inf64
Diamond Member
- Mar 11, 2011
- 3,884
- 4,692
- 136
The OP is listing rather old information.Then he is using this old information in order to compare this unreleased product to yet another unreleased product(Zambezi). Ok I can play the speculation game too
.
For top end SB-E we will have 6C/12T with same underlying core uarchitecture as SB(2600K),only with somewhat more L3 cache per core than Westmere 6C(980x,990x,unreleased 995x),15MB vs 12MB. Turbo has a slight bump to 3.9GHz versus 3.8Ghz for SB and 3.8 or 3.9Ghz for 995x.Since Turbo will run more often then not,I assume SB-E will almost always run at 3.6Ghz for 4 core workloads and 3.9Ghz for single core workloads.This is in reality around 2% better than what 2600K can do with its Turbo,both on 4C and single core load.
So summed up : L3 cache bump may bring ~3% or so higher performance and overall Turbo boost may be responsible for another 2%.Cumulative effect for SB-E versus 2600K on desktop,using hardware.fr chart : 213.4x1.03x1.02x1.16*=261 "points". But SB-E will have lower starting clock so we need to figure in results for 3.3Ghz 6 core workloads.This is 3.4/3.3=1.03 or 3% lower. So SB-E @ 3.3Ghz/3.9GhzT should land somewhere between 253pts and 261pts.
995x would in turn get 228.5x3.6/3.46=238 points.
*In intel's case,going from 4C/8T Nehalem @ 3.33Ghz to 6C/12T @ 3.33Ghz Westmere with 50% more L3 in desktop apps brings exactly 16.6% perf. improvement,courtesy of SMT which diminishes the effect of more real cores.This is what hardware.fr found out in their 980x review.AMD sees a bit better effect,24.5% higher overall score with 50% more cores (Deneb @ 3.2Ghz Vs Thuban @ 3.2Ghz,Turbo brings just over 1% so it's irrelevant).
So going from Westmere 6C @ 3.6Ghz/3.8GhzT to SB-E 6C @ 3.3Ghz/3.9GhzT we have approximately ~9.6% improvement in best case(261/238=1.096) or 6.3% in worst case (253/238).
On to Zambezi.I have written a blog that covers this.To cut the story short,my estimate for a rumored 8C 3.6Ghz/4.2GhzT Zambezi is 229pts,or 39% better overall performance than Thuban @ 3.3Ghz. This would tie the score of 990x and trail yet to be launched 995x(Q3 or Q4) by around 3%,while costing 3x less than 900+$ 995x. SB-E looks to be delayed a bit,to Q1 2012.If it launches with the specs we know now,it will be around 10-14%(worst-best case) faster overall than 3.6Ghz 8C Zambezi,while costing roughly 3x more(probably replacing the price spot of 995x).
For top end SB-E we will have 6C/12T with same underlying core uarchitecture as SB(2600K),only with somewhat more L3 cache per core than Westmere 6C(980x,990x,unreleased 995x),15MB vs 12MB. Turbo has a slight bump to 3.9GHz versus 3.8Ghz for SB and 3.8 or 3.9Ghz for 995x.Since Turbo will run more often then not,I assume SB-E will almost always run at 3.6Ghz for 4 core workloads and 3.9Ghz for single core workloads.This is in reality around 2% better than what 2600K can do with its Turbo,both on 4C and single core load.
So summed up : L3 cache bump may bring ~3% or so higher performance and overall Turbo boost may be responsible for another 2%.Cumulative effect for SB-E versus 2600K on desktop,using hardware.fr chart : 213.4x1.03x1.02x1.16*=261 "points". But SB-E will have lower starting clock so we need to figure in results for 3.3Ghz 6 core workloads.This is 3.4/3.3=1.03 or 3% lower. So SB-E @ 3.3Ghz/3.9GhzT should land somewhere between 253pts and 261pts.
995x would in turn get 228.5x3.6/3.46=238 points.
*In intel's case,going from 4C/8T Nehalem @ 3.33Ghz to 6C/12T @ 3.33Ghz Westmere with 50% more L3 in desktop apps brings exactly 16.6% perf. improvement,courtesy of SMT which diminishes the effect of more real cores.This is what hardware.fr found out in their 980x review.AMD sees a bit better effect,24.5% higher overall score with 50% more cores (Deneb @ 3.2Ghz Vs Thuban @ 3.2Ghz,Turbo brings just over 1% so it's irrelevant).
So going from Westmere 6C @ 3.6Ghz/3.8GhzT to SB-E 6C @ 3.3Ghz/3.9GhzT we have approximately ~9.6% improvement in best case(261/238=1.096) or 6.3% in worst case (253/238).
On to Zambezi.I have written a blog that covers this.To cut the story short,my estimate for a rumored 8C 3.6Ghz/4.2GhzT Zambezi is 229pts,or 39% better overall performance than Thuban @ 3.3Ghz. This would tie the score of 990x and trail yet to be launched 995x(Q3 or Q4) by around 3%,while costing 3x less than 900+$ 995x. SB-E looks to be delayed a bit,to Q1 2012.If it launches with the specs we know now,it will be around 10-14%(worst-best case) faster overall than 3.6Ghz 8C Zambezi,while costing roughly 3x more(probably replacing the price spot of 995x).
Last edited:
