-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
XBone AMD CPU @ 1750 MHz, memory @ 2133 MHz: CPU bottlenecking memory?
- Thread starter magellan
- Start date
- Joined
- Aug 4, 2011
EDIT: The memory benchmark will have a poorer score.
Original post:
Memory transfer operations are kind of funny. For every clock cycle the memory goes through, the amount of data it transfers can vary significantly.
http://www.anandtech.com/show/3851/...w-about-sdram-memory-but-were-afraid-to-ask/2
And the processor is in a similar boat. You have heard of IPC, or instructions per clock. How much work can a processor do per clock cycle? It varies, depending on the instruction in question. It's why Intel keeps introducing new instruction sets. The new ones are physically more efficient for certain operation types.
And then there is the cache. If I feed a processor code to calculate PI, all of my code will fit in the cache. The answer it spits out will also fit in the cache. There is no bottleneck here. The cache is a buffer against the terrible speed and latency of the DRAM.
The ram speed starts to matter more when you do an operation that doesn't fit in the cache very well. Like zipping up a file. All that data has to get pulled from main memory, in its entirety, compressed, and written back to main memory. Fast ram makes that work rather better.
But even then, the processor is largely the bottleneck, because the compression process takes computation time. Much more than the transfer time from the ram.
I'm very sleepy. I hope that helps.
Original post:
Memory transfer operations are kind of funny. For every clock cycle the memory goes through, the amount of data it transfers can vary significantly.
http://www.anandtech.com/show/3851/...w-about-sdram-memory-but-were-afraid-to-ask/2
And the processor is in a similar boat. You have heard of IPC, or instructions per clock. How much work can a processor do per clock cycle? It varies, depending on the instruction in question. It's why Intel keeps introducing new instruction sets. The new ones are physically more efficient for certain operation types.
And then there is the cache. If I feed a processor code to calculate PI, all of my code will fit in the cache. The answer it spits out will also fit in the cache. There is no bottleneck here. The cache is a buffer against the terrible speed and latency of the DRAM.
The ram speed starts to matter more when you do an operation that doesn't fit in the cache very well. Like zipping up a file. All that data has to get pulled from main memory, in its entirety, compressed, and written back to main memory. Fast ram makes that work rather better.
But even then, the processor is largely the bottleneck, because the compression process takes computation time. Much more than the transfer time from the ram.
I'm very sleepy. I hope that helps.
Last edited:
- Joined
- May 15, 2006
Yes, because there are many other factors than just frequency of the part.
Magellan, take a look at any of my(the) motherboard reviews on the front page here. Pay particular attention to the AIDA64 memory benchmarks and how little they change from 4.2Ghz to 4.7Ghz.
OP
- Joined
- Jul 20, 2002
- Thread Starter
- #8
That's nice, but how much would they change if you went from 4.2 GHz down to 1.750 GHz? A frequency drop that represents not only a decrease in CPU speed of ~60%, but an operating frequency below that of the RAM.
I'd figure this issue would be even more pronounced on the XBone because the memory bandwidth is being shared between the GPU and CPU.
I'd bet MS's published specs for the theoretical memory bandwidth of the XBone (as calculated based on the 2133 MHz DDR3 memory) aren't anywhere near the reality. Ditto for the PS4.
- - - Updated - - -
But Odie, isn't the cache also running much, much slower on the XBone and PS4 than on any Haswell-E, SB-E, IVB-E CPU?
I'm not happy the L3 cache on my Haswell-E is running at 3500 MHz, but that's still twice the speed of any of the caches on the XBone or PS4 -- unless the L3 cache resides in the IMC, which I figure must be running @ 2133 MHz.
not sure how that unrealistic situation would affect it. Test it out, if possible, and show us what you found.
I don't understand the logic in being upset with your cache running at whatever speed it is. When you buy a CPU, you buy the package. You buy performance as a whole, not in part. Your are 'not happy' with what is essentially an arbitrary bit of information that doesn't define the whole.
Now, system memory does play a role in the iGPU's performance because the igpu uses system memory as a discrete card would it's onboard memory. There are reviews out there that show the increase and not sure if/where that curve flattens out.
Last edited:
Similar threads
