[nando4]

Sandy Bridge Preview: i5-430M 2.26 HD5750@x2 vs GTX470@x2

The Sandy Bridge platform, scheduled for release in Q1 2011, will bring with it pci-e 2.0 ports providing 4GBps bandwidth after overhead. That is equivalent to x2 1.0.

A HP Pavilion DV4-2000 with 2xmPCIe slots designated as port1+port2 was used do x2 link testing of a HD5750 versus a GTX470 below. The system also has Intel HD + HD4550 switchable graphics, so Optimus was tested as well. Unfortunately the Optimus performance tweak only engaged on a x1 link, so x2.Opt is not possible. x2.Opt would be the equivalent of x1 2.0 that Sandy Bridge will offer.

System	RAM	GPU	3dmark		RE5		dmcv4.scene4		FFXIV		MafiaII	Ports	OS
			06	vant.g	dx9	dx10	dx9	dx10	high	low	1080p
i5-430M 2.26 HP Pav DV4-2000	4.0	GTX470@x2.Opt	9226	11183	102.2	97.5	104.6	116.4	2424	2607	49.3	HM55	Win7/64
		GTX470@x1.Opt	13205	12859 &	100.8	97.4	97.2	114.3	2410	2438	48.1	HM55
		GTX470@x1	5351	7715	70.2	69.9	61.0	75.3	2101	2252	29.0
		HD5750@x2	11540	6186	90.0	62.7	86.8	57.3	1917	2742	34.6	HM55
		HD5750@x1	10930	5194	73.0	43.5	66.7	40.9	1804	2685	28.8	HM55
T6600-2.2 Inspiron_1440	4.0	GTX470@x1.Opt	10264	10972	58.4	61.8	90.0	96.5	2242	2252	32.1	ICH9M	Win7/32

& = 820/1900 overclock giving ~GTX480 performance
x2.Opt = x2 link, Intel HD graphics, Optimus driver, but Optimus performance tweak does not engage. Only engages for a x1 link

Observations

* the Optimus performance tweak only engages upon detecting a x1 link. A x2 link reverts to non-Optimus performance levels, probably because NVidia assumes if it's not x1, then it's x8 (SLI) or x16. Means I cannot do a full preview of x1 2.0 Optimus preformance levels until it's tested on a Sandy Bridge platform that can do x1 pci-e 2.0 link speeds. Let's hope NVidia still release the x1 tweak with the driver set for that platform.

* I tried to trick the driver after being initialised in x1.Opt mode, changing the link to be a x2 link. System froze. If enable/disable the GTX470, then the Optimus driver will once again re-check the link width and only enable the Optimus tweak if it's x1.

* synthetic benchmarks of x1 1.0 Optimus comparison systems sees them outperform x2 1.0 non-Optimus. Real life gaming results do however see better results with x2 1.0. 3dmark06's scene1: Return to Proxycon pans smoother with a x1 1.0 Optimus setup than a x2 1.0 non-Optimus setup. Other benchmarks place x1.Opt to be within a whisker of x2 non-Optimus performance levels.

* I suspect the x1.Opt tweak is setting up a end-to-end compressed link across the bottleneck x1 link given it sees increased performance with faster CPUs (see first page for comparative performance).

* x1E/x2E tweak gives no performance increase over x1/x2 on a series 5 chipsets. So I got crippled x1/x2 performance with the HD5750 which was 15-30% slower than it would be with the tweak on an older system. We see JameBond007's P8400@2.44+HD5770@x2E here significantly outperform my i5-430M 2.26+(crippled)HD5750@x2 results above.

Conclusion

NVidia. Why did you have to lock down the Optimus performance tweak to x1 only?? The x2.Opt setup would fly otherwise. More comments about this in the NVidia Optimus works for Fermi desktop GPUS - more performance + Internal LCD setup!! post.