RnR_au
Diamond Member
The big one is doable with llama.cpp inference engine and cpu offloading;GPT-OSS at what size?
The big one is doable with llama.cpp inference engine and cpu offloading;GPT-OSS at what size?
32MB L3? Surprising. Is each core, dense or classic, getting the full 4MB of L3 per core? And are they all in one cluster, unlike strix point?Zen0.6%
"processor_frequency": {
"frequencies": [
2006,
1529,
1388,
1387,
1387,
1369,
1389,
1388,
1387,
1383,
1387,
1386,
1388,
1388,
1387
]
},
Not really, it's still Family 1Ah.Also, still only 32KB of LIi? Is this not surprising?
No idea.32MB L3? Surprising. Is each core, dense or classic, getting the full 4MB of L3 per core? And are they all in one cluster, unlike strix point?
RWC from GLC doubled the L1i cache on a tick core T-TNot really, it's still Family 1Ah.
The only time they changed L1 capacity within a family was Zen 1 -> 2. Them changing even the scheduler layout within a family is surprising.
I do think it's a "phoenix-2" like setup where it's all in one cluster. Since this is an ES, it could be misreporting, but if they were in separate clusters I imagine it would just not be able to differentiate or add the L3 from another cluster rather than accidentally combining them.No idea.
Zen6 is supposed to get the performance boost from clocks, right? A low-clocked ES should not tell you anything about the performance then.Zen0.6%
And IPC.Zen6 is supposed to get the performance boost from clocks, right? A low-clocked ES should not tell you anything about the performance then.
Don't read into it at all.32MB L3? Surprising. Is each core, dense or classic, getting the full 4MB of L3 per core? And are they all in one cluster, unlike strix point?
Also, still only 32KB of LIi? Is this not surprising?
I'm memeing.Zen6 is supposed to get the performance boost from clocks, right? A low-clocked ES should not tell you anything about the performance then.
Been wondering when Geekbench 7 will come out. Has to be getting pretty close.Don't read into it at all.
You won't ever know what's actually inside unless you have a special signed build of GB6/7.
Pretty sure it should've been out already.Been wondering when Geekbench 7 will come out. Has to be getting pretty close.
Why are there 15 processor frequencies reported? I am entirely ignorant of what GB6 is doing, but I would have expected it to show as many frequencies as there are logical CPUs in a system. Is this a new 5-core CPU with SMT3? ;-)"processor_frequency": {
"frequencies": [
2006,
1529,
1388,
1387,
1387,
1369,
1389,
1388,
1387,
1383,
1387,
1386,
1388,
1388,
1387
]
},
Of course the first Z6 leak is MDS1 with clearly unrepresentative results.Zen0.6%
"processor_frequency": {
"frequencies": [
2006,
1529,
1388,
1387,
1387,
1369,
1389,
1388,
1387,
1383,
1387,
1386,
1388,
1388,
1387
]
},
I'm just guessing, but I think these are clock readings at different points in time. Some 10-core chips have like 30 readings over the course of a run for example.Why are there 15 processor frequencies reported? I am entirely ignorant of what GB6 is doing, but I would have expected it to show as many frequencies as there are logical CPUs in a system. Is this a new 5-core CPU with SMT3? ;-)
Si's been out for a while too. funny.Of course the first Z6 leak is MDS1 with clearly unrepresentative results.
Well things will only accelerate from here.
John said sometime this yearBeen wondering when Geekbench 7 will come out. Has to be getting pretty close.
John who?John said sometime this year
Poole.John who?
John Poole, the guy who made GeekbenchJohn who?
If you extrapolate the score from the 1.4 GHz clock to 5.7 GHz and assume the usual GB6 score for Zen 5 core at 5.7 GHz is 3550 points, it turns out the IPC has gone up by... /drumroll/Zen0.6%
"processor_frequency": {
"frequencies": [
2006,
1529,
1388,
1387,
1387,
1369,
1389,
1388,
1387,
1383,
1387,
1386,
1388,
1388,
1387
]
},
John Cena? The Pluribus cameo actor?I always assume it's John Cena whenever someone mentions John.
Indeed.I'm just guessing, but I think these are clock readings at different points in time. Some 10-core chips have like 30 readings over the course of a run for example.
Indeed. Looking at the ST score if the frequency is 2.0 then performance is flat over Lion Cove. If frequency is 1.4 then we're looking at 45% better than Lion Cove. Obviously reality is somewhere in the middle, which we could have guessed 10 years ago so the information is meaningless.I'm memeing.
Trying to read into performance from 1.4-2.0 GHz region is indeed pointless.