TR Forums

@Meadows:

I think you're failing to see my point here.

Tie up all those 8 Jaguar cores and you just might be able to feed those Radeon cores well enough. Use 2-3 of those, and you're looking at starving the GPU and ultimately the gameplay experience.

Going back, I said devs must use all those Jaguar cores, given how each of them isn't particularly fast, at least from what we know thus far.

Or are you suggesting those Jaguar cores are particularly fast enough that devs don't have to use all of them to drive a beefy GCN-based GPU?

I'm sure Sony has a good reason for putting eight of those cores in there.

ronch wrote:

...
I'm sure Sony has a good reason for putting eight of those cores in there.

Take a look at Crysis3 thread usage and its clear that next gen title will have no problem using 6+ cores.

Jaguar include a brand new 'FPU' that more then double the theoretical peak compute of Bobcat. Double the throughput, but also optimized units and SSE4 & AVX support.

I also expect Sony to 'reserve' a core for background OS task and other always on features.

Also console 'driver' stack is possibly an order of magnitude more efficient then a classic windows graphic driver. So pushing more batch of 'triangles' is way less taxing on consoles.

I cant find the orignal blog post, but you can see from this that PC gaming require higher end HW for many reason to acomplish the 'same' thing.
http://www.psu.com/forums/showthread.ph ... -potential

So I wouldn't be surprised if you will need an overclocked i7-3770k + 680 GTX to run console title at default console settings.

Ryu Connor wrote:

HT can only show great benefits in niche scenarios: for example during a branch prediction failure and the subsequent pipeline flush, HT can run another waiting thread through the pipeline.

I disagree about the 'niche' tag; HT is great for loads that involve similar amounts of FP and INT processing. A single pure INT or (as close as possible to) pure FP thread may be able to saturate all of the associated units most of the time, but a mixed thread can't possibly saturate all of a core's FP and INT resources. I can't say much about that more than the theory and my one prior example, though, because I usually don't use FP at all. Cache misses are also a good scenario for HT. I have a hash-counting algorithm that splits hashing (integer bitshifts and xors) into some threads, and writing to the hash table (many gigs, presumably incurring a cache miss on every read or write) into other threads, and it scales decently with hyperthreading - up to about 5x singlethreaded on my 4-core machine with 8 threads, vs around 3.7x with 4 threads and HT disabled. Unfortunately in this case I have no control over which thread executes on which core; I might be able to get it to scale even better if I prevented same-type threads from occupying the same core.

sschaem wrote:

Jaguar include a brand new 'FPU' that more then double the theoretical peak compute of Bobcat. Double the throughput, but also optimized units and SSE4 & AVX support.

I thought of this too. If Jaguar has a saving grace it would be its beefed up FPU and newer instructions.

sschaem wrote:

I cant find the orignal blog post, but you can see from this that PC gaming require higher end HW for many reason to acomplish the 'same' thing.

For many years I've always thought of the PC as somewhat less efficient than other platforms. For example, I've always wondered how systems, such as the Playstation (1) can make do with a 'slow' 33MHz CPU to be able to push games such as Gran Turismo, with all those pretty (for its time) car models. Back in 1995 we can only run those kinds of games using a much faster Pentium at, say, 166MHz. I'm sure the Pentium also has an edge on IPC. This is perhaps true to this day, but we've reached a point where most platforms have plateaued already in terms of graphics capability that they're often less noticed than they were back in the 90's.

ronch wrote:

I've always thought of the PC as somewhat less efficient than other platforms. For example, I've always wondered how systems, such as the Playstation (1) can make do with a 'slow' 33MHz CPU to be able to push games such as Gran Turismo, with all those pretty (for its time) car models. Back in 1995 we can only run those kinds of games using a much faster Pentium at, say, 166MHz.

Because the PC used to have to do all the work on the CPU, with very minor help (like in sound) from other components. Only in the early 90s (or late 80s at best) did we have simple things like not having the CPU take care of every little disk transfer and sound buffering - an advancement that came courtesy of DMA.

On the other hand, with older consoles (and the Amiga / Atari ST / etc), the CPU was always a relatively minor player. All of them relied on their "GPUs", sound chips and other helps - all fixed-function - to do all the heavy lifting. This was both good and bad. Good because you could make a lot of things happen with very little hardware, but it made for difficult programming and porting, and provided little flexibility when you came across a situation that the helper chips couldn't handle.

PS - This is one of the reasons why one should never, ever compare MHz except when talking about different versions of the same chip, whatever it is

Benchmarks from pclab shows very different results



Source
http://www.overclock.net/t/1364211/pcla ... benchmarks

This is what sux about games with no built-in benchmarking tools - everybody are testing different areas/parts of the game and getting different results...

maroon1 wrote:

Benchmarks from pclab shows very different results



Source
http://www.overclock.net/t/1364211/pcla ... benchmarks

Going back to the very first post, why are the FX-8150 and FX-6300 so close if this is all about thrededness? Why is the FX-8350 so much faster than the FX-8150?