Speaking of duct taping, what clocks are needed to get Drake to land in the ballpark of three PS4s?
CPU's easy enough; I'm feeling 1.6 Ghz (matching the core count and clock) could do it. For some workloads, A78 should hit triple the Jaguar's IPC, and that's the 3x right there. And of course, clocks can be pushed a bit higher if we want to cover more workloads.
I'm blanking on what's needed for the GPU though.
And why I am asking about three Playstation Four's? Change the Four to something like, I dunno, Fource or something... and when taping them together, they could be arranged in like, a triangle, I guess?
Just three? Or over three? Bear with me, this post is going to be all over the place. At the end youโll imagine that Iโm Charlie Day and Iโm insane
Anywayโฆ
The Jaguar in the PS4, XB1, Pro and 1X use the same old Jaguar architecture.
Looking at the GB5 numbers for A5150, A5370, A5350 and Sempron 3850
@ 1GHz, I get: 119.375, 120.9, 120.4, 123.3. Respectively. Iโll average it out and round up.
I get 121 when normalized to 1GHz
In comparison, I found a few results of the TX1 (Android shield TV, not the switch) on GB5: 138.5, 140, 139.5, 139.6. Iโll round down for this.
So, 139 when normalized to 1GHz
(only for comparison sake)
Anyway, the PS4 only uses 6.5 cores for games on average and itโs clocked to 1.6GHz.
When figuring out the Multicore score, I noticed it didnโt translate linearly. So, I took the difference and averaged that multi core score is only 83.5% of the (single core score) x (number of cores) scoreโฆ. with that saidโฆ.
A
theoretical score:
(121*6.5*1.6)*0.835= 1,051 as a theoretical multi core score on geekbench 5 for the PS4 of what is actually usable. Or how it actually runs with those cores. Rounded up
Similarly happens to ARM, itโs 85% for that.
Soโฆ. (139*3*1.02)*0.85= 362 as a theoretical multi core score on GB5 for the Nintendo switch of what is actually usable. Rounded up
Nowโฆ. Onto your question of 3x the PS4โฆ well, we only have 1 sample which is the Tegra ORIN who has a 754 single core and a multi of 7773โฆ. So itโs not stellar sample size but Iโll try to make a theoretical. There was that other video so Iโll use those numbers and try to average the twoโฆ.
Normalized to the single GHz, it comes down to around 345 averaged of the 2 and rounded up.
With ORIN it is ~82% for multi like above.
So since the Drake will have 7 for games, it is a bonus going for it.
(345*7*1.6)*0.82= 3,169 theoretical score when applied to how it would actually run if at those clocks. Rounded up.
So, thatโs 3.015x the PS4 in theoretical world modeโฆ.
What about the other mentioned consoles?
Glad you asked.
XBox One runs at 1.75GHz and it is also doing 6.5 Cores.. ish. PS4 Pro has 6.5 and at 2.13GHz. 1X is at 2.3GHzโฆ
So, 1S: (121*6.5*1.75)*0.85= 1,150MT/ 212ST
Theoretical Drake(1.75) vs 1S: 3,466MT/ 604ST
Pro: 1,399MT/ 258ST
Drake(2.13): 4,219MT/ 735ST
One X: 1,511MT/ 279ST
Drake(2.3): 4,639MT/ 794ST
All of this at less than 4.5W no less (on 5N). Give or take.
PHEW! I finished the CPU theoretical Magnum opus, however, I should stress several caveats with this, first off geekbench is
very memory bandwidth sensitive. The more memory bandwidth you have, the far better your score is. Second of all, geekbench is a synthetic benchmark and cannot present true real world tests to the table, even bringing things not related to gaming at all.
Third of all, this is making several assumptions to get these numbers, although they are more closer to how other people would do it to find an actual answer, we will never truly know what it is unless it is placed in that position in a real world scenario that can be observed. That is to say this is just for fun and not meant to be taken as gospel.
For the GPU thoughโฆ. Drake would need to be clocked to 1.8GHz to match 3x the PS4 GPU on paper.
Edit: I also want to show something to people=
Oh and for Zen 2 in the PS5 and series, it should be this
261 single core normalized to 1GHz.
So, 914ST for the PS5
and this should be about where the MT score is with an optimized game that gets a 35% CPU uplift:
6,418MT score.
This is the closest to your real world comparison youโll get.
I did this one specifically to show people, everyone that keeps on saying that the next generation consoles will have โsignificantly stronger CPUโs in every scenario to switch Drakeโ need to reel in their expectations with respect to those consoles. Geekbench 5 shows software that is very optimized for taking that โup to 50% perf increase from SMTโ and completely for multithreaded work loads,
Well spoiler: that is completely unrealistic for games. I showed everyone here what it would be like, depending on the clock frequencies, and I kept hearing about โsignificantly strongerโ which I had my doubts because of the caveat of the platform testing.
I hope this brings people back into a more realistic world when it comes to this. I am aware of the caveat that I mentioned about geekbench, however, if these mobile CPUs can get that close to this desktop configuration, I think people need to keep their expectations in check with respect to the CPU in the other platforms because itโs not as high as people make it seem.
Hell, a goddamn 12400 competes and does very fine against the desktop class CPUs in these consoles.
And I was generous with that 35%.
Itโs more like 25% for games. They arenโt really
that great scalers to large threads like that.