Open-Source AMD HSA Should Come To Fruition This Year

[profoundWHALE]

Nothing has perfect scaling, even atomic instructions used in lockless methods have overhead. Thus a 50GHz cpu would always be better than 50 1 GHz cpus in performance.

This leads to an interesting question: Why don't we have really ridiculously large CPUs that are about 50GHz and take up as much room as 50 1GHz CPUs?

Because it's cheaper and more effective to hire programmers to write good code. Also if one machine breaks they can usually just swap it out if I'm not mistaken. If that super expensive CPU burns out then that's it, toast, gone. I believe the word is scalability.

That being said, I wouldn't mind having a very large two core CPUs. Actually, I wouldn't mind having a FX-8350 that's 4 times the size if it means that it can use passive cooling. That would be awesome!

[bridgman]

This leads to an interesting question: Why don't we have really ridiculously large CPUs that are about 50GHz and take up as much room as 50 1GHz CPUs?

It's sort of a two part answer. For any given fab process/voltage and resulting gate delay there's certain relationship between clock frequency and the amount of pipelining required, and the amount of pipelining goes up "faster than linear" past a certain clock frequency, so...

1. The "silly frequency" (ie where the increase in pipelining gets silly relative to the increase in clock frequency it buys you) for current fab processes is probably below 10 GHz today

2. Even if you weren't dealing with diminishing returns in terms of clock frequencies alone, there's a performance hit you get from branching, and to some extent the inability to perfectly predict branches limits the amount of pipelining (and hence clock frequencies) even if the mfg was willing to spend silly amounts on pipelining

We may finally be at the point where 1024 chickens *can* plow the field better than two strong oxen. I would like to see *that* on Mythbusters

[asdfblah]

A 50GHz CPU (if possible at all) would use too much power, wouldn't it?


I wonder if AMD plans to use that disasterous OpenCL implementation... they would become the laughing stock (as they are now) if things don't work as expected.
As it has been repeatedly over and over again, AMD hardware is the best, AMD drivers/software are the worst. Meanwhile, nVidia says they are a software company.
It's a good thing that we have a whole new driver/OpenCL implementation in Linux, but it's a terrible thing that there are so few developers..

[smitty3268]

This leads to an interesting question: Why don't we have really ridiculously large CPUs that are about 50GHz and take up as much room as 50 1GHz CPUs?

At 50Ghz you're most likely either going to run into power issues (too much in too small a space) or speed-of-light issues trying to keep the clock in sync across an entire massive die.

Believe me, if it were technically feasible, that's the direction the CPU companies would have gone in. Intel certainly thought they were going there in the P4 days, and they completely gave up that strategy when they ran into a brick wall trying to implement it.

[Bucic]

And just what real world application do you need more single thread performance for as opposed to more cores?
Games? Nope, Core 2s are fine in terms of single thread performance for most games
Compiling? There's a lot more to be gained by going parallel
Spreadsheets? Nope
Video Encode/Decode? That's offloaded to the GPU

ArmA 3 and every other simulation/CPU-intensive game. Majority of games doesn't benefit from many cores. Even quad cores don't get a boost over dual cores in most games.

[Luke_Wolf]

ArmA 3 and every other simulation/CPU-intensive game. Majority of games doesn't benefit from many cores. Even quad cores don't get a boost over dual cores in most games.

I want you to notice something about this chart compare the FX-8350 to the i7-4770k
the difference? 5FPS, with both well above 30FPS with no CPU going above 60

now
compare

vs

overclocked to 4.5GHz clock for clock there is then only a 3 FPS difference

Further from the Battlefield 4 official site http://www.battlefield.com/battlefield-4/pc

PC SYSTEM REQUIREMENTS

Minimum requirements:

OS
WINDOWS VISTA SP2 32-BIT (WITH KB971512 PLATFORM UPDATE)
PROCESSOR
AMD ATHLON X2 2.8 GHZ
INTEL CORE 2 DUO 2.4 GHZ
MEMORY
4 GB
GRAPHICS CARD
AMD RADEON HD 3870
NVIDIA GEFORCE 8800 GT
GRAPHICS MEMORY
512 MB
HARD DRIVE
30 GB

Recommended requirements:

OS
WINDOWS 8 64-BIT
PROCESSOR
AMD SIX-CORE CPU
INTEL QUAD-CORE CPU
MEMORY
8 GB
GRAPHICS CARD
AMD RADEON HD 7870
NVIDIA GEFORCE GTX 660
GRAPHICS MEMORY
3 GB
HARD DRIVE
30 GB

and here's the one for Crysis 3 http://www.crysis.com/us/crysis-3/buy

MINIMUM SYSTEM OPERATING REQUIREMENTS FOR PC:
• Windows Vista, Windows 7 or Windows 8
• DirectX 11 graphics card with 1Gb Video RAM
• Dual core CPU
• 2GB Memory (3GB on Vista)
• Example 1 (Nvidia/Intel):
• Nvidia GTS 450
• Intel Core2 Duo 2.4 Ghz (E6600)
• Example 2 (AMD):
• AMD Radeon HD5770
• AMD Athlon64 X2 2.7 Ghz (5200+)

RECOMMENDED SYSTEM OPERATING REQUIREMENTS FOR PC:
• Windows Vista, Windows 7 or Windows 8
• DirectX 11 graphics card with 1GB Video RAM
• Quad core GPU
• 4GB Memory
• Example 1 (Nvidia/Intel):
• Nvidia GTX 560
• Intel Core i5-750
• Example 2 (AMD):
• AMD Radeon HD5870
• AMD Phenom II X4 805

HI-PERFORMANCE PC SPECIFICATIONS:
• Windows Vista, Windows 7 or Windows 8
• Latest DirectX 11 graphics card
• Latest quad core CPU
• SLI / Crossfire configurations will run even better
• 8GB Memory
• Example 1 (Nvidia/Intel):
• NVidia GTX 680
• Intel Core i7-2600k
• Example 2 (AMD):
• AMD Radeon HD7970
• AMD Bulldozer FX4150

And guess what? with the consoles running octocore netbook processors, now guess what game developers are going to be forced to optimize for?

[Luke_Wolf]

Nothing has perfect scaling, even atomic instructions used in lockless methods have overhead. Thus a 50GHz cpu would always be better than 50 1 GHz cpus in performance.

Always? no. Multi-threading doesn't have perfect scaling however neither does frequency.

• Can you run SuperPi faster on that 50GHz cpu? hell yes.
• Will LibreOffice run faster? no.
• Will a game run faster? maybe, maybe not since they're running at 1Ghz as opposed to the more usual 2 or 3+, plus there's a bunch of other factors in play like what kind of game and so on, but say 12 4Ghz CPUs will provide you a better experience for most games. why?
  • Throwing more Hz at the problem does nothing about the interrupts whereas more cores can potentially do so
  • a game might not be able to take advantage of more cores but keep in mind you've got other things running in the background like say... the OS and if you've just got a single core it's going to have to do a lot more task switching than if you've got more cores available for these long running processes to expand themselves out to and stay in
• Will your overall experience be better (assuming said 1Ghz CPUs are enough for most tasks)? no. It's not something that you can show in a benchmark but interactivity is improved when you have more cores, exactly because you can spread out those long running processes to more cores and thus remove a lot of task-switching on the CPU

[drSeehas]

... Kabini is the current low end line ...

No!
The current low power line is Beema/Mullins.

... replaces the older Stars cores of the Jaguar ... lines.

Kabini replaces itself? (Hint: Kabini has Jaguar cores)

... some confusion.

You enlarge the confusion.

[Bucic]

now
compare

vs

overclocked to 4.5GHz clock for clock there is then only a 3 FPS difference

Is this seriously your conclusion? No, wait. What was even your point exactly? What were you trying to prove by comparing the two?

[JS987]

Oh please, the average GUI application spends almost all of it's time listening on the event loop as opposed to actually doing anything, and I don't exactly need a xeon to render QtCreator or Dolphin. Also please link to this website that you visit that demands all this performance that you need "a lot more single thread performance" than AMD currently provides. Keep in mind that if you have to draw upon sunspider or other web benchmarking sites, that you'll be proving my point for me.

Latency of complex GUI / website to keyboard and mouse events usually depends on single thread performance. Every modern desktop CPU should be fast enough, but latency can be different.