3D Mark announces Steel Nomad benchmark, replaces Timespy

[doodah]

Not bad on my 3090.

 

[L0rdMike]

7900xt

Mem running at 2900mhz

I scored 6 334 in Steel Nomad

AMD Ryzen 7 7800X3D, AMD Radeon RX 7900 XT x 1, 32768 MB, 64-bit Windows 11}

www.3dmark.com

 

[wookiee87]

got around to running it but noticing low gpu core clocks when running the benchmark

gpu was oc'd to 2950mhz min, 3225mhz max, ram was running at 2774mhz, +15% power limit

I scored 7 097 in Steel Nomad

AMD Ryzen 9 7950X, AMD Radeon RX 7900 XTX x 1, 32768 MB, 64-bit Windows 11}

www.3dmark.com

but gpu core was around the 2.4ghz region, far below what i had set as the base speed, any reason for this or is it a bug?

 

[Audioboxer]

got around to running it but noticing low gpu core clocks when running the benchmark

gpu was oc'd to 2950mhz min, 3225mhz max, ram was running at 2774mhz, +15% power limit

I scored 7 097 in Steel Nomad

AMD Ryzen 9 7950X, AMD Radeon RX 7900 XTX x 1, 32768 MB, 64-bit Windows 11}

www.3dmark.com

but gpu core was around the 2.4ghz region, far below what i had set as the base speed, any reason for this or is it a bug?

Have you got HWINFO or something running? Check what power usage goes on when running the test.

 

[wookiee87]

just re-ran at stock with hwinfo up, gpu power is running at 347w with maximum power at 495w but front end clock is stuck around 2426mhz, gpu core voltage was at 0.845mv too

scored 6418

re launched with overclocked settings this time 3100mhz min, 3200mhz max, 2776mhz ram +15% pwr, card was pulling 398w with spikes 522w, gpu clock in hwinfo reported at 2640mhz but the benchmark is stuck around 2400mhz region, gpu voltage reported at 0.884mv.

scored 7097

 

[JediFragger]

4K really stresses the clocks on the XT/XTX anyhow, mine is lower than what I set. You're vastly limited by power if the chip isn't v good tbh.

Currently testing a 2700c/2720mem/1090mV/-10 PL for 24/7 setting and get roughly 7200 with my Nitro, happy with that. Along with Chill it really optimises power usage for gaming.

 

[wookiee87]

should i run tests at 1440p or lower?

did a vulcan run and got 7262, with oc settings from above

 

[JediFragger]

did a vulcan run and got 7262, with oc settings from above

Mine was with Vulkan so yours seems about right tbh for MBA, what is your undervolt set to?

 

[wookiee87]

Mine was with Vulkan so yours seems about right tbh for MBA, what is your undervolt set to?

not running a under volt, +15% on pwr target.

some freaky stuff happning, set 2700mhz min 2800mhz max core, ram at 2776mhz, -10% power limit, in bench gpu core clock hovering just over 2100mhz, just scrapped 6000 points.

 

[JediFragger]

not running a under volt, +15% on pwr target.

Try a small drop so it won't affect stability, say 1130mV. It helps to sustain higher clocks.

 

[wookiee87]

Try a small drop so it won't affect stability, say 1130mV. It helps to sustain higher clocks.

kinda working set 1105mv for core voltage, power limit is 0% clocks set at 2700,2800mhz, score jumped to 6830 but again core clocks during the test hover around 2250-2300mhz, i can get them to around 2600mhz but have to set 3200mhz min to get 2600mhz its very weird

 

[L0rdMike]

I have heard to leave the min value alone and just use the max to overclock, dunno if it makes a difference.

 

[Jay-G25]

Good to see 3DMark taking down the bugged runs but i fear they will have their work cut out .... they need to implement something where it notices a big fps spike and then makes it invalid and will save them then having to manually go through each score .... but at least they are working on it

From a 3DMark employee over on OCN : https://www.overclock.net/posts/29334312/

 

[nlel1975]

Good to see 3DMark taking down the bugged runs but i fear they will have their work cut out .... they need to implement something where it notices a big fps spike and then makes it invalid and will save them then having to manually go through each score .... but at least they are working on it

They need to do it for every graphics card not just the 4090. They’ve got their work cut out if doing manually.

 

[Jay-G25]

They need to do it for every graphics card not just the 4090. They’ve got their work cut out if doing manually.

Do AMD also have this issue ? i thought it was exclusively a 40 series problem as never had anything like this back on the 30 series cards.... i 1st encountered it just after getting the 4090 and running Speedway , the difference being the 1st words out of my mouth when posting the score were this is a bugged run and from then on was like wild fire Hopefully the fix they are implementing will stop all this, they need to apply it to Speedway as well as that is another with bugged scores.

 

[nlel1975]

Do AMD also have this issue ?

Yes it happens with every card if pushed too far.

 

[MadMatty]

The issue is limited to the 40 series or maybe just the 4090 due to the way its error correction works (or doesnt). Every other card, Amd/30 series will slow down when error correction kicks in so you get lower scores and no visual errors. Just a black screen and system reset when you've gone to far.

 

[humbug]

7900xt

Mem running at 2900mhz

I scored 6 334 in Steel Nomad

AMD Ryzen 7 7800X3D, AMD Radeon RX 7900 XT x 1, 32768 MB, 64-bit Windows 11}

www.3dmark.com

16.6% memory overclock, nice....

 

[TheRealDeal]

So literally the only reliable results are from AMD and previous Nvidia card gens. Obviously stock 4 series and mildly overclocked 4 series are valid as well but those pushing the memory are getting massively overinflated scores. From reading that thread this is also a problem in Timespy.

 

[humbug]

So literally the only reliable results are from AMD and previous Nvidia card gens. Obviously stock 4 series and mildly overclocked 4 series are valid as well but those pushing the memory are getting massively overinflated scores. From reading that thread this is also a problem in Timespy.

4090 scoring near 2X a 4070 Ti when in reality that GPU is only about 40% faster, what it does have is 2X the memory bandwidth.

A 7900 XTX scores about 50% higher than a 7800 XT, in reality that GPU is 45% faster, which fits.... But the 7900 XTX also has 50% more memory bandwidth which is not as extreme as the 100% 4070 Ti vs 4090.

It is weird.

A couple of things, i would like to see a 4070 to see how that compared to the 4070 Ti with them both having the same 504 GB/s memory bandwidth. (puke)

Also the RX 7700 XT, it has 54 CU's to my 60 CU's, 11%, not a lot of difference, but with a 192 Bit Bus and 18 Gb/s IC's it has 432 GB/s memory bandwidth to my 624 GB/s with 256 Bit and 19.5 Gb/s IC's.