serve all desktop users.. Though I expect them to just continue selling Vishera into 2014 (till when?) and Kabini is here. But for low end, a single module Kaveri would run the web browser faster than a quad core slow Kabini..
The current low end Trinity/Richland sucks, and it's a huge die with most of it disabled, dunno the economics of that.
Well they cancelled "true" dual core Trinity's I believe. I guess the problem for AMD is being able to afford the design costs for CPU's that are going to lose them money anyway, and that's why there has been so many cancellations.
Trinity will have ~230 good die per wafer (wafer costs are probably less than $2K-$3K per wafer now) but if you assume the cheapest chips sell more, many of them must be getting crippled for those. It's not a great situation to be in.
I wouldn't be surprised if they attempt to see out the Bulldozer fiasco with Steamroller, then a Steamroller refresh. Jaguar can hold the low end fine, up to 25W. I don't think we'll see another chip without GPU in it.
Trinity will have ~230 good die per wafer (wafer costs are probably less than $2K-$3K per wafer now) but if you assume the cheapest chips sell more, many of them must be getting crippled for those. It's not a great situation to be in.
I wouldn't be surprised if they attempt to see out the Bulldozer fiasco with Steamroller, then a Steamroller refresh. Jaguar can hold the low end fine, up to 25W. I don't think we'll see another chip without GPU in it.
That is rather optimistic, from what I have seen it is rather aligned with their 15% per gen hopes, and there isn't any huge architectural overhaul. The 30% figure being bandied around is derived from misinterpreting Papermaster's Hotchips claims.
So it won't be able to get at least 1.4 single threaded in cinebench 11.5?
That is the minimum i was expecting, my current Phenom 2 gets 1.15.
The single greatest improvement for Steamroller core will be the split of the per-module decoder in two. And such a change a looks to target multi-thread performance rather than single thread, IMO.
Hence I would expect Steamroller to look more favorable in multi threaded comparisons rather than single thread.
Hence I would expect Steamroller to look more favorable in multi threaded comparisons rather than single thread.
I was already expecting some multi-threaded improvements but without a large IPC improvement there is no way i can justify buying AMD for my next upgrade, and the lacklustre chipsets AMD have been putting out make it an easy choice.
Also i think this thread might need to be updated to Haswell Refresh vs Steamroller now that we know about Broadwell delay, Kaveri arriving Q1-2014 instead of late 2013 and we know nothing about the future FX line other than it might be scrapped.
Last edited by a moderator:
/ all that is so depressing, still look like a waste of great engineering teams to me
Agreed on the second part. On the first, you really think on the first they are not going to make any "big cores" <= 25W? That seems a little risky since I'm not sure how well Jaguar will compete with the big cores in the 15-25W range. Does anyone have numbers on this? Certainly in that range it starts to look a little weak on single-threaded and SIMD performance, but I guess it depends on what people are doing with those machines I guess.
The odds of no ~17W Kaveris making it out (eventually) are slim to none. Whether they will manage any sort of quantities or be stuck with the trickle they have had with Trinity, that is another question. How will Kaveri look in a gimped form? Who knows!
Here's the only review of the A6-5200 I know about - http://adrenaline.uol.com.br/biblioteca/analise/784/amd-a6-5200-kabini.html
It looks to be very well placed against the 100W APU's in perf/W terms. Note that it can drive high end GPU's at a similar level of bottleneck as the desktop A10's, where Brazos falters quite badly in Tomb Raider.
Comparison CPU scores vs the sub 25W Richlands are hard to find, but vs the 35W A10-5750M it scores 1.99 vs 2.28 in Cinebench 11.5, so it won't be far away to a 25W Richland at least. My guess would be losing single thread by 20% while winning multi thread by something similar.
The big difference is in graphics where the 15W A4-5000 Kabini struggles to be half as fast as the 35W A10, but with 1/3rd the SP's and single channel memory that's still pretty good tbh. The A6-5200 has 600MHz clocks vs 500MHz on the A4-5000. Battery life is undoubtedly far superior and that's with the southbridge included. It all adds up to what we already know, Bulldozer is a piece of shit, though Richland improved a few areas.
I'd imagine it'll basically like AlexV says and there will likely be a trickle of sub 25W big cores but nothing much to get excited about. For AMD it makes a lot more sense to use Kabini here.
Last edited by a moderator:
Good summary, thanks jimbo/Alex. Agreed with your analysis. I'm expecting AMD to try and drive the "it's like the PS4" angle as well marketing-wise; should be interesting!