R2R DAC as in, a pile of resistors straight off GPIO pins?
Mind that these will not be very clean, both in terms of noise (you get 100% digital supply noise in the output, result depends on the quality of a power supply that may not be very quiet to begin with), and in the conversion itself (the resistors won't be matched).
Parallel DACs are cheap and plentiful, or SPI if you have the port to spare and can run it fast enough (or can accept a lower sample rate). Highly recommended.
A DAC converts raw digital levels, to controlled reference levels. The voltages on the R2R (if the chip uses one internally) might even be very similar, but the point is those tiny fluctuations left behind that. And, what if anything drives them. Might be digital noise on the supply, related to CPU activity, or USB or other peripherals; dirty external power, or ground loop noise; and it's probably not all that quiet in the first place, as voltage regulators don't need to be low noise. The fluctuations matter, since you're doing, however many bits you are, for example if it's 8 bits or more, that's ~0.4% or less of a difference for the smallest bit. 8 bits is pretty rough for audio purposes for example, and other electrical applications will show similar results. The one common exception I suppose being oscilloscopes, which typically only have 8 bits themselves, so won't be able to resolve finer details -- though digital filtering or high-res modes can improve things, YMMV.
As for the amplifier, and impedances -- if that's straight from GPIO pins, the rPi itself is probably closer to 50-100 ohms, but then add on the R2R resistance. That will be the source resistance seen by the amplifier. (Note that pin resistance causes an error in the R2R network, also limiting performance. R2R DACs are usually designed with much larger resistors, to mitigate this error; downside, the RC time constant gets huge.)
The DAC output node's capacitance can be mitigated by shorting it out: if there's no voltage swing, there's no dV/dt for the capacitor to respond to. That doesn't sound very useful, but we can make a virtual short circuit using a feedback amplifier. This is the inverting op-amp configuration, or more specifically, a transresistance amplifier (resistance because its gain is Vout / Iin, trans- because the quantities measured are in different places).
A very simple transresistance amplifier can be made from a few transistors, or really even just one in common-base configuration, though the latter doesn't have any current gain so you wouldn't get much advantage just yet. This is common design practice for DACs, back in the days when they were all R2R or current steering types -- the compliance (output voltage) range of the DAC is typically small, so that receiving it into a virtual ground node gives the least distortion and most bandwidth.
And you can always synthesize op-amps from discrete transistors, if you like; 2N3904/6 or similar jellybeans will get you 10s of MHz bandwidth even on the breadboard, and MMBTH10/81 or faster types will go higher, though you'll need to ditch the breadboard as capacitance is just too large there.
And you can always simply divide down the DAC output into a smaller resistance; this reduces the source resistance and signal level proportionally, so isn't any advantage by itself, but may be better suited to an amplifier, or to a certain amount of wiring (both cases being limited by input capacitance say, so the bandwidth goes up as the gain goes down -- the 1:1 tradeoff gives a fixed gain-bandwidth product (GBW), a figure of merit for the system).
It actually gets much better than that, or it can; when the capacitance is distributed, as in a transmission line, it is desirable to match source and/or load (at least one) to the transmission line impedance, and under this condition it will have essentially unlimited bandwidth (depending on length due to losses, or at stupid high frequencies due to waveguide modes -- 10s GHz for typical coax). Amplifiers can be tuned in a similar way, though it takes more specialized transistors to do so; the 2N3866 and friends still have a dominant lumped-capacitance behavior, so have a fixed GBW response.
Tim