My prediction is core count will likely scale in the future similarly to how RAM scales today.Originally Posted by Shintai
My prediction is core count will likely scale in the future similarly to how RAM scales today.
Ryzen 3800X @ 4.4Ghz
MSI X570 Unify
32GB G.Skill 3600Mhz CL14
Sapphire Nitro Vega 64
OCZ Gold 850W ZX Series
Thermaltake LV10
there are 3 main factors that go into performance, speed and # of cores are the most important that people look at. but theres also the complexity per clock that can give an easy 20% bonus between chips and manufacturers. back 7-10 years ago we had the mhz/ghz war, in the last 2-3 years, and probably the next 5 years, its gonna be the multitasking war. but after that, i think its gonna go back to the ghz war again once things cant be threaded any more. and poor old IPC will be left alone and no one will still care about him.
Doesn't matter how good these processors are. You are still limited to the socket F and that means DDR2 667 or 800. So, I agree, these are not good for gamers.
ca´t imagine how many diseases this monster could possibly cure ..
"Study hard my young friend"[/B].
---------------------------------------
Woody: It's not a laser! It's a... [sighs in frustration]
What would be good was if they increased the speed between CPU and GPU.
What would be good is if "they" would create GPU's that use up the current level of bandwidth so that there was a need to increase the speed between the CPU and GPU...
I meant total speed (including latency). If that was fast you wouldn't need memory on the gpu
i7 did that, look at how much better CF & SLi scale on X58 systems. The next big leap may be an optical FSB.
Intel i7 920 C0 @ 3.67GHz
ASUS 6T Deluxe
Powercolor 7970 @ 1050/1475
12GB GSkill Ripjaws
Antec 850W TruePower Quattro
50" Full HD PDP
Red Cosmos 1000
Latency isn't all that high currently, and yes, you would still need memory.
How high/low is it?
It is high, very much of GPU programming is done to minimize calls to GPU and send as much data as you can in each call to avoid latency.
Not sure about that, I mean just look at IPC difference from P4 to I7, sure 2009 and 2010 are all about increasing cores and clocks, but then it will be Sandy bridge time and I bet IPC will be center of attention once again.
Assuming you could massively increase the bandwidth for memory on the CPUs...
IGPs dont need memory. And its not a latency issue or speed issue of the connection. Its simply the total memory speed issue on the CPU.
Crunching for Comrades and the Common good of the People.
the biggest problem with IPC, is that advertising it to the average person is really tough to do. back in the first days of athlon, we had the "2500+" equivalents, people needed to hear what they were equal too, inorder to figure out why they should buy something thats 700mhz less than the old chip. for IPC to really take over, companies would probably have to advertise synthetic benchmarks, like "total cpu now does 1.2 giga-flozers, 20% more than than the old chip". but even that would still be scaling much quicker just from speed and core count. and in real life that number probably would mean nothing. as nice as having good IPC is, you really need to beat the opponent hard for it to mean anything, and translate it so average joe gets it.
For the type of work gpu is doing today (graphics) it needs massive bandwidth (and that is what they are good at). But if you take it one step further it needs to improve latency. Physics is one area that will get much help if latency was lower.
Don't say the word physics. Talonman will be all over this thread.
Posting Permissions
Reply With Quote
Bookmarks