Re: AMD - Bulldozer Chitchat
Quote:
Originally Posted by
scaryjim
The Anandtech write up of the scheduling issue makes it fairly clear that threads that depend on each other for data will run more efficiently if scheduled on the same module: don't forget that each module has two complete integer cores with a shared L2 cache, so if the data from one thread can go into that L2 cache then be picked up by the second thread from L2 cache that's a big saving on trips out to L3 cache (with increased clock cycle latency on BD) or main memory. Also, the increased clock speed available due to gating idle cores not only pushes more instructions through but also helps to mask the higher cache latencies...
As to the FPU layout, as far as I can tell BD has no more FPU capacity than a 4 core Phenom II for a lot of floating point workloads. It needs code that is heavily optimised to really shine. It is, sadly, in many ways a processor that is ahead of its time in relation to the software environment it finds itself in. On the plus side, the fact that Microsoft have already written a scheduler that is better optimised towards BD suggests (I'm making wild assumptions here, btw, feel free to correct me ;) ) that it might be possible to provide either a patch or a kernal mode driver that will allow Windows 7 to make the same optimisations, so we could see those scheduler-related performance boosts sooner rather than later...
This is what gets me, i have read through a couple of reviews and unless i have missed it in my tiredness, i dont think there is any real mention of an official patch for win 7.
People have compared the release to that of the first dual core and the threads jumping around. now that's before i was heavily into tech so i don't really remember that but how long after they released the first X2 did they wait to announce/release a patch?
Also, given how poor the per core performance really is, is it actually going to make THAT much difference if the scheduler issue is resolved? I mean are we talking a 20% increase across the board, small 2-5% increase across the board or (and this would be my prediction) are we talking generally small increases (2-5%) then in a select few niche benchmarks we see a bit more (10-12%). Still a chip based on an architecture with a long way to go.
Re: AMD - Bulldozer Chitchat
Quote:
Originally Posted by
Biscuit
Also, given how poor the per core performance really is, is it actually going to make THAT much difference if the scheduler issue is resolved? I mean are we talking a 20% increase across the board, small 2-5% increase across the board or (and this would be my prediction) are we talking generally small increases (2-5%) then in a select few niche benchmarks we see a bit more (10-12%). Still a chip based on an architecture with a long way to go.
Toms Hardware looked into as part of their review:
http://www.tomshardware.co.uk/fx-815...-32295-23.html
Situations which are multi-threaded are not going to see a difference. However,in situations which are more lightly threaded it will make more of a difference.
An example is WoW.
http://media.bestofmicro.com/N/F/310...wow%201680.png
http://media.bestofmicro.com/N/G/310...wow%202560.png
Framerates at higher resolution showed a decent increase of around 10% so I suspect the low resolution testing may see a bigger increase too.
Core parking also seems to work so at idle power consumption dropped by 10% too.
However,the main issue is whether Microsoft will release a patch for Windows 7. A few websites have mentioned this but until it comes from the horses mouth current performance is where it is. Since Vishera and Trinity and evolved versions of Bulldozer AMD do need to do something about it as they will be probably affected by similar issues.
Re: AMD - Bulldozer Chitchat
Quote:
Originally Posted by
scaryjim
As to the FPU layout, as far as I can tell BD has no more FPU capacity than a 4 core Phenom II for a lot of floating point workloads.
Yes, and this is the bit I don't get. Run as a single core it should have better issue width, improved branch prediction, more cache. The FPU is largely the same, but with better avx support (and the fma instruction that currently nothing uses).
Yet at a faster clock speed it is beaten as a single core by the Phenom II. They must have a problem here, lets hope they fix it fast.
Re: AMD - Bulldozer Chitchat
Quote:
Originally Posted by
DanceswithUnix
Yet at a faster clock speed it is beaten as a single core by the Phenom II. They must have a problem here, lets hope they fix it fast.
I cant see a scenario where releasing it now and getting the **** beaten out of it in benchmarks is preferable to delaying "another" month and releasing it fixed. It makes no sense at all so i can only conclude its perma-busted.
Re: AMD - Bulldozer Chitchat
All teething troubles I'm guessing - AMD would have scrapped the idea early on if it didn't show potential. It's just something quite new and different so I'd expect it to take a little while for programs to start fully taking advantage of the architecture - look at multi-core, how long did it take for games to go multi-threaded? And then that was helped by the fact the current consoles have pretty terrible single threaded performance vs a PC so devs were forced to program to properly utilise the resources. Let's hope MS do the sensible thing and release a patch for at least 7. It's a bold move by AMD to release such a radically different architecture but everyone just keeping things the way they are for compatibility doesn't do much for progress!
Quote:
Originally Posted by
viet cong zombi
HAHAHA IT WAS SUPPOSED TO BE A HORSE, A BULL IN FACT,,,
LET IT BE KNOWN AS.... girbilDOZER
I LIKE AMD BUT HEY, AN APU COULD NEVER RIVAL AN ENTHUSIAST CPU+GPU. ITS A STREAM CPU THAT DOENST REALLY STREAM VERY MUCH!!
Lower case, please! Caps are slower to read and make you looks like an angry Youtube commenter. ;)
This isn't an APU BTW. ;)
Re: AMD - Bulldozer Chitchat
Quote:
Originally Posted by
watercooled
Lower case, please! Caps are slower to read and make you looks like an angry Youtube commenter. ;)
This isn't an APU BTW. ;)
definitely seen this guy before on fudzilla or semiaccurate, or maybe both! i wouldn't hold out much hope of anything changing.
Re: AMD - Bulldozer Chitchat
Quote:
Originally Posted by
MadduckUK
I cant see a scenario where releasing it now and getting the **** beaten out of it in benchmarks is preferable to delaying "another" month and releasing it fixed. It makes no sense at all so i can only conclude its perma-busted.
Cashflow? There's people who will buy it regardless, including at least one poster in this thread ;)
Re: AMD - Bulldozer Chitchat
Hey I'm happy with Thuban! :D
Re: AMD - Bulldozer Chitchat
I just don't understand AMD's vision with Bulldozer, not because it doesn't beat Sandy bridge; but more the fact it's not much better than Thuban.
Then the Power usage is CRAZY
Re: AMD - Bulldozer Chitchat
Re: AMD - Bulldozer Chitchat
Quote:
Originally Posted by
MadduckUK
I cant see a scenario where releasing it now and getting the **** beaten out of it in benchmarks is preferable to delaying "another" month and releasing it fixed. It makes no sense at all so i can only conclude its perma-busted.
What I don't understand is that they have limited 32NM production due to limited yields and then launch this CPU for retail?? Wouldn't it have made more sense for them to get more Llano CPUs out and also drop the price on these and concentrate on Interlagos only. There was noise of an unlocked A8-3870 and one of those for around £90 to £100 would have been far more interesting for many people.
Re: AMD - Bulldozer Chitchat
Supposedly they are trying to get a B3 revision out the door asap...
Re: AMD - Bulldozer Chitchat
Quote:
Originally Posted by
shaithis
Supposedly they are trying to get a B3 revision out the door asap...
Why? Is there a bug?
Re: AMD - Bulldozer Chitchat
Quote:
Originally Posted by
shaithis
Supposedly they are trying to get a B3 revision out the door asap...
Source?
Re: AMD - Bulldozer Chitchat
In my view its not as bad as I initially thought.
Still tempted to get 8150 once prices settle at little.
Unless there is more to this 'B3 revision' rumour?
Re: AMD - Bulldozer Chitchat
Quote:
Originally Posted by
kalniel
Why? Is there a bug?
Did you not see the review? :P
No, in all seriousness i expected this. Given how often the 8170 was mooted then suddenly disappeared i assumed it would come along with a new stepping, not saying it will come with the suggested new stepping of course.
The only obvious bug i can think of outside of the cache "issues" [seems a more important fix than a stepping revision surely?] is the odd power draw @ stock when under the kosh.