diff --git a/vp8/encoder/arm/quantize_arm.c b/vp8/encoder/arm/quantize_arm.c
index 65c616614..225feaca6 100644
--- a/vp8/encoder/arm/quantize_arm.c
+++ b/vp8/encoder/arm/quantize_arm.c
@@ -29,7 +29,7 @@ extern int vp8_fast_quantize_b_neon_func(short *coeff_ptr, short *zbin_ptr, shor
 
 void vp8_fast_quantize_b_neon(BLOCK *b, BLOCKD *d)
 {
-    d->eob = vp8_fast_quantize_b_neon_func(b->coeff, b->zbin, d->qcoeff, d->dqcoeff, d->dequant, vp8_rvsplus1_default_zig_zag1d, b->round, b->quant);
+    d->eob = vp8_fast_quantize_b_neon_func(b->coeff, b->zbin, d->qcoeff, d->dqcoeff, d->dequant, vp8_rvsplus1_default_zig_zag1d, b->round, b->quant_fast);
 }
 
 /*
diff --git a/vp8/encoder/block.h b/vp8/encoder/block.h
index 90b42c35c..bf94e508b 100644
--- a/vp8/encoder/block.h
+++ b/vp8/encoder/block.h
@@ -33,6 +33,7 @@ typedef struct
 
     // 16 Y blocks, 4 U blocks, 4 V blocks each with 16 entries
     short *quant;
+    short *quant_fast;
     short *quant_shift;
     short *zbin;
     short *zrun_zbin_boost;
diff --git a/vp8/encoder/encodeframe.c b/vp8/encoder/encodeframe.c
index 152838946..cb7cc65d7 100644
--- a/vp8/encoder/encodeframe.c
+++ b/vp8/encoder/encodeframe.c
@@ -179,6 +179,7 @@ void vp8cx_init_quantizer(VP8_COMP *cpi)
     {
         // dc values
         quant_val = vp8_dc_quant(Q, cpi->common.y1dc_delta_q);
+        cpi->Y1quant_fast[Q][0] = (1 << 16) / quant_val;
         vp8cx_invert_quant(cpi->sf.improved_quant, cpi->Y1quant[Q] + 0,
                            cpi->Y1quant_shift[Q] + 0, quant_val);
         cpi->Y1zbin[Q][0] = ((qzbin_factors[Q] * quant_val) + 64) >> 7;
@@ -187,6 +188,7 @@ void vp8cx_init_quantizer(VP8_COMP *cpi)
         cpi->zrun_zbin_boost_y1[Q][0] = (quant_val * zbin_boost[0]) >> 7;
 
         quant_val = vp8_dc2quant(Q, cpi->common.y2dc_delta_q);
+        cpi->Y2quant_fast[Q][0] = (1 << 16) / quant_val;
         vp8cx_invert_quant(cpi->sf.improved_quant, cpi->Y2quant[Q] + 0,
                            cpi->Y2quant_shift[Q] + 0, quant_val);
         cpi->Y2zbin[Q][0] = ((qzbin_factors_y2[Q] * quant_val) + 64) >> 7;
@@ -195,6 +197,7 @@ void vp8cx_init_quantizer(VP8_COMP *cpi)
         cpi->zrun_zbin_boost_y2[Q][0] = (quant_val * zbin_boost[0]) >> 7;
 
         quant_val = vp8_dc_uv_quant(Q, cpi->common.uvdc_delta_q);
+        cpi->UVquant_fast[Q][0] = (1 << 16) / quant_val;
         vp8cx_invert_quant(cpi->sf.improved_quant, cpi->UVquant[Q] + 0,
                            cpi->UVquant_shift[Q] + 0, quant_val);
         cpi->UVzbin[Q][0] = ((qzbin_factors[Q] * quant_val) + 64) >> 7;;
@@ -208,6 +211,7 @@ void vp8cx_init_quantizer(VP8_COMP *cpi)
             int rc = vp8_default_zig_zag1d[i];
 
             quant_val = vp8_ac_yquant(Q);
+            cpi->Y1quant_fast[Q][rc] = (1 << 16) / quant_val;
             vp8cx_invert_quant(cpi->sf.improved_quant, cpi->Y1quant[Q] + rc,
                                cpi->Y1quant_shift[Q] + rc, quant_val);
             cpi->Y1zbin[Q][rc] = ((qzbin_factors[Q] * quant_val) + 64) >> 7;
@@ -216,6 +220,7 @@ void vp8cx_init_quantizer(VP8_COMP *cpi)
             cpi->zrun_zbin_boost_y1[Q][i] = (quant_val * zbin_boost[i]) >> 7;
 
             quant_val = vp8_ac2quant(Q, cpi->common.y2ac_delta_q);
+            cpi->Y2quant_fast[Q][rc] = (1 << 16) / quant_val;
             vp8cx_invert_quant(cpi->sf.improved_quant, cpi->Y2quant[Q] + rc,
                                cpi->Y2quant_shift[Q] + rc, quant_val);
             cpi->Y2zbin[Q][rc] = ((qzbin_factors_y2[Q] * quant_val) + 64) >> 7;
@@ -224,6 +229,7 @@ void vp8cx_init_quantizer(VP8_COMP *cpi)
             cpi->zrun_zbin_boost_y2[Q][i] = (quant_val * zbin_boost[i]) >> 7;
 
             quant_val = vp8_ac_uv_quant(Q, cpi->common.uvac_delta_q);
+            cpi->UVquant_fast[Q][rc] = (1 << 16) / quant_val;
             vp8cx_invert_quant(cpi->sf.improved_quant, cpi->UVquant[Q] + rc,
                                cpi->UVquant_shift[Q] + rc, quant_val);
             cpi->UVzbin[Q][rc] = ((qzbin_factors[Q] * quant_val) + 64) >> 7;
@@ -325,6 +331,7 @@ void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x)
     for (i = 0; i < 16; i++)
     {
         x->block[i].quant = cpi->Y1quant[QIndex];
+        x->block[i].quant_fast = cpi->Y1quant_fast[QIndex];
         x->block[i].quant_shift = cpi->Y1quant_shift[QIndex];
         x->block[i].zbin = cpi->Y1zbin[QIndex];
         x->block[i].round = cpi->Y1round[QIndex];
@@ -339,6 +346,7 @@ void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x)
     for (i = 16; i < 24; i++)
     {
         x->block[i].quant = cpi->UVquant[QIndex];
+        x->block[i].quant_fast = cpi->UVquant_fast[QIndex];
         x->block[i].quant_shift = cpi->UVquant_shift[QIndex];
         x->block[i].zbin = cpi->UVzbin[QIndex];
         x->block[i].round = cpi->UVround[QIndex];
@@ -349,6 +357,7 @@ void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x)
 
     // Y2
     zbin_extra = (cpi->common.Y2dequant[QIndex][1] * ((cpi->zbin_over_quant / 2) + cpi->zbin_mode_boost)) >> 7;
+    x->block[24].quant_fast = cpi->Y2quant_fast[QIndex];
     x->block[24].quant = cpi->Y2quant[QIndex];
     x->block[24].quant_shift = cpi->Y2quant_shift[QIndex];
     x->block[24].zbin = cpi->Y2zbin[QIndex];
@@ -1270,7 +1279,18 @@ int vp8cx_encode_inter_macroblock
 
     if (cpi->sf.RD)
     {
+        /* Are we using the fast quantizer for the mode selection? */
+        if(cpi->sf.use_fastquant_for_pick)
+            cpi->mb.quantize_b      = QUANTIZE_INVOKE(&cpi->rtcd.quantize, fastquantb);
+
         inter_error = vp8_rd_pick_inter_mode(cpi, x, recon_yoffset, recon_uvoffset, &rate, &distortion, &intra_error);
+
+        /* switch back to the regular quantizer for the encode */
+        if (cpi->sf.improved_quant)
+        {
+            cpi->mb.quantize_b    = QUANTIZE_INVOKE(&cpi->rtcd.quantize, quantb);
+        }
+
     }
     else
 #endif
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index 1f890790c..05a1338dc 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -591,6 +591,7 @@ void vp8_set_speed_features(VP8_COMP *cpi)
     sf->max_fs_radius = 32;
     sf->iterative_sub_pixel = 1;
     sf->optimize_coefficients = 1;
+    sf->use_fastquant_for_pick = 0;
 
     sf->first_step = 0;
     sf->max_step_search_steps = MAX_MVSEARCH_STEPS;
@@ -758,7 +759,7 @@ void vp8_set_speed_features(VP8_COMP *cpi)
 
             cpi->mode_check_freq[THR_SPLITG] = 4;
             cpi->mode_check_freq[THR_SPLITA] = 4;
-            cpi->mode_check_freq[THR_SPLITMV] = 0;
+            cpi->mode_check_freq[THR_SPLITMV] = 2;
 
             sf->thresh_mult[THR_TM       ] = 1500;
             sf->thresh_mult[THR_V_PRED   ] = 1500;
@@ -789,8 +790,7 @@ void vp8_set_speed_features(VP8_COMP *cpi)
                 sf->thresh_mult[THR_SPLITA   ] = 20000;
             }
 
-            sf->improved_quant = 0;
-            sf->improved_dct = 0;
+            sf->use_fastquant_for_pick = 1;
 
             sf->first_step = 1;
             sf->max_step_search_steps = MAX_MVSEARCH_STEPS;
@@ -798,6 +798,8 @@ void vp8_set_speed_features(VP8_COMP *cpi)
 
         if (Speed > 1)
         {
+            sf->use_fastquant_for_pick = 0;
+
             cpi->mode_check_freq[THR_SPLITG] = 15;
             cpi->mode_check_freq[THR_SPLITA] = 15;
             cpi->mode_check_freq[THR_SPLITMV] = 7;
@@ -831,6 +833,11 @@ void vp8_set_speed_features(VP8_COMP *cpi)
                 sf->thresh_mult[THR_SPLITA   ] = 50000;
             }
 
+            sf->first_step = 1;
+
+            sf->improved_quant = 0;
+            sf->improved_dct = 0;
+
             // Only do recode loop on key frames, golden frames and
             // alt ref frames
             sf->recode_loop = 2;
diff --git a/vp8/encoder/onyx_int.h b/vp8/encoder/onyx_int.h
index 990ae1d9e..ab270ca5f 100644
--- a/vp8/encoder/onyx_int.h
+++ b/vp8/encoder/onyx_int.h
@@ -182,6 +182,8 @@ typedef struct
     int first_step;
     int optimize_coefficients;
 
+    int use_fastquant_for_pick;
+
 } SPEED_FEATURES;
 
 typedef struct
@@ -269,6 +271,9 @@ typedef struct
     DECLARE_ALIGNED(16, short, zrun_zbin_boost_y1[QINDEX_RANGE][16]);
     DECLARE_ALIGNED(16, short, zrun_zbin_boost_y2[QINDEX_RANGE][16]);
     DECLARE_ALIGNED(16, short, zrun_zbin_boost_uv[QINDEX_RANGE][16]);
+    DECLARE_ALIGNED(16, short, Y1quant_fast[QINDEX_RANGE][16]);
+    DECLARE_ALIGNED(16, short, Y2quant_fast[QINDEX_RANGE][16]);
+    DECLARE_ALIGNED(16, short, UVquant_fast[QINDEX_RANGE][16]);
 
 
     MACROBLOCK mb;
diff --git a/vp8/encoder/quantize.c b/vp8/encoder/quantize.c
index a1be6614b..a67299487 100644
--- a/vp8/encoder/quantize.c
+++ b/vp8/encoder/quantize.c
@@ -27,7 +27,7 @@ void vp8_fast_quantize_b_c(BLOCK *b, BLOCKD *d)
     short *coeff_ptr       = b->coeff;
     short *zbin_ptr        = b->zbin;
     short *round_ptr       = b->round;
-    short *quant_ptr       = b->quant;
+    short *quant_ptr       = b->quant_fast;
     short *quant_shift_ptr = b->quant_shift;
     short *qcoeff_ptr      = d->qcoeff;
     short *dqcoeff_ptr     = d->dqcoeff;
@@ -74,7 +74,7 @@ void vp8_fast_quantize_b_c(BLOCK *b, BLOCKD *d)
     int x, y, z, sz;
     short *coeff_ptr   = b->coeff;
     short *round_ptr   = b->round;
-    short *quant_ptr   = b->quant;
+    short *quant_ptr   = b->quant_fast;
     short *qcoeff_ptr  = d->qcoeff;
     short *dqcoeff_ptr = d->dqcoeff;
     short *dequant_ptr = d->dequant;
diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c
index d2199a499..6e317e2a2 100644
--- a/vp8/encoder/x86/x86_csystemdependent.c
+++ b/vp8/encoder/x86/x86_csystemdependent.c
@@ -32,7 +32,7 @@ void vp8_fast_quantize_b_mmx(BLOCK *b, BLOCKD *d)
     short *coeff_ptr   = b->coeff;
     short *zbin_ptr    = b->zbin;
     short *round_ptr   = b->round;
-    short *quant_ptr   = b->quant;
+    short *quant_ptr   = b->quant_fast;
     short *qcoeff_ptr  = d->qcoeff;
     short *dqcoeff_ptr = d->dqcoeff;
     short *dequant_ptr = d->dequant;
@@ -90,7 +90,7 @@ void vp8_fast_quantize_b_sse2(BLOCK *b, BLOCKD *d)
     short *scan_mask   = vp8_default_zig_zag_mask;//d->scan_order_mask_ptr;
     short *coeff_ptr   = b->coeff;
     short *round_ptr   = b->round;
-    short *quant_ptr   = b->quant;
+    short *quant_ptr   = b->quant_fast;
     short *qcoeff_ptr  = d->qcoeff;
     short *dqcoeff_ptr = d->dqcoeff;
     short *dequant_ptr = d->dequant;
@@ -183,7 +183,7 @@ void vp8_fast_quantize_b_ssse3(BLOCK *b, BLOCKD *d)
                     d->qcoeff,
                     d->dequant,
                     b->round,
-                    b->quant,
+                    b->quant_fast,
                     d->dqcoeff
                );
 }