Intriguing Saint's Row 4 CPU benchmarks

[cytg111]

0% HT usage on the dual core, but some HT usage on the hexa core?
...

The 3970 and 2600 is both sandys and clocked about the same so total added total compute should be comparable ..

3970 : 56+62+6+17+11+17 = 169
2600 : 73+55+20+52 = 200

Even adjusting for a 100mhz lead, the 2600 is computing more and performing less .. sort of weird, it would be nice to know how much of those figures are kernel times (quad channel ddr3 minimizing some of that?). As non of the cores (besides the i3) is maxed, the game is obviously waiting for something else to complete.

 

[crashtech]

What amazes me the most is how well the 2100 does. I'd love to see a 3245 in those results. It's position in the ranks is a poke in the eye to those who say dual cores are dead. Looks like a viable low-end gaming choice to me.

 

[horza]

..

 

[Rvenger]

it just goes to show how far AMD has fallen behind... rather sad actually

They improved over Phenom and BD. This game is a console port so it's poorly programmed. I am sure some patches will help performance down the road.

 

[Essence_of_War]

They improved over Phenom and BD.

Yeah.

Except BD was actually a step-backwards. And not in the way that everyone jokes about "has-fail". It's not terribly praise-worthy to improve over a step-backwards, and improving over the 3ish year old PhenomII w/ Vishera was just expected.

 

[Erenhardt]

Yeah.

Except BD was actually a step-backwards.

How can be?

Clearly clock to clock 6 core phenom2 is slower than 6 core BD. BD might not live up to the hype, but it was not a step back.

 

[mrmt]

Well, as the saying goes, "it is what it is." One can blame the programmers for poor coding, but they could just as easily blame amd for designing processors with low ipc. It doesn't really matter, blaming either or both doesn't change the results.

Bulldozer/Piledriver chips are very sensitive to the quality of the code running. They shine once code is tuned for them by issuing less instructions to more cores but they suck big time once you need high IPC and don't use many cores. Programers need to heavily tune the code for Bulldozer and pray that the task at hand can make use of many threads, a true glass jaw of the architecture.

It's quite the opposite of what Intel is doing. HT gives MT performance bonuses but without compromises on single threaded performance, AVX2 and TSX might give you performance bonuses just by recompilling. That's more performance with less efforts demanded from the developers.

In the end we have one company asking programers to heavily tune code for their chips and giving mediocre software support for them, and we have another company trying to make the programers' lives easier. I wonder which one they will pick.

 

[Enigmoid]

How can be?

Clearly clock to clock 6 core phenom2 is slower than 6 core BD. BD might not live up to the hype, but it was not a step back.

Piledriver is equal to or slightly behind Phenom 2 in a bunch of games in terms of IPC. Had phenom 2 been ported over to 32nm it would have performed similarly and saved AMD tons of R&D.

Normalized to clockspeed the situation is even worse in applications.

If you look at the difference between the x4 965 and the 4350 we have a whopping increase of 7% stock vs stock.

 

[aShufflinZombie]

Does this game even support crossfire? 7990 only beat the 7970 by four frames.

 

[MarkLuvsCS]

Does this game even support crossfire? 7990 only beat the 7970 by four frames.

It's a new release. Doesn't it take amd a while to release the profiles for new games, or can you try to force crossfire onto applications like nvidia's sli?

 

[Blastman]

I'm thinking the Phenom cores falling back may have to do with SSE support -- specifically SSE4.1/4.2 support. The newer Bulldozer and later cores from AMD have full SSE4.1/4.2 support while Phenom does not.

There was an issue with Planetside 2 performance on older CPU's that was thought to be about lack of SSE4.1 support.

 

[Essence_of_War]

How can be?

Clearly clock to clock 6 core phenom2 is slower than 6 core BD. BD might not live up to the hype, but it was not a step back.

 

[AtenRa]

Piledriver is equal to or slightly behind Phenom 2 in a bunch of games in terms of IPC. Had phenom 2 been ported over to 32nm it would have performed similarly and saved AMD tons of R&D.

Llano is a Phenom 2 at 32nm and NO it is not faster in games than BD/PD.

Normalized to clockspeed the situation is even worse in applications.

If you look at the difference between the x4 965 and the 4350 we have a whopping increase of 7% stock vs stock.

Different Architectures, you cannot compare the two in the same clock.

BD is faster in 90% of today's games than Phenom x6 and PD increased on that another 10-15%.

 

[SPBHM]

Llano is a Phenom 2 at 32nm and NO it is not faster in games than BD/PD.

llano lacks l3 cache, you should only compare llano with trinity/richland, and llano does beat trinity at similar clocks for many things... but... llano can't go higher than 3GHz without bad power usage, and simply can't go over 3.6GHz for some reason... also it lacks avx, sse4.1 and more

still, with more time and money spent on improving "k10" it's not impossible to imagine that it would have been a better product than BD...