The TLB erratum will probably never be seen by most consumers, but in the server arena its a different matter. Details below:
254 Internal Resource Livelock Involving Cached TLB Reload
Description
Under a highly specific and detailed set of conditions, an internal resource livelock may occur
between a TLB reload and other cached operations.
Potential Effect on System
The system may hang.
Suggested Workaround
BIOS should apply the workaround based on the extended model, base model, and stepping reported
by CPUID Fn8000_0001 as described in the following table:
Fix Planned
Yes
CPUID
Fn8000_0001_EAX[19:16]
(Extended Model)
CPUID
Fn8000_0001_EAX[7:4]
(Base Model)
CPUID
Fn8000_0001_EAX[3:0]
(Stepping)
HWCR[3]
(MSRC001_0015h)
BU_CFG[1]
(MSRC001_1023h)
BU_CFG[21]
(MSRC001_1023h)
0h 2h Ah 0b 0b 1b
0h 2h 2h 0b 0b 1b
(sorry, its a cut and paste from a pdf....)
Our server engineers report that it is very prevelant using any type of virtualization, and also shows up under heavy DB usage. As such, we won't be looking at this chip before the B3, and will test the hell out of it again. On the positive side, the cores do scale well, 8P performs as well as or better than 4P, and 16P scales better than 8P. On the negative side, it's power hungry and also BTU hungry.
Our preliminary Xeon tests are due to start, hopefully, next month.