Haswell vs Kaveri

Alexko said:
They probably just used the latest Catalyst, which do not support Kaveri. Obviously it still works, but the poor performance should not come as a surprise.
Click to expand...
How would a driver not supporting Kaveri work on it at all? That doesn't make much sense to me. It could be there's still optimization to do, but I doubt you could just substitute a driver with no support and have it work.
 
Andrew Lauritzen said:
How would a driver not supporting Kaveri work on it at all? That doesn't make much sense to me. It could be there's still optimization to do, but I doubt you could just substitute a driver with no support and have it work.
Click to expand...

Not sure about what drivers one used, but this (used in other recent leak) suggests there might not be drivers yet:
129c.jpg
 
How do they say about this? :D Result of using unknown hardware on software (in this case GPU-Z) which doesn't support it yet...

I also think that if Catalyst doesn't support the hardware, it won't install at all
 
ToTTenTranz said:
It could be the rumoured testing platform with single-channel memory. It would explain the good CPU scores and abysmal GPU scores.
Click to expand...


From the leak with gaming benchmarks the tester wrote DDR3-2133 8GBx2....which implies Dualchannel.


Kaotik said:
Not sure about what drivers one used, but this (used in other recent leak) suggests there might not be drivers yet:
Click to expand...


This screenshot is from a different leak with just a Cinebench R15 run, it has nothing to do with the recent gaming benchmarks. As for the drivers there are drivers with Kaveri support available, here the first one: http://www.rage3d.com/board/showthread.php?t=34001853


Otherwise he couldn't run the benchmarks. Also he could be a beta tester for some OEM, they should have drivers available.
 
Right, yeah GPU-Z doesn't actually query all of that data from hardware or anything (and in fact sometimes it gets it wrong or inconsistent), it basically just has a database of device IDs for a lot of it. Unrecognized device ID = "unknown"/zeroed fields.
 
Andrew Lauritzen said:
Right, yeah GPU-Z doesn't actually query all of that data from hardware or anything (and in fact sometimes it gets it wrong or inconsistent), it basically just has a database of device IDs for a lot of it. Unrecognized device ID = "unknown"/zeroed fields.
Click to expand...

Didn't mean the 0's etc, but rather the name which to my understanding comes from drivers and/or hardware rather than database?
 
Andrew Lauritzen said:
How would a driver not supporting Kaveri work on it at all? That doesn't make much sense to me. It could be there's still optimization to do, but I doubt you could just substitute a driver with no support and have it work.
Click to expand...

I would assume that drivers fall back to some kind of basic, poorly optimized support.

In any case, Trinity/Richland has 6 VLIW4 compute units, while Kaveri has 8 GCN ones. We know from numerous discrete graphics benchmarks that the latter ought to be much faster. At worst, it could be too limited by memory bandwidth to significantly pull away, but there's no reason for it to be slower, as it is in a few benchmarks here.

To me, this points to poor drivers, which is to be expected when someone gets their hands on an engineering sample they shouldn't have.
 
Alexko said:
At worst, it could be too limited by memory bandwidth to significantly pull away, but there's no reason for it to be slower, as it is in a few benchmarks here.
Click to expand...

The A10-6800k has a maximum GPU clock of 844mhz, the A10-7800 720mhz.
That is a reasonable explanation for the deficit on some benchmarks.
 
Kaotik said:
Didn't mean the 0's etc, but rather the name which to my understanding comes from drivers and/or hardware rather than database?
Click to expand...

Yeah, kind of running Kaveri on the Windows Microsoft's driver? :LOL:

No, it could be just a fault on the GPU-Z side of things. TPU should be able to explain more why it shows 'Microsoft' there
 
Where are the slides that show raw performance and not just perf/watt? I think those rumours about it being only marginally better than Richland could be true.
 
Wynix said:
Where are the slides that show raw performance and not just perf/watt? I think those rumours about it being only marginally better than Richland could be true.
Click to expand...

When the said chips are running at 45W/65W/95W perf/watt difference is same as performance difference.
It's not like "65W Richland" runs at 65W and "65W Kaveri" at 50W or something
 
I like what Mantle does for gaming on Kaveri!
It's also funny that now cryptocurrencies are important enough to appear on upcoming APU slide deck :smile:

All in all I like perf/watt improvements and will upgrade my HTPC to top Kaveri from my good old A10-5800K.
Shame my motherboard is too old to support FM2+ CPU's but it served me well for over 1 year now so all is good.

One more thing I noticed is just how much denser transistors are packed in Kaveri compared to Trinity!
246mm2 32nm SOI - 1.3b transistors VS 245mm2 28nm Bulk - 2.4b transistors
 
Lightman said:
One more thing I noticed is just how much denser transistors are packed in Kaveri compared to Trinity!
246mm2 32nm SOI - 1.3b transistors VS 245mm2 28nm Bulk - 2.4b transistors
Click to expand...

You know that with these 85% more transistors you should get relatively comparable jump in performance as well...
That's a very huge jump in transistors count, I am impressed! :oops: SOI sucks big way :LOL:
 
You must log in or register to reply here.
Back
Top