Quite frankly two occurrences which at least one appears to be the user's fault out of tens of thousands CPU sold are statistically insignificant. In any case, there's Buildzoid's take on it:
We can rule that out I think. Wouldn't have lasted 20 days unless there's some process that causes unconductive crap to turn into conductive crap through heat over a period of three weeks. Most likely ASROCK BIOS defaults or EXPO voltage handling are to blame.
Or it stops making proper contact and at that point the consequences are similar.If there's a bent pin, it may not cause issues at first, but after many thermal cycles the pin continues to deform until it gets too close to a neighboring pad/pin and then the problem becomes spectacularly apparent.
In the classroom, kids doing PCs unbuild/rebuild break pins half of the time.if I asked them to remove CPU. Thanks LGABring back the pin grid array
Well, it is conceivable Nvidia for example could try try to divert attention due to their melting issue (or at least give fans ammo for whataboutisms cope), but I *really* doubt it. Such conspiracy explanations are unlikely. Maybe more to be expected from a single silly fan (think Userbenchmark guy), but in that case the dude would have to sacrifice a lot of money.smells like FUD
I suggest sacrifying the dude insteadWell, it is conceivable Nvidia for example could try try to divert attention due to their melting issue (or at least give fans ammo for whataboutisms cope), but I *really* doubt it. Such conspiracy explanations are unlikely. Maybe more to be expected from a single silly fan (think Userbenchmark guy), but in that case the dude would have to sacrifice a lot of money.
It's likely just normal failures. Hidden manufacturing fault, board somehow frying the CPU (wouldn't be the first time on AMD or Intel platforms - remember V-Cache is voltage sensitive)...
What is the significance of the yellow hypervisor text above the Save button. Could the test be running inside a VM?Read bandwidth seems like a huge disappointment
Seems AIDA64 has detected sandbox or virtualized environment. Hmmm...What is the significance of the yellow hypervisor text above the Save button. Could the test be running inside a VM?
I think from the chips and cheese interview, most is provisioned to the graphics. They kept saying something like all the fabric goes there.
They said in the interview that Halo CCD has the same BW (actually it's higher as Granite Ridge has 32B/c read and 16B/c write BW between CCD and IOD, and Halo is supposed to have 32B/c BW for both R/W) that desktop parts, so both CCDs together should be able to reach 200GB/s read. It makes little sense for write to be able to reach that and read not being able to do so, therefore I would wait for tests with final bioses before reaching any conclusions.I think from the chips and cheese interview, most is provisioned to the graphics. They kept saying something like all the fabric goes there.
Ooouh, now you got me worried, not 9800x3d but 7800x3d, a friend gave me an Asus x670e Gene that they had damaged several pins. I managed to straighten them and am using the system as my daily. Its been functioning fine since I repaired it and thats around 6 months ago.If there's a bent pin, it may not cause issues at first, but after many thermal cycles the pin continues to deform until it gets too close to a neighboring pad/pin and then the problem becomes spectacularly apparent.
Just don't watch anything on it that God doesn't approve ofI managed to straighten them and am using the system as my daily. Its been functioning fine since I repaired it and thats around 6 months ago.
No, that latency is a disappointment. The R/W speeds are fine, IMO.
That's what you get with LPDDR any version. Only way to avoid that is to not use LPDDR. Or come up with some sort of FrankenDDR tech that uses both DDR4 and LPDDR5 together in such a way that low latency is provided by the former and high bandwidth by the latter.No, that latency is a disappointment.
140nS is really high if that's accurate. Most mobile systems are in the 100nS range. It makes Arrowlake look very good.That's what you get with LPDDR any version. Only way to avoid that is to not use LPDDR. Or come up with some sort of FrankenDDR tech that uses both DDR4 and LPDDR5 together in such a way that low latency is provided by the former and high bandwidth by the latter.
Doesn't work like that, the channels are not dedicated.It's lpddr5x which should have high latency high bandwidth, quad or octo channel. I think the system can do weird subchannel dual quad or something which is probably what were seeing here. Video gets a dedicated quad channel plus the MALL cache. CPU and GPU share the other quad channel.