My 4070 Ti + i5 13600K[F] gained over 50 FPS in game after upgrading & manually tuning the RAM

O

OwlyEagle

Who cares?

Intel Core i5-13600KF @ Stock
GeForce RTX 4070 Ti Phoenix GS @3Ghz
2x 32GB - G.Skill Ripjaws V F4-4000C18Q-128GVK
Memory is manually tuned @ 17-21-21-41 1.45v
Dual Rank, Dual Channel - Gear 1 @ CR1
Hynix MJR IC - VDDQ & VCCSA @ 1.4v

Old ram @ XMP profile: HyperX Fury black HX432C16FB3K2/16

Results (after tuning):

www.3dmark.com

I scored 22 424 in Time Spy

Intel Core i5-13600KF Processor, NVIDIA GeForce RTX 4070 Ti x 1, 65536 MB, 64-bit Windows 11}
www.3dmark.com www.3dmark.com
www.3dmark.com

I scored 22 391 in Time Spy

Intel Core i5-13600KF Processor, NVIDIA GeForce RTX 4070 Ti x 1, 65536 MB, 64-bit Windows 11}
www.3dmark.com www.3dmark.com

Gameplay (before):

Gameplay (after):

Looking for feedback to improve on.
1.png2.pngScreenshot 2023-01-31 070111.pngScreenshot 2023-01-31 065036.png3.png
 
With a stock 13600K and 6700XT my CPU score is 18393. DDR5-5600 32-36-36-76-112 2T.

Overall score is 13857 because my GPU is much weaker.
 
Memory latency will have a big impact on a lot of games. Wasting cpu cycles waiting for data is a cpu killer. One reason I dislike a lot of cpu reviews is they test with stock/recommended memory specs which are a lot lower than what you can get if you max out. Intel cpus tend to allow lower latency and faster memory so if you’re willing to put the time in you can get a lot more out of the Intel platform for gaming. 5800x3d is interesting because it kind of mitigates latency with a huge cache. I think a lot of cpu reviews are done poorly and will give false impressions of how well a cpu will perform long term.
 
Scott_Arm said:
One reason I dislike a lot of cpu reviews is they test with stock/recommended memory specs which are a lot lower than what you can get if you max out.
Click to expand...
The problem with that is it''s going to vary depending on ram manufacturer model number and the type of motherboard used
and if recommended speed for a cpu is 4800mhz ddr5 but you can also buy 5200,5600,6000,6200 what do you test with
 
Davros said:
The problem with that is it''s going to vary depending on ram manufacturer model number and the type of motherboard used
and if recommended speed for a cpu is 4800mhz ddr5 but you can also buy 5200,5600,6000,6200 what do you test with
Click to expand...

I think reviewers focus too much on having 100 different cpus in their reviews, vs having a smaller number showing both stock and optimized. They also mix in games that are gpu limited for some reason, or running at gpu limited settings (even at 1080p ultra) for reasons I cannot understand.
 
This is Interesting
1676490711274.png

all the timings on the 2nd one are tighter apart from the final number 96vs112 yet it seems to be the final number that matters (what does that number represent)
 
Thanks to you guys I decided to mess with my RAM clocks, but I couldn't get even a slight OC or tighter timings on it. The consequence of cheapening out on RAM I guess lol.
 
Reynaldo said:
Thanks to you guys I decided to mess with my RAM clocks, but I couldn't get even a slight OC or tighter timings on it. The consequence of cheapening out on RAM I guess lol.
Click to expand...
Yeah I've never had any luck with optimizing my RAM. Even with first a 16Gb set and now a 32Gb set, both Samsung B-die, I can barely get them stable at normal speeds.
 
Reynaldo said:
Thanks to you guys I decided to mess with my RAM clocks, but I couldn't get even a slight OC or tighter timings on it. The consequence of cheapening out on RAM I guess lol.
Click to expand...

Malo said:
Yeah I've never had any luck with optimizing my RAM. Even with first a 16Gb set and now a 32Gb set, both Samsung B-die, I can barely get them stable at normal speeds.
Click to expand...

Achievable memory speeds are also very dependent on the CPU and motherboard (yes even with the IMCs now on the CPU, it's still motherboard dependent).

For example in the HUB video there is a reason why they only show DDR5 7200 for Raptor Lake and not Zen 4. Zen 4's top end is essentially 6000 (and the reports are mixed if the cheaper, for AM5 anyways, motherboards can even hit this consistently). While Raptor Lake on the 7xx series boards can hit 7200.

Some of the higher speeds you see sometimes being posted also leave out that they do need to increase voltage into either the memory or IMC (or equivalent terminology, it's slightly different) or both. The famed B-die for DDR4 for example is known for being able to achieve higher speeds/low latency but sometimes it gets left out that it's because it has very good voltage scaling.
 
arandomguy said:
Achievable memory speeds are also very dependent on the CPU and motherboard (yes even with the IMCs now on the CPU, it's still motherboard dependent).

For example in the HUB video there is a reason why they only show DDR5 7200 for Raptor Lake and not Zen 4. Zen 4's top end is essentially 6000 (and the reports are mixed if the cheaper, for AM5 anyways, motherboards can even hit this consistently). While Raptor Lake on the 7xx series boards can hit 7200.

Some of the higher speeds you see sometimes being posted also leave out that they do need to increase voltage into either the memory or IMC (or equivalent terminology, it's slightly different) or both. The famed B-die for DDR4 for example is known for being able to achieve higher speeds/low latency but sometimes it gets left out that it's because it has very good voltage scaling.
Click to expand...

Yah, I think ITX motherboards with two slots for DDR4 were considered good for memory overclocking, at least with certain chipsets. I guess signal integrity on those boards can be a bit better. It's definitely not easy to push the limits of DDR4, though I wouldn't say it's incredibly hard with B-die. It'll mostly be limited by the motherboard and not the RAM itself with b-die. That said, I think a lot of people could get decent performance increases with a little knowledge and some careful purchasing. A little time and effort into knowing which motherboards and which dimms to buy really help.
 
Davros said:
Is it still the case that you want to keep memory speeds in sync with the speed of the infinity fabric ?
Click to expand...

Are you asking about Zen 4? AMDs official stance is to leave fabric speeds at "Auto" and use DDR5 6000 (which represents the practical limit with current Zen 4/Mobos), which results in an async ratio and that there's accommodations in design for this compared to earlier. But I don't think anyone's tested in depth to see what the behavior actually is with different combinations. Not that there's really much practical configuration variants.

For Zen 3 and earlier you'd still want to leave it in sync in practice outside of maybe some edge cases/configurations.
 
Davros said:
This is Interesting
View attachment 8325

all the timings on the 2nd one are tighter apart from the final number 96vs112 yet it seems to be the final number that matters (what does that number represent)
Click to expand...
2nd to last one looks like tRAS to me, although it's normally expressed as the 4th timing. (96 and 76 are common XMP values for that timing, though!)
last one looks like tRc to me as 96 and 112 are pretty common there.

DDR5 seems to be a lot more sensitive to secondary timings than primary timings compared to DDR4, in particular CAS latency seems to matter much less than it used to.
It's really easy to get a DDR5 profile that has worse primary timings but better secondary timings and have it perform significantly better overall.

<--- as one example
 
And if you have 'good' (Hynix) DDR5 and want to optimize it on your Ryzen 7000 series, this is a really good start:

 
You must log in or register to reply here.
Back
Top