LTT 3DMark Thread

[arnavvr]

53 minutes ago, DXMember said:

I honestly don't see a difference between the two for the end game - as in to display the results

MS is more convenient for me as I can use desktop Excel and publish it directly, it also offers more stuff to do and I'm more familiar with it

On desktop it's fine. But mobile it's kind of lackluster. 

[done12many2]

6 hours ago, Kryptyx said:

Well I don't really think this is "my" thread as it is more for the community as a whole. I just happened to be the one who started it. I guess I was just trying to see what everyone else thought. 

I appreciate you taking everyone's opinion into consideration.

 

6 hours ago, Kryptyx said:

You could have a higher overall 3dmark score but much less FPS in gaming applications.

Following this train of thought, couldn't you have an extremely high graphics score influenced by multi GPUs, which scale well in FireStrike and Time Spy, but experience much lower FPS in actual games?  This of course due to the imperfect scaling or lack of support in a lot of games.

 

I ended up copying all of the entries for Firestrike 1.1 over to my own spreadsheet in order to look at things better with the use of some formulas.

 

Higher GPU count is the largest influencer, more so than physics scores when we look at all 390 submissions.  In the end, the guys with the most cards win when comparing similar GPUs so the guy with the most money to throw at either CPU or GPUs is still going to come out on top.  

 

Like @Vellinious said, "money always wins", but it's not just because of overall scores or physics.  A guy with a 5820k or 5930k and three 980 Ti's is going to do well in both graphics and overall scores.  His overall score will more than likely be higher than a guy with a 6950x and two 1080s for a considerable amount less.  His graphics score will definitely be higher, but does that mean his rig games better?

 

I agree that this is a graphics thread, but how much of a reflection are FireStrike and Time Spy graphics scores on gaming performance especially with regards to multi GPUs?    With that said, I don't believe that switching to graphics scores as the means of ranking accurately reflects gaming performance at all.  I'm not blind to the fact that physics scores are also influencing overall scores, but since it is a synthetic benchmark and not a measurement of actual gaming performance, I think we should just leave it as a benchmark and use it the way the designers intended.

 

I'm personally not affected if this change happens.  I stay in the same spot in Time Spy and move one down in FireStrike 1.1.  

 

Others stand to gain quite a bit of upward movement, especially in the 4 core, multi Maxwell GPU group.  Naturally so as it's the largest group.

 

Maybe we should consider breaking off into single and multi GPU categories/tabs.  I think that serves the community more than switching from overall score to graphics score.

 

Hopefully the guys over in the Cinebench thread aren't talking about switching to single core performance.  After all, single core is a subcategory and could be argued as more pertinent to daily computing.  

 

Either way, thanks for lending an ear.

[Jumper118]

5 hours ago, done12many2 said:

e 

 

Hopefully the guys over in the Cinebench thread aren't talking about switching to single core performance.  After all, single core is a subcategory and could be argued as more pertinent to daily computing.  

 

Either way, thanks for lending an ear.

No chance.  I likes the Xeons too much if someone wants to spend £100k to get 8 22 Xeons to be at the top of the board then they are welcome to do that

[Vellinious]

6 hours ago, done12many2 said:

I appreciate you taking everyone's opinion into consideration.

 

Following this train of thought, couldn't you have an extremely high graphics score influenced by multi GPUs, which scale well in FireStrike and Time Spy, but experience much lower FPS in actual games?  This of course due to the imperfect scaling or lack of support in a lot of games.

 

I ended up copying all of the entries for Firestrike 1.1 over to my own spreadsheet in order to look at things better with the use of some formulas.

 

Higher GPU count is the largest influencer, more so than physics scores when we look at all 390 submissions.  In the end, the guys with the most cards win when comparing similar GPUs so the guy with the most money to throw at either CPU or GPUs is still going to come out on top.  

 

Like @Vellinious said, "money always wins", but it's not just because of overall scores or physics.  A guy with a 5820k or 5930k and three 980 Ti's is going to do well in both graphics and overall scores.  His overall score will more than likely be higher than a guy with a 6950x and two 1080s for a considerable amount less.  His graphics score will definitely be higher, but does that mean his rig games better?

 

I agree that this is a graphics thread, but how much of a reflection are FireStrike and Time Spy graphics scores on gaming performance especially with regards to multi GPUs?    With that said, I don't believe that switching to graphics scores as the means of ranking accurately reflects gaming performance at all.  I'm not blind to the fact that physics scores are also influencing overall scores, but since it is a synthetic benchmark and not a measurement of actual gaming performance, I think we should just leave it as a benchmark and use it the way the designers intended.

 

I'm personally not affected if this change happens.  I stay in the same spot in Time Spy and move one down in FireStrike 1.1.  

 

Others stand to gain quite a bit of upward movement, especially in the 4 core, multi Maxwell GPU group.  Naturally so as it's the largest group.

 

Maybe we should consider breaking off into single and multi GPU categories/tabs.  I think that serves the community more than switching from overall score to graphics score.

 

Hopefully the guys over in the Cinebench thread aren't talking about switching to single core performance.  After all, single core is a subcategory and could be argued as more pertinent to daily computing.  

 

Either way, thanks for lending an ear.

For multi-GPUs it's not a good indication of gaming performance at all.  The SLI in the 3D Mark benches scales at nearly 100%.  You won't find that in games.  As for overall performance in games, you're not going to find a lot of games that are threaded heavily enough to make the E series processors worth it for gaming (I can't even think of any right off hand), which is why the quad cores with the higher clocks are considered better gaming processors.  DX12 MIGHT change that, and bring the power of the E series processors more to the forefront, as they should be, but.....we're not there yet.

I agree that separate tabs for single vs multi GPU setups would be awesome.  You're talking about a lot of tabs though....almost easier at that point, to segregate the benchmarks to their own threads, maybe, to make tracking it easier?  I dunno...this system works fine.  Maybe just an added field for number of GPUs, so you can filter down on x1 / x2 / x3 or something.  /shrug  I know that's the first thing I look at, when I'm running for score, "who's the top dog in graphics score for 2x 'insert my card here'".

Cinebench is all about the CPU...I would think you'd want to measure all of the CPU performance metrics.  I don't even know what they measure on....I've never been overly concerned with that benchmark.  Now, if they started talking about using the GPU test to sort on, now that'd be pretty silly....

 

[done12many2]

2 hours ago, Vellinious said:

For multi-GPUs it's not a good indication of gaming performance at all.  The SLI in the 3D Mark benches scales at nearly 100%.  You won't find that in games.  As for overall performance in games, you're not going to find a lot of games that are threaded heavily enough to make the E series processors worth it for gaming (I can't even think of any right off hand), which is why the quad cores with the higher clocks are considered better gaming processors.  DX12 MIGHT change that, and bring the power of the E series processors more to the forefront, as they should be, but.....we're not there yet.

 

We're on the same sheet of music here.

 

2 hours ago, Vellinious said:

I agree that separate tabs for single vs multi GPU setups would be awesome.  You're talking about a lot of tabs though....almost easier at that point, to segregate the benchmarks to their own threads, maybe, to make tracking it easier?  I dunno...this system works fine.  Maybe just an added field for number of GPUs, so you can filter down on x1 / x2 / x3 or something.  /shrug  I know that's the first thing I look at, when I'm running for score, "who's the top dog in graphics score for 2x 'insert my card here'".

 

I think the best idea is segregation.  One thread that focuses on graphics scores with another that maintains scoring per 3DMark's design.  I haven't seen this practice in the other forums that you and I both frequent, however they haven't pushed for alternate ranking systems.  I've got all the time in the world so maybe I can look into this further.

 

3 hours ago, Vellinious said:

Cinebench is all about the CPU...I would think you'd want to measure all of the CPU performance metrics.  I don't even know what they measure on....I've never been overly concerned with that benchmark.  Now, if they started talking about using the GPU test to sort on, now that'd be pretty silly....

I mostly mentioned that out of sarcasm, which I'm glad @Jumper118 found humor in.

 

Speaking of FireStrike, Time Spy or Cinebench, where are the 6050x submissions?  I'm looking forward to seeing that rig of yours sail to the top both here and in the HOF.  I know @DXMember is waiting to shout, "called it!"

[Vellinious]

CPU arrives on Wednesday.  I'll tear down the loop and get everything flushed and refilled over the weekend, then start my baseline benches.  Hopefully it won't be too much longer for my turn in the step up queue to come around, and I can start with the 1080s.

[arnavvr]

8 minutes ago, Vellinious said:

CPU arrives on Wednesday.  I'll tear down the loop and get everything flushed and refilled over the weekend, then start my baseline benches.  Hopefully it won't be too much longer for my turn in the step up queue to come around, and I can start with the 1080s.

Wow! I would never spend $1700 on a proc... The SR-2 costed enough already...

[Vellinious]

4 minutes ago, arnavvr said:

Wow! I would never spend $1700 on a proc... The SR-2 costed enough already...

I work with very large data sets and databases for logistical analysis of transportation networks....it just so happens, that I also like to benchmark, so it's going to serve multiple purposes.  Work, benchmark, and gaming.  

[arnavvr]

3 minutes ago, Vellinious said:

I work with very large data sets and databases for logistical analysis of transportation networks....it just so happens, that I also like to benchmark, so it's going to serve multiple purposes.  Work, benchmark, and gaming.  

Wouldn't be cheaper to just build two rigs? One Z9PED9WS and Dual 2670s, and one 6700K Gaming Rig?

 

4000th Post in Comic Sans

[Vellinious]

7 minutes ago, arnavvr said:

Wouldn't be cheaper to just build two rigs? One Z9PED9WS and Dual 2670s, and one 6700K Gaming Rig?

 

4000th Post in Comic Sans

Might be....hadn't really considered that, but....eh

[VictorRoco]

Benchmark: Sky Diver

CPU: i7-6700HQ

GPU: GTX 965M

GPU Core: 935MHz

GPU Memory: 1253MHz

Score: 11959

3DMark Link: (Not Valid So No Link But Score is real)

[IGJoe2192]

New personal best,

Benchmark: Timespy - default

CPU: i7-5930K @ 4.5Ghz

Memory: 16GB 4x4 G.Skill 3000mhz

Motherboard: Asus X99-A

Graphics Card(s): Reference Sapphire RX480 8GB (x2)

PSU: Corsair HX1000i

 

CPU Score: 7257

Graphics Score: 8594

3DMark Score: 8362

 

Gfx Test 1: 57.45fps

Gfx Test 2: 48.21fps

CPU Test: 24.38fps

Link: http://www.3dmark.com/spy/151102

 

[Vellinious]

1 hour ago, IGJoe2192 said:

New personal best,

Benchmark: Timespy - default

CPU: i7-5930K @ 4.5Ghz

Memory: 16GB 4x4 G.Skill 3000mhz

Motherboard: Asus X99-A

Graphics Card(s): Reference Sapphire RX480 8GB (x2)

PSU: Corsair HX1000i

 

CPU Score: 7257

Graphics Score: 8594

3DMark Score: 8362

 

Gfx Test 1: 57.45fps

Gfx Test 2: 48.21fps

CPU Test: 24.38fps

Link: http://www.3dmark.com/spy/151102

 

I'm having a bit of trouble understanding what makes this CPU test "tick".  By all rights, at 4.5, the 5930k should be posting a higher CPU score than a 5820 at 4.625....but it's not.  Likewise, the 4.75 that I had always used when running Firestrike, yielded a lower score, than the 4.625 I ended up using.  

Just an observation....has anyone done any extensive testing with the Timespy CPU score?  I'll do some more this weekend, it just seems that the CPU clock increases don't increase the CPU score as much as they did in previous versions.  

[done12many2]

4 minutes ago, Vellinious said:

I'm having a bit of trouble understanding what makes this CPU test "tick".  By all rights, at 4.5, the 5930k should be posting a higher CPU score than a 5820 at 4.625....but it's not.  Likewise, the 4.75 that I had always used when running Firestrike, yielded a lower score, than the 4.625 I ended up using.  

Just an observation....has anyone done any extensive testing with the Timespy CPU score?  I'll do some more this weekend, it just seems that the CPU clock increases don't increase the CPU score as much as they did in previous versions.  

 

Why would you expect a 5930k at a lower speed to score higher than a 5820k at a higher speed?

[done12many2]

Also, Time Spy scores seem to scale well on my end with higher clock speeds.

[IGJoe2192]

Timespy seems to be more core count bound than speed bound, though the speed does help but not as much. 5820 and 5930 have the same core count.

[arnavvr]

12 minutes ago, IGJoe2192 said:

Timespy seems to be more core count bound than speed bound, though the speed does help but not as much. 5820 and 5930 have the same core count.

 

What about 22 Core Xeons ?

[IGJoe2192]

Just now, arnavvr said:

What about 22 Core Xeons ?

I haven't seen any scores but I bet they would eat it up. 

[arnavvr]

Just now, IGJoe2192 said:

I haven't seen any scores but I bet they would eat it up. 

They are cheaper than 6950X 

[IGJoe2192]

45 minutes ago, Vellinious said:

 

For a bit of comparison, here is a 5820K overclocked to 4.5ghz with 2133mhz memory vs my 5930k @ 4.5ghz with 3000mhz memory. Also check out those OC'd RX480s vs a single 1080 OC =D I hope most games scale like this in the future. 

 

http://www.3dmark.com/compare/spy/117391/spy/151102#

[arnavvr]

Just now, IGJoe2192 said:

For a bit of comparison, here is a 5820K overclocked to 4.5ghz with 2133mhz memory vs my 5930k @ 4.5ghz with 3000mhz memory. Also check out those OC'd RX480s vs a single 1080 OC =D I hope most games scale like this in the future. 

 

lol i wish 1060s did SLI that would have been the perfect bang/buck

[IGJoe2192]

Just now, arnavvr said:

lol i wish 1060s did SLI that would have been the perfect bang/buck

They will in games that support Explicit Multi-Adapter

[grayperview]

Not very good scaling for me on my unlocked fury.

 

http://www.3dmark.com/compare/spy/46736/spy/133920

[Vellinious]

1 hour ago, done12many2 said:

 

Why would you expect a 5930k at a lower speed to score higher than a 5820k at a higher speed?

Because they would in the physics test in Firestrike.  This test changed things.

[done12many2]

Just now, Vellinious said:

Because they would in the physics test in Firestrike.  This test changed things.

Am I missing something?  The only difference between the two are PCIe lane count.  Other than that they are the same when run at the same clock speed