By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Turkish on 31 January 2014
Edit Thread
globalisateur said:
Turkish said:
globalisateur said:
I am sorry but I have to disagree with Famousmortimer here.

Esram alone is in fact in reality slower than GDDR5.

Real benchmarks during ideal conditions have been done on both machines:

- 140-150GB/s averaged at 145GB/s in best conditions using the best test available for Esram. Those are Microsoft ideal/best condition numbers and work for rare utilizations, most cases will reach maybe 100GB/s in fact.
- 160GB/s done with a similar test on the GDDR5 ram by some developers, they can reach 91% of the total 176GB/s which is compatible with similar, on best conditions tests on GDDR5 ram on PC, maybe like 135GB/s in less ideal conditions.

Hmm this is interesting, where were this tests done?

The 140-150 GB/s esram max real case bandwidth (in a real life test) comes from Microsoft found in a Eurogamer article. No developer will probably have a better real bandwidth than those 145GB/S Microsoft numbers I guess.

The 160GB/s comes from the 90%-91% bandwidth utilization that I read several times in hardware forums (years ago), 91% being the maxium real bandwidth utilization of the typical GDDR5 ram in a ideal simple test (without contention form the CPU from example). But we can directly compare those 145 and 160 test which are maximum real bandwidth in ideal (rare) scenarii.

But except those simple cases, I think the average bandwith (with latency and contention in any test) with GDDR5 ram is like 75% so 132GB/s. But with Esram it's even worse because those 145GB/S cases are if you use both direction at the same time of the bus. It can only be done in some cases like framebuffer simple alpha operations. The rest (which depends of the game) will be limited by the peak 109GB/S so 100GB/S realistically in the best case, I feel generous here (because esram low latency powa!!!).

I am using my fuzzy logic in many of my latter assumptions here. But I think, based from the info I gathered I must not be far away from the truth.

But of course for X1 you have to think about the 68GB/s main ram (peak bandwidth number maybe 50GB/S-ish in average cases) but when you transfer memory from DDR3 to esram (because duh 32MB is nothing in a next gen game, KZSF uses 800MB for just the objects rendered in one level) you are doing nothing but just using the DDR3 bandwidth (with probably CPU work but in most cases it should not need more than 5-10GB/S in fact) so that's why you can't directly add 50 + 100 for XB1.

You may better understand why if the PS4 GPU has only more 41% more grunt that X1, the PS4 games shows in most cases an average of 70% advantage over X1. (70% with BF4 and with Tomb Raider DE, 125% with COD and 44% with AC4). The 70% number on average and particularlry with BF4 which is the only game  where we could really compare both performance objectively in both unlocked (60fps) game which had stable gap performance visible (45fps for X1, 55fps for PS4 for long periods + resolution difference = 70% PS4 advantage).


hmmm thanks for the explanation, you and drkohler. So the PS4 ram advantage is bigger than I thought.