Upvote Upvoted 114 Downvote Downvoted
1 ⋅⋅ 18 19 20 21
TF2 benchmarks
601
#601
0 Frags +
kamild1996_v2500 reboots later (not really, just under a hundred actually; don't do B450 boards, kids), here I am testing the 3700X. Results are in.

Setup details:
- Windows 10 1903
- GeForce driver 431.60
- GeForce Experience with replay enabled, no game optimization
- game details: mastercomfig v6 latest, 1080p, DX9, mastercomfig's launch parameters, mods: budhud, Less Distracting Explosions, mastercomfig addons (mouse tweaks, no pyroland, no soundscapes, no tutorial)
- tested with timedemo_runcount set at 3, so stuff being loaded in doesn't lower the results

Since there's a ton of data, I figured it'd be better to show it all in a table, and since TFTV doesn't have tables... Here's a Google Docs link: https://docs.google.com/spreadsheets/d/1AGtyRBvgcpYv2T8h9bSgx9IDd8fXFES175tPrsK5hMI/edit?usp=sharing

whitepuzzle, I think you'll be happy with the results ;)

I was worried that Zen's higher memory latency compared to CFL would hurt TF2 performance (seeing how it's so sensitive to memory frequency and timings) but it looks like it doesn't matter too much.

At this rate, Zen 3 should be straight up faster than CFL, if not later batch Zen 2s with a more mature process and better clocks.

[quote=kamild1996_v2]500 reboots later (not really, just under a hundred actually; don't do B450 boards, kids), here I am testing the 3700X. Results are in.

Setup details:
- Windows 10 1903
- GeForce driver 431.60
- GeForce Experience with replay enabled, no game optimization
- game details: mastercomfig v6 latest, 1080p, DX9, mastercomfig's launch parameters, mods: budhud, Less Distracting Explosions, mastercomfig addons (mouse tweaks, no pyroland, no soundscapes, no tutorial)
- tested with timedemo_runcount set at 3, so stuff being loaded in doesn't lower the results

Since there's a ton of data, I figured it'd be better to show it all in a table, and since TFTV doesn't have tables... Here's a Google Docs link: https://docs.google.com/spreadsheets/d/1AGtyRBvgcpYv2T8h9bSgx9IDd8fXFES175tPrsK5hMI/edit?usp=sharing

[b]whitepuzzle[/b], I think you'll be happy with the results ;)[/quote]

I was worried that Zen's higher memory latency compared to CFL would hurt TF2 performance (seeing how it's so sensitive to memory frequency and timings) but it looks like it doesn't matter too much.


At this rate, Zen 3 should be straight up faster than CFL, if not later batch Zen 2s with a more mature process and better clocks.
602
#602
3 Frags +

Well thanks to AMD not locking the cheaper chipsets down to 2666 MHz like Intel and people being aware that Zen needs a bit more than that to get decent latency it's not nearly as bad as with the Zen1 launch.

Would be great to see a comparision with 2667, 3200, 3733 and possibly higher though.

Maybe we'll get an N7+ (7nm + EUV) refresh similar to Zen+ (14->12nm which is just 14+) next year, that could be interesting.

Well thanks to AMD not locking the cheaper chipsets down to 2666 MHz like Intel and people being aware that Zen needs a bit more than that to get decent latency it's not nearly as bad as with the Zen1 launch.

Would be great to see a comparision with 2667, 3200, 3733 and possibly higher though.

Maybe we'll get an N7+ (7nm + EUV) refresh similar to Zen+ (14->12nm which is just 14+) next year, that could be interesting.
603
#603
0 Frags +

According to roadmaps there is no Zen 2 refresh planned, it's straight on to Zen 3.

According to roadmaps there is no Zen 2 refresh planned, it's straight on to Zen 3.
604
#604
2 Frags +

Which roadmaps? A refresh probably wouldn't be listed as architecture.
There's a lot of options. Zen3 is probably simply not be very different (Zen+ was literally identical except for the L2 fix so it couldn't be called Zen2), otherwise AMD wouldn't be able to release it just a year later. It is listed as "optimization" just like Zen+ for a reason. And just because Zen3 exists doesn't mean the whole Desktop lineup will get Zen3. E.g. no 2200, no 2400, 2300X and 2500X OEM only, no 2800X and Ryzen 3000 APUs are still Zen+. It's entirely possible for example that they'll do a die with more cores to bump up the per socket core count, but can't use that for the 6/8 core SKUs because it would mean disabling half the cores. Or that year desktop gets the short end of the stick and APUs get the new architecture first. EPYC only got 2 SKUs with the obligatory clock bump last year so after the big Zen2 update this year they might sit out next year again.

Let's just wait and see.

Which roadmaps? A refresh probably wouldn't be listed as architecture.
There's a lot of options. Zen3 is probably simply not be very different (Zen+ was literally identical except for the L2 fix so it couldn't be called Zen2), otherwise AMD wouldn't be able to release it just a year later. It is listed as "optimization" just like Zen+ for a reason. And just because Zen3 exists doesn't mean the whole Desktop lineup will get Zen3. E.g. no 2200, no 2400, 2300X and 2500X OEM only, no 2800X and Ryzen 3000 APUs are still Zen+. It's entirely possible for example that they'll do a die with more cores to bump up the per socket core count, but can't use that for the 6/8 core SKUs because it would mean disabling half the cores. Or that year desktop gets the short end of the stick and APUs get the new architecture first. EPYC only got 2 SKUs with the obligatory clock bump last year so after the big Zen2 update this year they might sit out next year again.

Let's just wait and see.
605
#605
1 Frags +

The roadmaps listed Zen+ but show Zen 3 coming straight after Zen 2.

https://www.anandtech.com/show/12233/amd-tech-day-at-ces-2018-roadmap-revealed-with-ryzen-apus-zen-on-12nm-vega-on-7nm

The roadmaps listed Zen+ but show Zen 3 coming straight after Zen 2.

https://www.anandtech.com/show/12233/amd-tech-day-at-ces-2018-roadmap-revealed-with-ryzen-apus-zen-on-12nm-vega-on-7nm
606
#606
0 Frags +

where the 5700 benchmarks at woop woop

where the 5700 benchmarks at woop woop
607
#607
4 Frags +
sniffwhere the 5700 benchmarks at woop woop

it's a gpu so it wont do anything to tf2 performance whatsoever (unless you're comparing it to apu)

[quote=sniff]where the 5700 benchmarks at woop woop[/quote]
it's a gpu so it wont do anything to tf2 performance whatsoever (unless you're comparing it to apu)
608
#608
3 Frags +

#605
Sorry I didn't reply sooner, but read again what I wrote.
Zen2 was on the roadmap for 2019 and the 2019 APUs are ... Zen+.
Zen+ was on the roadmap for 2018 and EPYC got ... nothing.

The roadmap only implies that there will be at least one Zen3 SKU in 2020, not that everything will be replaced.
It could be anything from no new Desktop SKUs, a 7nm refresh, 7nm+ refresh to a full Zen3 lineup.

There are no detailed roadmaps split by segment for AMD yet afaik. Intel does that and throws around more codenames than anyone can keep track of. Cascade Lake for this, Whiskey Lake for that, Amber Lake here and some Ice Lake sprinkled in there. AMD doesn't do that, they might not be sure yet themselves.

#605
Sorry I didn't reply sooner, but read again what I wrote.
Zen2 was on the roadmap for 2019 and the 2019 APUs are ... Zen+.
Zen+ was on the roadmap for 2018 and EPYC got ... nothing.

The roadmap only implies that there will be at least one Zen3 SKU in 2020, not that everything will be replaced.
It could be anything from no new Desktop SKUs, a 7nm refresh, 7nm+ refresh to a full Zen3 lineup.

There are no detailed roadmaps split by segment for AMD yet afaik. Intel does that and throws around more codenames than anyone can keep track of. Cascade Lake for this, Whiskey Lake for that, Amber Lake here and some Ice Lake sprinkled in there. AMD doesn't do that, they might not be sure yet themselves.
609
#609
0 Frags +
SetsulWould be great to see a comparision with 2667, 3200, 3733 and possibly higher though.

A little something for you, then.

Now that the BIOSes have matured a fair bit, I decided to OC my RAM to see how much of a difference would that make. Couldn't overclock my sticks as much as I hoped for, but there's something at least.
As a reminder: R7 3700X (PBO enabled), Win10 1903, 16 GB RAM; customized mastercomfig medium preset (afaik it's just textures set to very high and ragdolls enabled); DX9; launch parameters recommended by mastercomfig; using benchmark_test (the newer one)

Before (XMP profile, 3000 MHz CL16):

4812 frames 21.475 seconds 224.07 fps ( 4.46 ms/f) 31.826 fps variability

After (OC, 3466 MHz CL16):

4812 frames 20.741 seconds 232.01 fps ( 4.31 ms/f) 31.111 fps variability

(got screenshots of before/after with CPU-Z window open, I can send them if anyone wants to)

Just about 3.5% of difference. Whether or not it's worth the hassle and the potentially shorter lifespan I'll leave to your judgement ¯\_(ツ)_/¯

[quote=Setsul]Would be great to see a comparision with 2667, 3200, 3733 and possibly higher though.[/quote]
A little something for you, then.

Now that the BIOSes have matured a fair bit, I decided to OC my RAM to see how much of a difference would that make. Couldn't overclock my sticks as much as I hoped for, but there's something at least.
As a reminder: R7 3700X (PBO enabled), Win10 1903, 16 GB RAM; customized mastercomfig medium preset (afaik it's just textures set to very high and ragdolls enabled); DX9; launch parameters recommended by mastercomfig; using benchmark_test (the newer one)

Before (XMP profile, 3000 MHz CL16):
[code]4812 frames 21.475 seconds 224.07 fps ( 4.46 ms/f) 31.826 fps variability[/code]

After (OC, 3466 MHz CL16):
[code]4812 frames 20.741 seconds 232.01 fps ( 4.31 ms/f) 31.111 fps variability[/code]

(got screenshots of before/after with CPU-Z window open, I can send them if anyone wants to)

Just about 3.5% of difference. Whether or not it's worth the hassle and the potentially shorter lifespan I'll leave to your judgement ¯\_(ツ)_/¯
610
#610
0 Frags +

benchmark1
2639 frames 13.942 seconds 189.29 fps ( 5.28 ms/f) 12.811 fps variability

benchmark3
22256 frames 83.923 seconds 265.19 fps ( 3.77 ms/f) 24.588 fps variability

benchmark_test
4812 frames 20.341 seconds 236.56 fps ( 4.23 ms/f) 36.665 fps variability

Ryzen 7 3700X
16GB DDR4-3200 CL16
MSI B450M Gaming Plus (27/08/2019 BIOS)
RX 5700
Dxlevel 90
1920x1080 fullscreen
Config: none

I've done no optimizations or tweaking to this system yet so I'm sure more frames can be pulled out once that's done and If I use an appropriate config

benchmark1
2639 frames 13.942 seconds 189.29 fps ( 5.28 ms/f) 12.811 fps variability

benchmark3
22256 frames 83.923 seconds 265.19 fps ( 3.77 ms/f) 24.588 fps variability

benchmark_test
4812 frames 20.341 seconds 236.56 fps ( 4.23 ms/f) 36.665 fps variability

Ryzen 7 3700X
16GB DDR4-3200 CL16
MSI B450M Gaming Plus (27/08/2019 BIOS)
RX 5700
Dxlevel 90
1920x1080 fullscreen
Config: none

I've done no optimizations or tweaking to this system yet so I'm sure more frames can be pulled out once that's done and If I use an appropriate config
611
#611
1 Frags +

Just upgraded because my brother's PC died and i gave him my CPU.
benchmark1

Ryzen 7 2700x PBO maxed + undervolt
DDR4 3200 CL16
Asus ROG STRIX X470-F GAMING BIOS 5220 AGESA 1.0.0.3ABBA
Vega 56 with 64 bios Undervolted 985mv +50% power limit 1632 mhz core 1050mhz mem
Windows 10 1903
mastercoms medium, no rag dolls/gibs, 4xaa, 8xaf, mat_picmip -1, dxlevel 95

2639 frames 13.445 seconds 196.28 fps ( 5.09 ms/f) 12.436 fps variability

Ryzen 7 3700x PBO maxed
DDR4 3466 CL16
Asus ROG STRIX X470-F GAMING BIOS 5220 AGESA 1.0.0.3ABBA
Vega 56 with 64 bios Undervolted 985mv +50% power limit 1632 mhz core 1050mhz mem
Windows 10 1903
mastercoms medium, no rag dolls/gibs, 4xaa, 8xaf, mat_picmip -1, dxlevel 95

2639 frames 10.470 seconds 252.06 fps ( 3.97 ms/f) 17.821 fps variability

Ryzen 7 3700x PBO maxed
DDR4 ‭3534‬ CL16
Asus ROG STRIX X470-F GAMING BIOS 5220 AGESA 1.0.0.3ABBA
Vega 56 with 64 bios Undervolted 985mv +50% power limit 1632 mhz core 1050mhz mem
Windows 10 1903
mastercoms medium, no rag dolls/gibs, 4xaa, 8xaf, mat_picmip -1, dxlevel 95

2639 frames 9.685 seconds 272.47 fps ( 3.67 ms/f) 19.875 fps variability

Just upgraded because my brother's PC died and i gave him my CPU.
benchmark1

Ryzen 7 2700x PBO maxed + undervolt
DDR4 3200 CL16
Asus ROG STRIX X470-F GAMING BIOS 5220 AGESA 1.0.0.3ABBA
Vega 56 with 64 bios Undervolted 985mv +50% power limit 1632 mhz core 1050mhz mem
Windows 10 1903
mastercoms medium, no rag dolls/gibs, 4xaa, 8xaf, mat_picmip -1, dxlevel 95

2639 frames 13.445 seconds 196.28 fps ( 5.09 ms/f) 12.436 fps variability

Ryzen 7 3700x PBO maxed
DDR4 3466 CL16
Asus ROG STRIX X470-F GAMING BIOS 5220 AGESA 1.0.0.3ABBA
Vega 56 with 64 bios Undervolted 985mv +50% power limit 1632 mhz core 1050mhz mem
Windows 10 1903
mastercoms medium, no rag dolls/gibs, 4xaa, 8xaf, mat_picmip -1, dxlevel 95

2639 frames 10.470 seconds 252.06 fps ( 3.97 ms/f) 17.821 fps variability

Ryzen 7 3700x PBO maxed
DDR4 ‭3534‬ CL16
Asus ROG STRIX X470-F GAMING BIOS 5220 AGESA 1.0.0.3ABBA
Vega 56 with 64 bios Undervolted 985mv +50% power limit 1632 mhz core 1050mhz mem
Windows 10 1903
mastercoms medium, no rag dolls/gibs, 4xaa, 8xaf, mat_picmip -1, dxlevel 95

2639 frames 9.685 seconds 272.47 fps ( 3.67 ms/f) 19.875 fps variability
1 ⋅⋅ 18 19 20 21
Please sign in through STEAM to post a comment.