Which test of the three torture tests? Blend is more memory/NB, while the other two focus more on the core.
What about your ram? What is it running at? memtest will usually pick out defective ram, however it sometimes doesn't throw errors if good ram is clocked too high (or too tight timings).
And finally, what VCore is the chip defaulting to. Defective CPUs are pretty rare, so you want to be certain it has to be the CPU before sending it back.
Also, try other programs such as LinX if you can.