Yea the thing that everyone would like to know is,
If ZEN at the end of 2016 will be performance or Energy or Price or any combination competitive against Kabylake at the time of release.
I think it's clear, Shintai doesn't want to downgrade his current build with Zen.
As for me, I want 6700K performance cheaper then I could upgrade my other systems with Zen. If AMD is targeting IB/HW level of performance, it's a fail, imo. At the very least it should be able to beat i3 6100 in 4T workloads to be even considered in budget builds. Yeah, 2016 is around the corner.
It seems to be more likely as new info is becoming available that AMD seems to be designing Zen for the server market mainly with TCO as the prime design driver. Less likely that Zen will have an outright performance win/thread. The enthusiast segment is quite small and with the apparent brand loyalty as seen here, not worth pursuing relative to the needed effort.Yea the thing that everyone would like to know is,
If ZEN at the end of 2016 will be performance or Energy or Price or any combination competitive against Kabylake at the time of release.
It seems to be more likely as new info is becoming available that AMD seems to be designing Zen for the server market mainly with TCO as the prime design driver. Less likely that Zen will have an outright performance win/thread. The enthusiast segment is quite small and with the apparent brand loyalty as seen here, not worth pursuing relative to the needed effort.
Very small cores [lower cost + this will also be a benefit for APUs ]
Power efficiency [Techniques used with Carrizo mobile on 28nm is impressive]
I have agreed with you on that part, however my question is: how it's possible that upcoming top mobile APU (FX-9830P?) have so much higher base clock than FX-8800P (3.0 GHz vs 2.1)?
Keep in mind that in case of Carrizo / Bristol Ridge there is no such thing as the "base clock". The clocks can drop below the "base clock" level when and if the power limit is exceeded.
After reading in this thread about the unlikeliness of AM4 boards in March, how does the appearance of MYRTLE testing boards and BR samples in Oct and Nov affect this?
Interesting, so base clocks and TDP have no actual meaning with AMD chips. This is FX8350 on steroids.
Google "2d34e2agm44ab" and thou shalt see.I'd like to know more about that. If AM4 + Bristol Ridge show up in March, great! But AMD is apparently doing a GPU launch in June/July, so that would be the time when I expect them to dump new products on the market, sort of like they did in 2015.
That being said, I'd like to be wrong when AMD does launch Bristol Ridge in March.
That's not what I said, I said IB/HW-E performance regardless if that is achieved through higher IPC or higher clocks. But if you're assuming IB/Haswell IPC, then you should be comparing to 5960X which is 3.0 Ghz, not 2.5 Ghz.8 cores at 2.5Ghz with IB/HW IPC and 95W, no I dont want it. I wouldn't want it at 3Ghz either for that matter.
I think it's your memory that is failing you. Or you're misinterpreting conclusions about a potential outcome based on assumptions on this forum with my own expectations.If your memory is that short, that you forgot your own statements...sure.
That's not what I said, I said IB/HW-E performance regardless if that is achieved through higher IPC or higher clocks. But if you're assuming IB/Haswell IPC, then you should be comparing to 5960X which is 3.0 Ghz, not 2.5 Ghz.
But since you said you wouldn't want it at 3.0 GHz either, you're basically saying you'd not buy the 5960X at $600, even if it was 95W TDP? Is it because it would have AMD stamped on it instead of Intel?
Did you even read what I wrote? I said IB/HW-E performance. That is not faster than Broadwell-E, is it?8 Core Broadwell is 3.3Ghz at 140W. You somehow expect a complete miracle from AMD beating Intel in all metrics. You are way deep into your hyperbole.
You don't need to OC to get better MT performance from a 5960X.Also there is the thread about what would you pick if it was free. I picked a 6700K over a 5960X because I dont OC.
Again, I said IB/HW-E performance, not Broadwell-E performance. So no, I don't expect Zen to perform at the same level as Broadwell-E.Broadwell-E is 14nm like Zen will be. What Intel needs 140W for at 3.3Ghz, you expect AMD to need 95W on a crappy 14LPP node that even Apple wont use with A10. And even better, it will be done at GloFo.
Ok, then that's because of your specific use cases. You're basically only interested in ST performance. And thus you are neither interested in Intel HEDT nor Zen. So that you're not interested in buying Zen does not really say much about it's potential among customers looking for that kind of CPU.And I cant use the extra MT performance to anything. Hence I picked the 6700K over the 5960X.
Google "2d34e2agm44ab" and thou shalt see.
No I need 4 cores with high ST, I just dont need 6 or 8.
I see that now. Hmmmmm! I'm a good bit more optimistic about the March launch now. AMD really does need AM4 out ASAP to keep their lineup fresh (and whet appetites for Zen). A March launch would be a great thing . . .
Here are the CPU & GPU clocks I´ve measured on FX-8800P during GTA V benchmark (at default 15W base and 25W boost TDP).
Is that guaranteed 2.1GHz / 800MHz?
Do you have some performance estimates of what we could expect from a A10-7870K with GDDR5?
Based on the tests I´ve made in the past the A10-7870K (8 CUs & 2 RBs @ 866MHz) would have ~33% performance benefit in GPU performance from 6Gbps GDDR5 (with 128-bit bus). At 4Gbps the benefit is already ~24%.
HBM2 appears to be the only way for AMD to achieve better iGPU performance. They can add more units to their iGPUs but it is nothing but waste of power and die area. The bandwidth available from the 128-bit DDR3 / DDR4 bus in completely inadequate to keep even their current iGPUs fed at lower clocks than they currently are available. The new framebuffer compressions algorithm introduced in Carrizo (originally in Tonga) definitely helps, but even with it the bandwidth is simply inadequate even for the current generation iGPUs.
4x x64 DDR4 would be the most obvious choice (other than HBM2 or on-die cache for FB), however it is very likely too expensive to implement on platforms of this segment. It would significantly increase the motherboard complexity and also increase the die size and the package size of the final product quite alot (more pins / bumps required). Given how much bandwidth even the current generation APUs are lacking, the quad channel DDR4 wouldn´t be too long term solution either.
Triple channel is not unheard of; 192 bit bus and >3GHz DDR4 would be a nice improvement.