Techspot：Core i7 7700k@4.9G vs. Ryzen 5 1600@4G with Vega 64 & GTX 1080

[coercitiv]

Could you point me to the post where I did all that?

 

[Zucker2k]

Actually, I 'liked' that post because it argues that the illusive "'Bang for buck' is subjective because it depends where on the performance curve you wish to be."
Adding to the above, the three biggest issues I personally have against the "best bang for buck" argument is that:
1. It freezes everything in time, and is therefore a shifting target as new chips are released and game codes change.
2. You're never getting the best result now.
3. You're never getting the best result tomorrow.

 

[AtenRa]

Since the Core i7 7700K is the best Gaming CPU today, i will say the R5 1600 currently is the best bang for the buck of the high end gaming CPUs.

 

[IRobot23]

Since the Core i7 7700K is the best Gaming CPU today, i will say the R5 1600 currently is the best bang for the buck of the high end gaming CPUs.

That is very relative. I would prefer i7 6800K or R5 1600 over i7 7700K.

 

[coercitiv]

That is very relative. I would prefer i7 6800K or R5 1600 over i7 7700K.

It may be relative for you or me, but it's not relative if we consider a larger group of people. The majority of both users and reviewers have decided that 7700K is today's best gaming CPU and R5 1600 is by far the best bang for the buck. Things will likely change come Coffee Lake, for one or maybe even both categories.

 

[IRobot23]

It may be relative for you or me, but it's not relative if we consider a larger group of people. The majority of both users and reviewers have decided that 7700K is today's best gaming CPU and R5 1600 is by far the best bang for the buck. Things will likely change come Coffee Lake, for one or maybe even both categories.

You are right. Also most PC users do not need i7 7700K.
What happens next year, if BF 2018 wants to push for 128player map? game physics is progressing really fast and it is well threaded.

 

[Dave2150]

Yup! That's why I upgraded all of my main desktop rigs to R5 1600. The value-for-money is really pretty-much unbeatable these days.

Why do you have more than one desktop?

 

[VirtualLarry]

Why do you have more than one desktop?

Why do you have more than one pair of shoes?

 

[Madpacket]

Why are so many people in denial when presented with clear evidence? Most of us who own Ryzen R5/R7 systems paired with LL 3200Mhz RAM understand there's virtually no gaming difference with a high-end GPU compared to a system with a high-speed Core i7. Many of us still own i7 systems or upgraded/switched from i7 to Ryzen and can confirm this. This isn't some conspiracy against Intel by Techspot. AMD pulled a rabbit out their ass with 1/10th the R&D budget and caught Intel off guard. Intel is now responding with Coffee Lake and it'll likely perform really well and that should force prices down across the board. This is good for everyone!

 

[raghu78]

Why are so many people in denial when presented with clear evidence? Most of us who own Ryzen R5/R7 systems paired with LL 3200Mhz RAM understand there's virtually no gaming difference with a high-end GPU compared to a system with a high-speed Core i7. Many of us still own i7 systems or upgraded/switched from i7 to Ryzen and can confirm this. This isn't some conspiracy against Intel by Techspot. AMD pulled a rabbit out their ass with 1/10th the R&D budget and caught Intel off guard. Intel is now responding with Coffee Lake and it'll likely perform really well and that should force prices down across the board. This is good for everyone!

Very well said. Ryzen performs very well with DDR4 3200 for gaming. Coffeelake is going to be a very strong response to Ryzen. AMD might have to make some price cuts till they can get Pinnacle Ridge out with higher clocks on Ryzen. AMD needs 10-12% higher clocks with Pinnacle Ridge to remain competitive against Coffeelake and even then I think AMD might have to price the high end Ryzen 2800x SKU at USD 430 - USD 450 to retain a significant perf/$ advantage over 7820x .

 

[Reinvented]

Why do you have more than one pair of shoes?

Wait, people have more than one pair of shoes?! Madness!

Honestly, I can't wait to see what else AMD has in store. Good times in 2017 and ahead. More competition, and better products all around.

 

[VirtualLarry]

Wait, people have more than one pair of shoes?! Madness!

I guess I'm like the Emielda Marcos of PCs...

 

[Carfax83]

Because the 7700K doesn't scale as well with memory as Ryzen does.

All dual channel CPUs gain significant amounts of performance from faster memory, including the 7700K. The reason why the 7700K may not gain as much as Ryzen, is probably because it has a more efficient, lower latency memory controller so the gains aren't as drastic (this goes for losses from using slower RAM as well), but they are still present.

In fact, here is a review which shows RAM speed scaling between several Intel architectures from Haswell to Broadwell-C to Skylake, and while the scaling isn't as great as with Ryzen, it's still significant with the exception of Broadwell-C due to it having a massive L4 cache which mimics high speed RAM.

I've done memory speed testing on my 6900K X99 platform, and the scaling is very slight or non existent most of the time since it uses quad channel memory and L3 cache is much bigger than in the mainstream CPUs. Still, the hardware enthusiast in me loves to have fast RAM

 

[Carfax83]

Why are so many people in denial when presented with clear evidence? Most of us who own Ryzen R5/R7 systems paired with LL 3200Mhz RAM understand there's virtually no gaming difference with a high-end GPU compared to a system with a high-speed Core i7.

Probably because the evidence isn't as clear as you think? The test that the OP presented from Techspot is decent, but it's obvious that the 4.9ghz Core i7 is being bottlenecked by both the Vega 64 and the GTX 1080 FE. They should have used a faster GPU like the 1080 Ti with more headroom.

Or better yet, they should have tested at 720p so they could load the CPUs more. I understand that testing at 1080p and 1440p presents a more realistic scenario, but if you want to test a CPUs gaming performance, it's always better to test at lower resolutions so that the burden is placed on the CPU where it belongs.

Also, frame latency testing should be in order as well, as that is a more accurate depiction of what is actually going on.

 

[IRobot23]

All dual channel CPUs gain significant amounts of performance from faster memory, including the 7700K. The reason why the 7700K may not gain as much as Ryzen, is probably because it has a more efficient, lower latency memory controller so the gains aren't as drastic (this goes for losses from using slower RAM as well), but they are still present.

In fact, here is a review which shows RAM speed scaling between several Intel architectures from Haswell to Broadwell-C to Skylake, and while the scaling isn't as great as with Ryzen, it's still significant with the exception of Broadwell-C due to it having a massive L4 cache which mimics high speed RAM.

I've done memory speed testing on my 6900K X99 platform, and the scaling is very slight or non existent most of the time since it uses quad channel memory and L3 cache is much bigger than in the mainstream CPUs. Still, the hardware enthusiast in me loves to have fast RAM

Has has most efficient memory controller. I thing that Ryzen 2 with 4000MT/s DDR4 we could see over 60GB/s with dual channel ram for the first time.

Only problem to the latency is NB (Datafabric) speed, try having NB clocked at 1600MHz with i7 7700K. Otherwise Ryzen will have better write/copy speeds...
http://www.guru3d.com/articles-pages/amd-ryzen-5-1500x-and-1600x-review,16.html

problem is latency.

 

[Carfax83]

Has has most efficient memory controller. I thing that Ryzen 2 with 4000MT/s DDR4 we could see over 60GB/s with dual channel ram for the first time.

Sorry, not convinced because latency is a crucial aspect of overall memory efficiency. I did some checking and I found that the infinity fabric interconnect runs at the same speed as the memory controller, which explains why Ryzen is so sensitive to memory speed. That to me is bad design, if you make something as important as the memory controller run at only 1333Mhz, since the highest officially supported memory speed for Ryzen is DDR4 2666.

If getting the most out of Ryzen requires super fast RAM, the platform may actually end up being more expensive than a comparable Intel solution, since very fast RAM ain't cheap these days. AMD definitely needs to fix that problem in Zen 2 if they are serious about competing with Intel.

 

[tamz_msc]

Sorry, not convinced because latency is a crucial aspect of overall memory efficiency. I did some checking and I found that the infinity fabric interconnect runs at the same speed as the memory controller, which explains why Ryzen is so sensitive to memory speed. That to me is bad design, if you make something as important as the memory controller run at only 1333Mhz, since the highest officially supported memory speed for Ryzen is DDR4 2666.

If getting the most out of Ryzen requires super fast RAM, the platform may actually end up being more expensive than a comparable Intel solution, since very fast RAM ain't cheap these days. AMD definitely needs to fix that problem in Zen 2 if they are serious about competing with Intel.

Ryzen actually incentivizes the need to get fast memory. The Ryzen IMC gets closer to peak theoretical bandwidth as measured by AIDA64 than Kaby Lake IMC, and you don't need super-high end 4000MHz RAM because the best Ryzen can do is around 3600MHz CL16. Intel chips also scale with memory speeds, but the scaling is weaker than Ryzen.

Memory always runs at JEDEC specs which is half the MT/s. The speed of the IF links is the same as the actual DRAM frequency as measured by CPU-Z.

 

[epsilon84]

Ryzen actually incentivizes the need to get fast memory. The Ryzen IMC gets closer to peak theoretical bandwidth as measured by AIDA64 than Kaby Lake IMC, and you don't need super-high end 4000MHz RAM because the best Ryzen can do is around 3600MHz CL16. Intel chips also scale with memory speeds, but the scaling is weaker than Ryzen.

Memory always runs at JEDEC specs which is half the MT/s. The speed of the IF links is the same as the actual DRAM frequency as measured by CPU-Z.

This is a double edged sword, due to the currently inflated prices of DDR4. I actually found a very good deal online for a used 1700X/X370 combo for not much more than the cost of a 1700X by itself, I was really tempted as I haven't built an AMD rig in forever and wanted to play around with Ryzen, but then I checked the prices of LL DDR4 3466 and that really put me off.

If Ryzen actually performed within a few percent of top end DDR4 3466/3600 with say, regular DDR4 2666 I probably would have jumped on board, but I couldn't stomach the price jump to the higher end DDR4 kits. Hopefully RAM prices will settle down soon but I'm not holding my breath.

 

[tamz_msc]

This is a double edged sword, due to the currently inflated prices of DDR4. I actually found a very good deal online for a used 1700X/X370 combo for not much more than the cost of a 1700X by itself, I was really tempted as I haven't built an AMD rig in forever and wanted to play around with Ryzen, but then I checked the prices of LL DDR4 3466 and that really put me off.

If Ryzen actually performed within a few percent of top end DDR4 3466/3600 with say, regular DDR4 2666 I probably would have jumped on board, but I couldn't stomach the price jump to the higher end DDR4 kits. Hopefully RAM prices will settle down soon but I'm not holding my breath.

You don't need 3466LL, those are achievable only with 3600 CL16 kits. 3200CL14 Samsung B-die works with Auto XMP on latest BIOS, and it's not that much more expensive:
GSkill FlareX 3200CL14(Samsung B-die), 180$
GSkill RipJaws V 3200CL16(Hynix), 139$. These will do 2933.

 

[IRobot23]

What are you guys talking about,

Ryzen has most efficient IMC on market. That is a fact!

Sorry, not convinced because latency is a crucial aspect of overall memory efficiency. I did some checking and I found that the infinity fabric interconnect runs at the same speed as the memory controller, which explains why Ryzen is so sensitive to memory speed. That to me is bad design, if you make something as important as the memory controller run at only 1333Mhz, since the highest officially supported memory speed for Ryzen is DDR4 2666.

If getting the most out of Ryzen requires super fast RAM, the platform may actually end up being more expensive than a comparable Intel solution, since very fast RAM ain't cheap these days. AMD definitely needs to fix that problem in Zen 2 if they are serious about competing with Intel.

So you say that IMC is bad, because ryzens NB (DF - not IF) is clocked lower?
Who

DF links to the IMC. Again DF = NB!

Here is Ryzen 3600MT/s vs intel 3866MT/s ddr4 speed ( got over 53GB/s with 3466MT/s and really good timings)

So basically Ryzen IMC at 1500MHz@3000MT/s ddr4 is almost as fast as intel 1833MHz@3866MHz...
Please put I7 7700K NB to IMC speeds and try to compare it to ryzen. IF is blessing (Control fabric and Data fabric = NB).

Check out how much can FX do at 1600MT/s. I get over 50GB/s at 3200MHz...

I beg you to not talk about poor IMC, if link to it make all latency that even then at 1600MHz R7 1800X goes near 60ns at 4GHz (which is really good for an 8 core with lots of cache).

 

[Carfax83]

Ryzen actually incentivizes the need to get fast memory. The Ryzen IMC gets closer to peak theoretical bandwidth as measured by AIDA64 than Kaby Lake IMC, and you don't need super-high end 4000MHz RAM because the best Ryzen can do is around 3600MHz CL16. Intel chips also scale with memory speeds, but the scaling is weaker than Ryzen.

Yes, but like I told the other guy, bandwidth is only half of the equation. The other half is latency, and there Ryzen falls flat on its face. Latency is actually more important for performance for most consumer applications (gaming included) than bandwidth.

You can't say that Ryzen's memory controller is more efficient than Intel's when it has such poor latency.

 

[Carfax83]

What are you guys talking about,

Ryzen has most efficient IMC on market. That is a fact!

Is it now?

So you say that IMC is bad, because ryzens NB (DF - not IF) is clocked lower?
Who

I was implying that the reason for Ryzen's ridiculously high memory latency is because the memory controller runs at such a low clock speed. If the memory controller runs at the same speed as the memory, then that might possibly explain the high latencies.

With Intel CPUs, the memory controller runs at a set clock speed which is typically faster than the memory itself. On my CPU, I think the MC runs at 2800mhz at stock clocks, (though I currently have mine set to 3400mhz) which is much faster than the fastest officially supported memory that my CPU supports; DDR4 2400 or 1200mhz.

DF links to the IMC. Again DF = NB!

Do you even know what a North Bridge is? Obviously you don't, otherwise you wouldn't be trying to compare infinity fabric to it.

Here is Ryzen 3600MT/s vs intel 3866MT/s ddr4 speed ( got over 53GB/s with 3466MT/s and really good timings)

So basically Ryzen IMC at 1500MHz@3000MT/s ddr4 is almost as fast as intel 1833MHz@3866MHz...
Please put I7 7700K NB to IMC speeds and try to compare it to ryzen. IF is blessing (Control fabric and Data fabric = NB).

Look, I'll tell you again one last time. Memory bandwidth is only half of what makes up memory performance. So citing high bandwidth scores is cool and all, but if the latency scores are abysmal, then you can't claim that Ryzen has the most efficient IMC on the market now can you?

 

[tamz_msc]

Yes, but like I told the other guy, bandwidth is only half of the equation. The other half is latency, and there Ryzen falls flat on its face. Latency is actually more important for performance for most consumer applications (gaming included) than bandwidth.

You can't say that Ryzen's memory controller is more efficient than Intel's when it has such poor latency.

I meant efficiency in the sense that how close one can get to the peak memory bandwidth based on DRAM frequency. Latency is no doubt important in a variety of applications, but bandwidth efficiency is also important in things like memory bound HPC applications. Both Ryzen and Intel desktop chips(non-HEDT, excluding Xeons as well) have the property that a single core can utilize almost all of the available bandwidth.

 

[IRobot23]

Is it now?



I was implying that the reason for Ryzen's ridiculously high memory latency is because the memory controller runs at such a low clock speed. If the memory controller runs at the same speed as the memory, then that might possibly explain the high latencies.

With Intel CPUs, the memory controller runs at a set clock speed which is typically faster than the memory itself. On my CPU, I think the MC runs at 2800mhz at stock clocks, (though I currently have mine set to 3400mhz) which is much faster than the fastest officially supported memory that my CPU supports; DDR4 2400 or 1200mhz.



Do you even know what a North Bridge is? Obviously you don't, otherwise you wouldn't be trying to compare infinity fabric to it.



Look, I'll tell you again one last time. Memory bandwidth is only half of what makes up memory performance. So citing high bandwidth scores is cool and all, but if the latency scores are abysmal, then you can't claim that Ryzen has the most efficient IMC on the market now can you?

Only for you:
1. The latency that is not produced by IMC on RYZEN
2. Latency is cause by poor clocked NB = data fabric

Why are you comparing i7 7700K with latency?
FX 8350 has lower latency that Ryzen.

So does i7 7820X has poor IMC because of latency? You can check i7 6900K will have lower latency
DDR4 latency is not cause only by IMC, NB (DF)... is basically everything on its way.

Why is IF (Data fabric) blessing and curse?
This test tells everything

100 GB/s

 

[Carfax83]

Both Ryzen and Intel desktop chips(non-HEDT, excluding Xeons as well) have the property that a single core can utilize almost all of the available bandwidth.

Yep, and that's where the cache comes in. The cache hierarchies have way more bandwidth and far lower latencies than ram, which is why CPUs are designed to actually minimize RAM access as much as possible because it's so expensive in terms of delay. But depending on the application, cache misses will occur very often and the CPU will have to go to RAM.