Ryzen 5 1600, Core i5 7500, or Core i7 7700 for deep learning (Tensorflow) on GPU?

[shab_2008]

Hi there,
I plan to work with some deep pretrained convolutional neural networks for an object-detection task. I recently bought a GTX 1080 Ti to perform all the computations on it. My main problem, however, is that which CPU should be paired with this GPU to avoid any bottleneck. In this regard, I’d like to ask you two questions.
I have two main choices: Ryzen 5 1600 and Core i5 7500. As I saw on some websites that evaluated the performances of these CPUs, Ryzen 5 1600 beat Core i5 7500 in most tests. Based on this, my first question is that which of these CPUs is a better choice for my need?
Apart from these two CPUs, I’ve heard from some people that a Core i7 is the minimum requirement for doing deep learning on my GPU; otherwise, the GPU could be bottlenecked and could not exert its potential. In this respect, my second question is that, is it worth increasing my budget to get a Core i7 7700 (or 6700) rather than a Ryzen 5 1600 or a Core i5 7500?
Thanks

 

[Bungle11]

I would guess for deep learning more threads would be better. 6core/12 threads vs 4/4 (i5) vs 4/8(i7). I would say 6/12 is more advantageous than 4/4 on the i5. In your situation, as the cpu can spread the work load around. Out of the box the r5 1600 is about the same as the i5 7500, but it has a much better workstation score (see 2nd link below) the r5 1600 can also be overclocked, on the stock cooler. The i5 7500 does not OC.

If you are considering spending £250 - £300 on an i7 6700/7700, you will be better off with a Ryzen 1700 or 1700x with its 8/16 layout OC'd to 3.8ghz. the i7s only really beat the r7 at gaming because games prefer higher clock speeds. My 2 cents a r7 1700 with a decent air cooler would be the best option for a single gpu 'budget' deep learning machine. An r7 should not 'bottleneck' a single 1080Ti. If you do go for an r7 for your needs, also get an x370 chipset motherboard. Also check the memory compatibility list for the motherboard you intend to get. 3200mhz ddr4 seams to be the sweet 'affordable spot' for ryzen 7. Ryzen loves fast memory.


https://www.quora.com/Should-I-buy-Ryzen-or-Intel-for-deep-learning-2xgpu-workstation

http://cpu.userbenchmark.com/Compare/Intel-Core-i5-7500-vs-AMD-Ryzen-5-1600/3648vs3919

r7 1700 has much better multicore preformance, and costs less.
http://cpu.userbenchmark.com/Compare/Intel-Core-i7-7700K-vs-AMD-Ryzen-7-1700/3647vs3917

 

[Will_49]

Given the seriousness of the situation, I would recommend that you go with the i7 7700 over the Ryzen. Intel still remains king in reliability. Never had any problems with my 7700k.

 

[Bungle11]

Given the seriousness of the situation, I would recommend that you go with the i7 7700 over the Ryzen. Intel still remains king in reliability. Never had any problems with my 7700k.

more cores and more threads are preferable for compute/ deep learning tasks. r7 1700 or 1700x would be better at the price point of a 7700k.

 

[Will_49]

Given the seriousness of the situation, I would recommend that you go with the i7 7700 over the Ryzen. Intel still remains king in reliability. Never had any problems with my 7700k.

more cores and more threads are preferable for compute/ deep learning tasks. r7 1700 or 1700x would be better at the price point of a 7700k.

I would have to agree with that.The RYZEN is probably best due to the amount of cores.

 

[shab_2008]

I would guess for deep learning more threads would be better. 6core/12 threads vs 4/4 (i5) vs 4/8(i7). I would say 6/12 is more advantageous than 4/4 on the i5. In your situation, as the cpu can spread the work load around. Out of the box the r5 1600 is about the same as the i5 7500, but it has a much better workstation score (see 2nd link below) the r5 1600 can also be overclocked, on the stock cooler. The i5 7500 does not OC.

If you are considering spending £250 - £300 on an i7 6700/7700, you will be better off with a Ryzen 1700 or 1700x with its 8/16 layout OC'd to 3.8ghz. the i7s only really beat the r7 at gaming because games prefer higher clock speeds. My 2 cents a r7 1700 with a decent air cooler would be the best option for a single gpu 'budget' deep learning machine. An r7 should not 'bottleneck' a single 1080Ti. If you do go for an r7 for your needs, also get an x370 chipset motherboard. Also check the memory compatibility list for the motherboard you intend to get. 3200mhz ddr4 seams to be the sweet 'affordable spot' for ryzen 7. Ryzen loves fast memory.


https://www.quora.com/Should-I-buy-Ryzen-or-Intel-for-deep-learning-2xgpu-workstation

http://cpu.userbenchmark.com/Compare/Intel-Core-i5-7500-vs-AMD-Ryzen-5-1600/3648vs3919

r7 1700 has much better multicore preformance, and costs less.
http://cpu.userbenchmark.com/Compare/Intel-Core-i7-7700K-vs-AMD-Ryzen-7-1700/3647vs3917

Thank you for your through answer. It really helped.
According to your descriptions, is it safe to say that it is very likely that Ryzen 5 1600 bottlenecks the GPU?
Thanks

 

[SENOR BURTOS]

Hi there,
I plan to work with some deep pretrained convolutional neural networks for an object-detection task.

From the CPUs you selected, the ryzen 5 1600 is the better choice.

The other two make no sense at all, they are 7th gen CPUs with very few cores, poor value and they are not upgradeable.

Go ryzen 5, ryzen 7 or get an 8th generation CPU, they have more cores and offer better performance for the price

If you are planning to get a $300 CPU like the 7700 you can afford a ryzen 1700/1700X or simply a core i7 8700/8700K

Minimum requirements? I don't think there's such thing. Mose people tend to undervalue all kinds of things. I'm pretty sure you can do deep learning with less than an i7, but obviously you won't get the same performance.

Again a ryzen 5 1600 is an excellent value, but if you work involves a good pay the ryzen 7 1700/1700X and their coffelake counterparts will pay for themselves eventually

 

[SENOR BURTOS]

bottlenecks exist everywhere, ther's no way you can avoid them completely. Indeed there's no such thing as "hardware bottlenekcs" they're all software bottlenecks.

It comes down to which appliation/taks you perform with your PC, in some programs the R5 1600 will be OK in others it simply won't

 

[Bungle11]

I would guess for deep learning more threads would be better. 6core/12 threads vs 4/4 (i5) vs 4/8(i7). I would say 6/12 is more advantageous than 4/4 on the i5. In your situation, as the cpu can spread the work load around. Out of the box the r5 1600 is about the same as the i5 7500, but it has a much better workstation score (see 2nd link below) the r5 1600 can also be overclocked, on the stock cooler. The i5 7500 does not OC.

If you are considering spending £250 - £300 on an i7 6700/7700, you will be better off with a Ryzen 1700 or 1700x with its 8/16 layout OC'd to 3.8ghz. the i7s only really beat the r7 at gaming because games prefer higher clock speeds. My 2 cents a r7 1700 with a decent air cooler would be the best option for a single gpu 'budget' deep learning machine. An r7 should not 'bottleneck' a single 1080Ti. If you do go for an r7 for your needs, also get an x370 chipset motherboard. Also check the memory compatibility list for the motherboard you intend to get. 3200mhz ddr4 seams to be the sweet 'affordable spot' for ryzen 7. Ryzen loves fast memory.


https://www.quora.com/Should-I-buy-Ryzen-or-Intel-for-deep-learning-2xgpu-workstation

http://cpu.userbenchmark.com/Compare/Intel-Core-i5-7500-vs-AMD-Ryzen-5-1600/3648vs3919

r7 1700 has much better multicore preformance, and costs less.
http://cpu.userbenchmark.com/Compare/Intel-Core-i7-7700K-vs-AMD-Ryzen-7-1700/3647vs3917

Thank you for your through answer. It really helped.
According to your descriptions, is it safe to say that it is very likely that Ryzen 5 1600 bottlenecks the GPU?
Thanks

the r5 1600 probably not. It will be far better than i5 7500 iwth its 4/4 c/t count. Personally, and if you were thinking of an i7 7700k because you can afford it, I would get a ryzen 1700 running a 3.8/3.9ghz instead with a good board and cooler. Its 8/16 C/T will be far better than the i7 for your needs. Also with your 1080Ti, a ryzen 7 will be better match. I would assume that can afford an r7, and you have a 1080Ti. So go for r7, with an ATX x370 board (So you can add a second gpu latter)

 

[Bungle11]

Hi there,
I plan to work with some deep pretrained convolutional neural networks for an object-detection task.

From the CPUs you selected, the ryzen 5 1600 is the better choice.

The other two make no sense at all, they are 7th gen CPUs with very few cores, poor value and they are not upgradeable.

Go ryzen 5, ryzen 7 or get an 8th generation CPU, they have more cores and offer better performance for the price

If you are planning to get a $300 CPU like the 7700 you can afford a ryzen 1700/1700X or simply a core i7 8700/8700K

Minimum requirements? I don't think there's such thing. Mose people tend to undervalue all kinds of things. I'm pretty sure you can do deep learning with less than an i7, but obviously you won't get the same performance.

Again a ryzen 5 1600 is an excellent value, but if you work involves a good pay the ryzen 7 1700/1700X and their coffelake counterparts will pay for themselves eventually

Decent motherboards for i7 8th gen chips are pretty expensive when compared to x370. Cost could spiral.

 

[SENOR BURTOS]

Decent motherboards for i7 8th gen chips are pretty expensive when compared to x370. Cost could spiral.

Even so, they are more suitable than current 7th gen CPUs

 

[Bungle11]

Decent motherboards for i7 8th gen chips are pretty expensive when compared to x370. Cost could spiral.

Even so, they are more suitable than current 7th gen CPUs

Fair point