baccilus
Cyborg Agent
Even Phenom II X4 can max all the games out there. That won't change till the next generation consoles come out. As long as they sell it for cheap and it is better than Phenom II, I will be happy.
Last edited:
AMD Bulldozer News and Discussion
Even Phenom II X4 can max all the games out there. That won't change till the next generation consoles come out. As long as they sell it for cheap and it is better than Phenom II, I will be happy.
If it performs anywhere closer as reported by the leaks, then AMD have done a good enough job. It will mean Trinity will become simply irresistible. Coming back to BD, I am a bit disappointed with the 990 FX/X chipset - I was expecting more.
Well I agree that more and more are closer to using 4 Cores as of now. But 6 & 8 will take more time. By the time games start supporting those a whole new league of Proccys will be out
I tend to agree with ico.
AMD's "Pure" CPU power is getting over powered by Intel so they are getting the Fusion & APU thing going so as to compensate for that.
AMD = Entry Mid Level (i.e masses) and Mid High + High will go with Intel.
It's not that you upgrade your CPU whenever there are some new releases. So when games will start using more than 4 cores, you "older" 6/8 core CPUs will just delay the inevitable upgrade, saving you some dough. So having more cores, albeit somewhat lower performing, at a similar price point is not bad at all.
vickybat
I am the night...I am...
^^ It has to be seen how much lower performing bulldozer will be compared to its competition. I think the fight won't even go to socket 2011 based sandybridge-E, but 1155 will be enough.
Remember that more cpu's are going to role out from the 1155 stable. That means 2600k might get a big brother.
Remember that more cpu's are going to role out from the 1155 stable. That means 2600k might get a big brother.
BD will ultimately more or less match SNB. Even if i7 2700/2800 comes, AMD is also keeping the option open for a speed bump. So they looks more or less even. SNB-E would be a different beast altogether, it would be interesting to see how AMD is going to counter that.
If it performs anywhere closer as reported by the leaks, then AMD have done a good enough job. It will mean Trinity will become simply irresistible. Coming back to BD, I am a bit disappointed with the 990 FX/X chipset - I was expecting more.
It's not that you upgrade your CPU whenever there are some new releases. So when games will start using more than 4 cores, you "older" 6/8 core CPUs will just delay the inevitable upgrade, saving you some dough. So having more cores, albeit somewhat lower performing, at a similar price point is not bad at all.
^^ It has to be seen how much lower performing bulldozer will be compared to its competition. I think the fight won't even go to socket 2011 based sandybridge-E, but 1155 will be enough.
Remember that more cpu's are going to role out from the 1155 stable. That means 2600k might get a big brother.
^^ or 2500k might get a younger brother but I'm sure Intel is delaying it for Final BullDozer cpu release.
For me ( as a gamer's perspective ) IF Bulldozer can give 60-80+ FPS on most current gen games with a decent GPU like say HD6970/GTX580 and if it's reasonably priced then I think it's enough as games are getting more GPU power hungry than CPU it's so good to invest more in a good GPU and if possible in multi gpu config
^^ or 2500k might get a younger brother but I'm sure Intel is delaying it for Final BullDozer cpu release.
For me ( as a gamer's perspective ) IF Bulldozer can give 60-80+ FPS on most current gen games with a decent GPU like say HD6970/GTX580 and if it's reasonably priced then I think it's enough as games are getting more GPU power hungry than CPU it's so good to invest more in a good GPU and if possible in multi gpu config
If I am not mistaken, to drive that GPU/s to its optimum potential a powerful CPU is necessary.
But Fermi(480) did take the single GPU crown from 5870. Only that the temps and noise were unbearable and the gap was not much to warrant any changes.
dont know why i feell this way but it appears that something is really rong with intels haswell /rocwell thats why AMD is releasing buldozer wth sandy bridge -E to lurk in the shadows to strike haswell.
AMD's Bulldozer-based FX-8130P benchmarked early - TechSpot News
*en.wikipedia.org/wiki/Bulldozer_(processor)
AMD's Bulldozer-based FX-8130P benchmarked early - TechSpot News
*en.wikipedia.org/wiki/Bulldozer_(processor)
OP
Cilus
laborare est orare
The release of Bulldozer is due to the performance issue but the general assumption that the performance is below the current generation Sandybridge processor is may not be true. It is highly possible that AMD has set their minimum threshold to a very high level to make everything correct after a long time.
One interesting thing, we were discussing in the ASUS Power user meetup, rather than 8 cores or 6 cores of Bulldozer, just consider the 8 core Bulldozer as 4 modules which is having almost 2 Physical cores, i.e. discrease execution units for most of the cases apart from the FPU and Fetch unit.
Intel 2600K is also a 4 module based design where each of the module has 2 logical cores but only one set of execution units. It has two thread register to hold two threads simultaneously.
Now in a single threaded applications, which has been a very strong point for Intel now, HT won't help and all the instructions present in the thread will be executed by one core with all its execution unit. It may be improved by out of Order execution, Pipelining, superscalar processing, but remember, inside a single core.
Now come to the case of a Bulldozer mudule. Here a single module is having two independent set of execution units. Now if a single thread is assigned to a BD module, the shared double Fetch and Dispatch unit will fetch the istructions present in that thread to the BD module as it is shared by both the cores. So the istructions present inside a single thread can be executed parallelly by two cores of a BD module. Suppose total number of instructions inside Thread T is 2N. just consider each of the core is processing N number of instructions and again N instructions for a core can be benifited rom the OOO, Supersacalar and Pipelining.
So actually even if a single thread is assigned to a BD module it can still use both the cores to implement a ILP which is not in the case of Sandybridge where ILP is possible only inside the one execution core and HT is not gonna help.
That's why in the single threaded Super PI, I guess bulldozer is performing better.
One interesting thing, we were discussing in the ASUS Power user meetup, rather than 8 cores or 6 cores of Bulldozer, just consider the 8 core Bulldozer as 4 modules which is having almost 2 Physical cores, i.e. discrease execution units for most of the cases apart from the FPU and Fetch unit.
Intel 2600K is also a 4 module based design where each of the module has 2 logical cores but only one set of execution units. It has two thread register to hold two threads simultaneously.
Now in a single threaded applications, which has been a very strong point for Intel now, HT won't help and all the instructions present in the thread will be executed by one core with all its execution unit. It may be improved by out of Order execution, Pipelining, superscalar processing, but remember, inside a single core.
Now come to the case of a Bulldozer mudule. Here a single module is having two independent set of execution units. Now if a single thread is assigned to a BD module, the shared double Fetch and Dispatch unit will fetch the istructions present in that thread to the BD module as it is shared by both the cores. So the istructions present inside a single thread can be executed parallelly by two cores of a BD module. Suppose total number of instructions inside Thread T is 2N. just consider each of the core is processing N number of instructions and again N instructions for a core can be benifited rom the OOO, Supersacalar and Pipelining.
So actually even if a single thread is assigned to a BD module it can still use both the cores to implement a ILP which is not in the case of Sandybridge where ILP is possible only inside the one execution core and HT is not gonna help.
That's why in the single threaded Super PI, I guess bulldozer is performing better.
Last edited:
MatchBoxx
the m0nk who sold his PC!
J
Joker
Guest
i will explain in simple words.
a normal core has one integer schedular and one floating point schedular.normally a core handles one thread.
a bulldozer module has two integer schedulars and one floating point schedular.
intel's hyper threading "forces" two threads through a core which has integer schedular and one floating schedular. hyperthreading in real world applications gives u 10-15% boost. synthetic benchmarks may show u 50-60% boost but that doesnt matter.
a bulldozer module will handle two threads and both threads will get their own integer schedular but will share the floating point schedular.
people think, bulldozer will obviously be good in multithreaded performance but are not sure about single threaded performance.
if bulldozer module handles a single thread and somehow the second integer schedular comes into play..then single threaded performance will also have some decent gain.
but this is all paper talk.we need bulldozer launching quickly.
core to core i still expect intel to be faster with their traditional approach but this will be of no practical significance as AMD with their bulldozer module approach will be able to pack in more cores in a limited die space and price compared to intel.
bulldozer also doesnt have to be faster than 2600k to win. history has taught us the fastest and better product doesnt always win. noobs are most likely to be fooled by the core count and insanely high clock speeds. another thing is...enthusiasts like us are a minority. real money is in server and OEM market.
a normal core has one integer schedular and one floating point schedular.normally a core handles one thread.
a bulldozer module has two integer schedulars and one floating point schedular.
intel's hyper threading "forces" two threads through a core which has integer schedular and one floating schedular. hyperthreading in real world applications gives u 10-15% boost. synthetic benchmarks may show u 50-60% boost but that doesnt matter.
a bulldozer module will handle two threads and both threads will get their own integer schedular but will share the floating point schedular.
people think, bulldozer will obviously be good in multithreaded performance but are not sure about single threaded performance.
if bulldozer module handles a single thread and somehow the second integer schedular comes into play..then single threaded performance will also have some decent gain.
but this is all paper talk.we need bulldozer launching quickly.
core to core i still expect intel to be faster with their traditional approach but this will be of no practical significance as AMD with their bulldozer module approach will be able to pack in more cores in a limited die space and price compared to intel.
bulldozer also doesnt have to be faster than 2600k to win. history has taught us the fastest and better product doesnt always win. noobs are most likely to be fooled by the core count and insanely high clock speeds. another thing is...enthusiasts like us are a minority. real money is in server and OEM market.
Last edited by a moderator:
Nice explanation but why AMD is delaying Bulldozer release day by day- are they thinking of competing with SB-E but with what price or performance
OP
Cilus
laborare est orare
Ya, Clock speed also plays a significant role in performance as architecture wise both the competitors are strong enough. AMD has faced lot of issues to deliver high clock speed in their previous gen architectures (who can forget the TLB Bug in their 1st series of Phenom Processors?) and their Trubo Core in Thuban is and also in Llano is simply not up to the mark. I guess they are making sure nothing of that kind will happen this time.
Example of Core clock speed is: If you overclock a 1090T or 1100T to 4 to 4.2 GHz, it beats Core i7 975 and perform closer to a i7 980X in most of the benchmarks.
By the way, yesterday, I have reached 3.9 GHz with my 1090T by using just the stock cooler. Idle was 32 degree and in load (while running 3D MArk 2011) it was 57 degree. I just increased the Multiplier to 19.5 and CPU V to 1.41 V. Great Oceing...na.
Example of Core clock speed is: If you overclock a 1090T or 1100T to 4 to 4.2 GHz, it beats Core i7 975 and perform closer to a i7 980X in most of the benchmarks.
By the way, yesterday, I have reached 3.9 GHz with my 1090T by using just the stock cooler. Idle was 32 degree and in load (while running 3D MArk 2011) it was 57 degree. I just increased the Multiplier to 19.5 and CPU V to 1.41 V. Great Oceing...na.
vickybat
I am the night...I am...
Actually a cpu is more biased towards integer operations than floating point operations. A gpu handles more floating point operations and is in contrast.
Amd after acquiring ATI has now a lot of expertise in gpu manufacturing and they know this fact well. Therefore, they shared the floating point unit and gave an extra integer unit to a single bulldozer module.
Thus not exactly having an extra core as a normal core, it surely has twice the integer units than a generic single core. Since floating point performance does not matter that much in cpu based tasks, amd reffers its modules as two cores because of these two execution units.
Therefore any thread having multiple instructions will be divided between the two units. The integer operations will be executed by the two integer units giving rise to ILP whereas the float operations will be shared by the single unit. Therefore instruction level parallelism will be restricted to integer operations and this is what is required for a cpu.
Thus this is more efficient form of smt (simultaneous multithreading) than intel's hyperthreading.