Kind of ironic that the most vocal critics of the pentium are the most vociferous advocates of very marginal performing APUs.
Well, I think the point was that even a "stuttering" pentium with any kind of decent dgpu, will still provide a better gaming experience than an apu.
Kind of ironic that the most vocal critics of the pentium are the most vociferous advocates of very marginal performing APUs.
I've built and used numerous systems with both AMD apus and intel pentiums and i3s. I have almost always been dissappointed with the general performance of the intel machines but the AMD apu machines were fine.
I'm stuck with an i3 machine now and it's torture just using the web browser.
I haven't gamed much on either, I use a i5 with dgpu for that. Just speaking of overall light duty speed and responsiveness.
Really?
What i3 can't handle web browsing? How can it be "torture" with an i3?
I can't tell the difference between my i3-4360 and my i7-4790k when I'm web browsing. Or my G3258 either, for that matter. They all have no problem with web browsing or watching HD videos. All using the igp.
Really?
What i3 can't handle web browsing? How can it be "torture" with an i3?
I can't tell the difference between my i3-4360 and my i7-4790k when I'm web browsing. Or my G3258 either, for that matter. They all have no problem with web browsing or watching HD videos. All using the igp.
Agreed. My work computers actually are core 2 duos, and even running disk encryption and antivirus in tbe background they are fine for web browsing and office use. It is pretty much inconceivable that a modern i3 would have problems.
Well, I think the point was that even a "stuttering" pentium with any kind of decent dgpu, will still provide a better gaming experience than an apu.
What's mystifying about the G3258 is the stuttering. Even when it manages some decent average FPS scores, it's the dips that kill it in SOME games (not all), and it's quite odd. You have to wonder if it's a misconfiguration issue, or something that couldn't be sorted out with higher uncore clocks or memory clockspeed. Or killing some background tasks maybe?
Starvation Prevention
Left alone, the FindReadyThread and ReadyThread might prevent low-priority threads from getting a chance to execute. For example, a priority 4 thread running on a system with continuously running priority 8 threads would be starved for CPU time. However, NT provides a mechanism that gives low-priority threads a shot at the CPU. The NT Balance Set Manager is a system thread that wakes up every second or so to perform memory tuning. As a secondary responsibility, Balance Set Manager executes the ScanReadyQueues algorithm, which implements NT's anti-CPU starvation policy.
ScanReadyQueues scans the Dispatcher Ready List, working down the list from priority 31. It looks for threads that haven't executed in more than 3 seconds. When it finds one, ScanReadyQueues gives the thread a special anti-starvation boost, doubles its quantum, and calls ReadyThread with the thread as a parameter. The anti-starvation boost differs from other boosts: Instead of applying a relative priority increment, the anti-starvation boost slams the thread's priority to the top of the dynamic range. (On pre-Service Pack 2--SP2--systems, the anti-starvation boost was to priority 14; post-SP2 systems boost to priority 15). When a thread that receives an anti-starvation boost finishes its extended quantum (or the thread blocks), its priority returns to the pre-starvation boost level and its quantum returns to its usual length.
You kidding yourself if you believe a Stuttering Pentium with a dGPU will provide a better gaming experience than a APU.
Just close two cores from your Core i5 and try it yourself and see. Although with the Pentium + dGPU you will be able to raise the resolution and perhaps the Image Quality, the stuttering makes the game unplayable. With a Quad Core APU you will play at lower resolution and or Image Quality but there will be no stuttering.
Did you just resurrect the slower but smoother mantra? :thumbsdown:
That is what the G3258 owners need to do in order to play games that Stutter. You know, fps limits, vSyncs, slower dGPUs etc.
This could simply be discovered by an user experience study instead of riding numbers.Did you just resurrect the slower but smoother mantra? :thumbsdown:
I know, it just seems, that it becomes the victim of argumentum ad ignorantiam, or that some project their own world of thinking onto every computer buyer out there. I think, I also brought it in this thread before.Brought up that point years ago...
"New" old data:HBM1 wont be faster than it is. Overclocking doesn't count. HBM2 will double the clock, still be far behind in latency.
And no, its still not better latency than DDR2/3/4. You really need to lookup latencies on DDR1/2/3/4 for more than the first word. But also 4th and 8th in the transfer.
"New" old data:
http://www.microarch.org/micro47/files/files/LightningTalks.pdf
Slide 2 (by Intel, GATech) says (stacked DRAM compared to standard DRAM):
Bandwidth: 2X to 8X
Latency: 0.5X to 1X
Capacity: 0.25X
So latency doesn't seem to be seen as a problem by Intel.
And funny: Intel will use Bulldozer for DRAM stacking
Now I see, where you got that high latency idea from. HBM and HMC both have stacked DRAMs. Both stack variants offer vast bandwidth via thousands of TSVs. Both work at energy efficient working points. The main difference is the communication with the connected processor. But that communication doesn't change the latency of the internal DRAM bank accesses, but just transfers the addresses and data in different ways.Stacked memory is irrelevant. This only confirms you dont have any understanding of the reason behind. But just blindly defends it without knowledge.
The paper involved takes base in HMC. A serial design that runs up to 15Gbps using SerDes. While the signaling rate is up to 1Gbps for HBM with its slow wide parallel bus.
That is what the G3258 owners need to do in order to play games that Stutter. You know, fps limits, vSyncs, slower dGPUs etc.
And what do APU users do? Turn the resolution to 720p or play sub 30fps at lowest settings at 1080p. Great alternative.
Nope, I ignored the stacked mobile dies, flash dies, Charlie's report of AMD's work with Amkor etc. Someone hacked my old blog and posted somewhen in 2012 or 2013, that the PS4 chip might possibly use stacked memory as AMD was working on it for many years already.I dont hope you think HBM or HMC is new as such in stacking. Plenty of products already there with stacked dies.
Now I see the other problem. You cannot compare the parallel communication on an interposer with more energy consuming high speed SERDES communication between dies via PCB. Where is the performance/energy crossover? Surely, it costs more energy and place to route parallel wires over PCB. But at very close distances with very tiny wiring (silicon metal layers), low communication voltages and such, the advantage might simply shift to parallel communication. High speed SERDES also has to take more care of the eye in the signal. One surely could plot a pareto frontier of serial vs. parallel communication in different scenarios with different HW implementations.HBM may solve its latency issues as speed ramps up. But its not in the current product. But they had the chance to remove the outdated parallel bus and then went all in on it instead. Its the last part in the PC that still use a parallel bus. Serial busses are vastly superior.
You also ask for a case that is already plenty of examples of. The slow bus always lose in latency. The penalty is a lot lower on GPUs that peak at 1Ghz rather than 4Ghz CPUs.
The difference is that G3258 users expectations were high and they were believing they could use a high-end dGPU for high 1080p gaming. Unfortunately for them that turned out to be a stuttering fest.
On the other hand, APU users expectations were lower from the beginning. APU users knew from the start they were going to play at 720p or have low fps at 1080p.