Kaveri vs FX: is single threaded performance improving?

Gref · January 18, 2014

It seems a lot of Kaveri benchmarks focus on the GPU side. What I'm interested in is the CPU performance alone. Has Kaveri improved on single threaded performance versus the FX series? I would like to see some gaming benchmarks with the A10-7850k vs FX-4350 with graphics cards.

How well does Kaveri overclock versus the 4350?

AMD hasn't made much improvements in single threaded performance for a long time, does Kaveri's new architecture and smaller manufacturing process bring noticeable improvements?

IVIeLt · January 18, 2014

Yeah there's improvements, but stock clocks are lower and OC potential doesn't seem all that great so FX is better IMO

Vitalius · January 18, 2014

Well FX's single core performance was majorly hindered by sharing cache and a single FPU shared between 2 cores.

The APUs have many FPUs (the GPU cores are basically FPUs) and IIRC each CPU core doesn't share resources the same way they did on FX.

So I imagine APU single core performance to be fairly awesome in comparison to FX CPUs.

D3MON · January 18, 2014

k10 was great, bulldozer was a bit worse per clock but clocked better but had a release microsoft would be proud of, Piledriver improved efficiency and temps allot and kept the same clocks (some higher) but outperformed k10 (so fx6300 compares to phenom x6)(id chose a phenom over a 4300 but a 6300 over a phenom x6 and an 8320 over anything upto i7-enthusiast), steamroller cores might add another 10-15%.
AMD is closing the gap on intel, because its easier for them to do it, they're still learning the module architecture and what matters to it and improving on it with regards to IPC efficiency etc.
all we seem to get form intel at the moment is 5% more ipc with a 5% reduction in average clock speeds, IB ran hot and everyone de-lidded them, then haswell came out and was worse, only reason to get haswell is to run it at basic clocks, by which i mean around 4ghz, (non-k, overclock 4 multipliers to 3.9ghz, done)
processors have been at 3ghz since the pentium d days, intel needs to stop playing the "but we saved you 13 watts" game and start raising those lame stock clocks.

single threaded performance is pretty dead anyway for allot of things, otherwise everyone would have pentium g3220's,
for that matter in 5 years im guessing most apps that would be used on 4-8 threaded processors like the fx8's and i7's will likely be run off discrete graphics with maybe a quad core relegated to the things that simply arnt written for it, GPU for gaming and production, iGPU for physx and or highly parrallel computing in the mainstream and quad cores (or whatever) for the other few apps and programs left.

p.s given the current way things are i can see the next few years slowing down with moores law fighting intel (the irony) they either have to continue their efficiency strive reducing power draw as node sizes decreate but leaving us with little to no performance gains (for us overclockers)
OR maintain the same power envelope but do as i pointed out and while the reduced nodes decrease power draw its offset by slowly increasing clock speeds from 3.5-3.7-to 4ghz...trying to keep going at the end of the marathon.

AMD in the meantime are a bit further back having stumbled a bit, but they have a bit left in thier tank are recovering and are getting back into thier stride picking up the page and catching up, for anyone thinking this is ballony think about this...
if amd could improve ipc by as much as they did from bulldozer to piledriver and gain another 15%, PLUS reduce node size by 45% by going to 22nm getting a double whammy of lower power consumption, more profit/wafer, a fractional increase in clocks while being able to fit more transistors on (for say APU's?) and then combine that with HKMG transistors, we could be looking at a 22nm 6core/12thread, HKMG, FX1250@4.5ghz and an FX1270@5ghz with a small APU taking care of HUMA applications and floating points with the whole system using ddr4 by then so the apu has double the memory bandwidth it has now with 32gb of 3200mhz CL15-15-15-45 becoming typical. This could be within 2-4 years.
4 years ago if you had the cash you had an i7 950, still a great chip, but compare it to a 4770k....

Jumper118 · January 18, 2014

k10 was great, bulldozer was a bit worse per clock but clocked better but had a release microsoft would be proud of, Piledriver improved efficiency and temps allot and kept the same clocks (some higher) but outperformed k10 (so fx6300 compares to phenom x6)(id chose a phenom over a 4300 but a 6300 over a phenom x6 and an 8320 over anything upto i7-enthusiast), steamroller cores might add another 10-15%.

AMD is closing the gap on intel, because its easier for them to do it, they're still learning the module architecture and what matters to it and improving on it with regards to IPC efficiency etc.

all we seem to get form intel at the moment is 5% more ipc with a 5% reduction in average clock speeds, IB ran hot and everyone de-lidded them, then haswell came out and was worse, only reason to get haswell is to run it at basic clocks, by which i mean around 4ghz, (non-k, overclock 4 multipliers to 3.9ghz, done)

processors have been at 3ghz since the pentium d days, intel needs to stop playing the "but we saved you 13 watts" game and start raising those lame stock clocks.

single threaded performance is pretty dead anyway for allot of things, otherwise everyone would have pentium g3220's,

for that matter in 5 years im guessing most apps that would be used on 4-8 threaded processors like the fx8's and i7's will likely be run off discrete graphics with maybe a quad core relegated to the things that simply arnt written for it, GPU for gaming and production, iGPU for physx and or highly parrallel computing in the mainstream and quad cores (or whatever) for the other few apps and programs left.

p.s given the current way things are i can see the next few years slowing down with moores law fighting intel (the irony) they either have to continue their efficiency strive reducing power draw as node sizes decreate but leaving us with little to no performance gains (for us overclockers)

OR maintain the same power envelope but do as i pointed out and while the reduced nodes decrease power draw its offset by slowly increasing clock speeds from 3.5-3.7-to 4ghz...trying to keep going at the end of the marathon.

AMD in the meantime are a bit further back having stumbled a bit, but they have a bit left in thier tank are recovering and are getting back into thier stride picking up the page and catching up, for anyone thinking this is ballony think about this...

if amd could improve ipc by as much as they did from bulldozer to piledriver and gain another 15%, PLUS reduce node size by 45% by going to 22nm getting a double whammy of lower power consumption, more profit/wafer, a fractional increase in clocks while being able to fit more transistors on (for say APU's?) and then combine that with HKMG transistors, we could be looking at a 22nm 6core/12thread, HKMG, FX1250@4.5ghz and an FX1270@5ghz with a small APU taking care of HUMA applications and floating points with the whole system using ddr4 by then so the apu has double the memory bandwidth it has now with 32gb of 3200mhz CL15-15-15-45 becoming typical. This could be within 2-4 years.

4 years ago if you had the cash you had an i7 950, still a great chip, but compare it to a 4770k....